1-Nov-83 10:43:08-PST,18980;000000000001 Mail-From: LAWS created at 1-Nov-83 09:58:29 Date: Tuesday, November 1, 1983 9:47AM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #87 To: AIList@SRI-AI AIList Digest Tuesday, 1 Nov 1983 Volume 1 : Issue 87 Today's Topics: Rational Psychology - Definition, Parallel Systems, Conciousness & Intelligence, Halting Problem, Molecular Computers ---------------------------------------------------------------------- Date: 29 Oct 83 23:57:36-PDT (Sat) From: hplabs!hao!csu-cs!denelcor!neal @ Ucb-Vax Subject: Re: Rational Psychology Article-I.D.: denelcor.182 I see what you are saying and I beg to disagree. I don't believe that the distinction between rational and irrational psychology (it's probably not that simple) as depending on whether or not the scientist is being rational but on whether or not the subject is (or rather which aspect of his behavior--or mentation if you accept the existence of that--is under consideration). More like the distinction between organic and inorganic chemistry. ------------------------------ Date: Mon, 31 Oct 83 10:16:00 PST From: Philip Kahn Subject: Sequential vs. parallel It was claimed that "parallel computation can always be done sequentially." I had thought that this naive concept had passed away into never never land, but I suppose not. I do not deny that MANY parallel computations can be accomplished sequentially, yet not ALL parallel computations can be made sequential. Those class of parallel computations that cannot be accomplished sequentially are those that involve the state of all variables in a single instant. This class of parallelism often arises in sensor applications. It would not be valid, for example, to raster-scan (sequential computation) a sensing field if the processing of that sensing field relied upon the quantization of elements in a single instant. I don't want to belabor this point, but it should be recognized that the common assertion that all parallel computation can be done sequentially is NOT ALWAYS VALID. In my own experience, I have found that artificial intelligence (and real biologic intelligence for that matter) relies heavily upon comparisons of various elements at a single time instant. As such, the assumption of sequentialty of parallelistic algorithms is often invalid. Something to think about. ------------------------------ Date: Saturday, 29 Oct 1983 21:05-PST From: sdcrdcf!trw-unix!scgvaxd!qsi03!achut@rand-relay Subject: Conciousness, Halting Problem, Intelligence I am new to this mailing list and I see there is some lively discussion going on. I am eager to contribute to it. Consciousness: I treat the words self-awareness, consciousness, and soul as synonyms in the context of these discussions. They are all epiphenomena of the phenomenon of intelligence, along with emotions, desires, etc. To say that machines can never be truly intelligent because they cannot have a "soul" is to be excessively naive and anthropocentric. Self- awareness is not a necessary prerequisite for intelligence; it arises naturally *because* of intelligence. All intelligent beings possess some degree of self-awareness; to perceive and interact with the world, there must be an internal model, and this invariably involves taking into account the "self". A very, very low intelligence, like that of a plant, will possess a very, very low self-awareness. Parallelism: The human brain resembles a parallel machine more than it does a purely sequential one. Parallel machines can do many things much quicker than their sequential counterpart. Parallel hardware may well make the difference between the attainment of AI in the near future and the unattainment for several decades. But I cannot understand those who claim that there is something *fundamentally* different between the two types of architectures. I am always amazed at the extremes to which some people will go to find the "magic spark" which separates intelligence from non- intelligence. Two of these are "continuousness vs. discreteness" and "non-determinism vs. determinism". Continuous? Nothing in the universe is continuous. (Except maybe arguments to the contrary :-)) Mass, energy, space and even time, at least according to current physical knowledge, are all quantized. Non-determinism? Many people feel that "randomness" is a necessary ingredient to intelligence. But why isn't this possible with a sequential architecture? I can construct a "discrete" random number generator for my sequential machine so that it behaves in a similar manner to your "non-deterministic" parallel machine, although perhaps slower. (See "Slow intelligence" below) Perhaps the "magic sparkers" should consider that difference they are searching for is merely one of complexity. (I really hate to use the word "merely", since I appreciate the vast scope of the complexity, but it seems appropriate here) There is no evidence, currently, to justify thinking otherwise. The Halting(?) Problem: What Stan referred to as the "Halting Problem" is really the "looping problem", hence the subsequent confusion. The Halting Problem is not really relevant to AI, but the looping problem *is* relevant. The question is not even "why don't humans get caught in loops", since, as Mr. Frederking aptly points out, "beings which aren't careful about this fail to breed, and are weeded out by evolution". (For an interesting story of what could happen if this were not the case, see "Riddle of the universe and its solution" by Christoper Cerniak in "The Mind's I") But rather, the more interesting question is "by what mechanisms do humans avoid them?", and then, "are these the best mechanisms to use in AI programs?". It not clear that this might not be a problem when AI is attempted on a machine whose internal states could conceivably recur. Now I am not saying that this an insurmountable problem by any means; I am merely saying that it might be a worthy topic of discussion. Slow intelligence: Intelligence is dependent on time? This would require a curious definition of intelligence. Suppose you played chess at strength 2000 given 5 seconds per move, 2010 given 5 minutes, and 2050 given as much time as you desired. Suppose the corresponding numbers for me were 1500, 2000, and 2500. Who is the better (more intelligent) player? True, I need 5 minutes per move just to play as good as you can at only 5 seconds. But shouldn't the "high end" be compared instead? There are many bases on which to decide the "greater" of two intelligences. One is (conceivably, but not exclusively) speed. Another is number and power of inferences it can make in a given situation. Another is memory, and ability to correlate current situations with previous ones. STRAZ@MIT-OZ has the right idea. Incidentally, I'm surprised that no one pointed out an example of an intelligence staring us in the face which is slower but smarter than us all, individually. Namely, this net! ------------------------------ Date: 25 Oct 83 13:34:02-PDT (Tue) From: harpo!eagle!mhuxl!ulysses!cbosgd!cbscd5!pmd @ Ucb-Vax Subject: Artificial Consciousness? [and Reply] I'm interested in getting some feedback on some philosophical questions that have been haunting me: 1) Is there any reason why developments in artificial intelligence and computer technology could not someday produce a machine with human consciousness (i.e. an I-story)? 2) If the answer to the above question is no, and such a machine were produced, what would distinguish it from humans as far as "human" rights were concerned? Would it be murder for us to destroy such a machine? What about letting it die of natural (?) causes if we have the ability to repair it indefinitely? (Note: Just having a unique, human genetic code does not legally make one human as per the 1973 *Row vs Wade* Supreme Court decision on abortion.) Thanks in advance. Paul Dubuc [For an excellent discussion of the rights and legal status of AI systems, see Marshal Willick's "Artificial Intelligence: Some Legal Approaches and Implications" in the Summer '83 issue (V. 4, N. 2) of AI magazine. The resolution of this issue will of course be up to the courts. -- KIL] ------------------------------ Date: 28 Oct 1983 21:01-PDT From: fc%usc-cse%USC-ECL@SRI-NIC Subject: Halting in learning programs If you restrict the class of things that can be learned by your program to those which don't cause infinite recursion or circularity, you will have a good solution to the halting problem you state. Although generalized learning might be nice, until we know more about learning, it might be more appropriate to select specific classes of adaption which lend themselves to analysis and development of new theories. As a simple example of a non halting problem learning automata, the Purr Puss system developed by John Andreas (from New Zealand) does an excellent job of learning without any such difficulty. Other such systems exist as well, all you have to do is look for them. I guess the point is that rather than pursue the impossible, find something possible that may lead to the solution of a bigger problem and pursue it with the passion and rigor worthy of the problem. An old saying: 'Problems worthy of attack prove their worth by fighting back' Fred ------------------------------ Date: Sat, 29 Oct 83 13:23:33 CDT From: Bob.Warfield Subject: Halting Problem Discussion It turns out that any computer program running on a real piece of hardware may be simulated by a deterministic finite automaton, since it only has a finite (but very large) number of possible states. This is usually not a productive observation to make, but it does present one solution to the halting problem for real (i.e. finite) computing hardware. Simulate the program in question as a DFA and look for loops. From this, one should be able to tell what input to the DFA would produce an infinite loop, and recognition of that input could be done by a smaller DFA (the old one sans loops) that gets incorporated into the learning program. It would run the DFA in parallel (or 1 step ahead?) and take action if a dangerous situation appeared. Bob Warfield warbob@rice ------------------------------ Date: Mon 31 Oct 83 15:45:12-PST From: Calton Pu Subject: Halting Problem: Resource Use From Shebs@Utah-20: The question is this: consider a learning program, or any program that is self-modifying in some way. What must I do to prevent it from getting caught in an infinite loop, or a stack overflow, or other unpleasantnesses? ... How can *it* know when it's stuck in a losing situation? Trying to come up with a loop detector program seemed to find few enthusiasts. The limited loop detector suggests another approach to the "halting problem". The question above does not require the solution of the halting problem, although that could help. The question posed is one of resource allocation and use. Self-awareness is only necessary for the program to watch itself and know whether it is making progress considering its resource consumption. Consequently it is not surprising that: The best answers I saw were along the lines of an operating system design, where a stuck process can be killed, or pushed to the bottom of an agenda, or whatever. However, Stan wants more: Workable, but unsatisfactory. In the case of an infinite loop (that nastiest of possible errors), the program can only guess that it has created a situation where infinite loops can happen. The real issue here is not whether the program is in loop, but whether the program will be able to find a solution in feasible time. Suppose a program will take a thousand years to find a solution, will you let it run that long? In other words, the problem is one of measuring gained progress versus spent resources. It may turn out that a program is not in loop but you choose to write another program instead of letting the first run to completion. Looping is just one of the losing situations. Summarizing, the learning program should be allowed to see a losing situation because it is unfeasible, whether the solution is possible or not. From this view, there are two aspects to the decision: the measurement of progress made by the program, and monitoring resource consumption. It is the second aspect that involves some "operating systems design". I would be interested to know whether your parser knows it is making progress. -Calton- Usenet: ...decvax!microsoft!uw-beaver!calton ------------------------------ Date: 31 Oct 83 2030 EST From: Dave.Touretzky@CMU-CS-A Subject: forwarded article - - - - Begin forwarded message - - - - Date: 31 Oct 1983 18:41 EST (Mon) From: Daniel S. Weld To: macmol%MIT-OZ@MIT-MC.ARPA Subject: Molecular Computers Below is a forwarded message: From: David Rogers I have always been confused by the people who work on "molecular computers", it seems so stupid. It seems much more reasonable to consider the reverse application: using computers to make better molecules. Is anyone out there excited by this stuff? MOLECULAR COMPUTERS by Lee Dembart, LA Times (reprinted from the San Jose Mercury News 31 Oct 83) SANTA MONICA - Scientists have dreamed for the past few years of building a radically different kind of computer, one based on molecular reactions rather than on silicon. With such a machine, they could pack circuits much more tightly than they can inside today's computers. More important, a molecular computer might not be bound by the rigid binary logic of conventional computers. Biological functions - the movement of information within a cell or between cells - are the models for molecular computers. If that basic process could be reproduced in a machine, it would be a very powerful machine. But such a machine is many, many years away. Some say the idea is science fiction. At the moment, it exists only in the minds of of several dozen computer scientists, biologists, chemists and engineers, many of whom met here last week under the aegis of the Crump Institute for Medical Engineering at the University of California at Los Angeles. "There are a number of ideas in place, a number of technologies in place, but no concrete results," said Michael Conrad, a biologist and computer scientist at Wayne State University in Detroit and a co-organizer of the conference. For all their strengths, today's digital computers have no ability to judge. They cannot recognize patterns. They cannot, for example, distinguish one face from another, as even babies can. A great deal of information can be packed on a computer chip, but it pales by comparison to the contents of the brain of an ant, which can protect itself against its environment. If scientists had a computer with more flexible logic and circuitry, they think they might be able to develop "a different style of computing", one less rigid than current computers, one that works more like a brain and less like a machine. The "mood" of such a device might affect the way scientists solve problems, just as people's moods affect their work. The computing molecules would be manufactured by genetically engineered bacteria, which has given rise to the name "biochip" to describe a network of them. "This is really the new gene technology", Conrad said. The conference was a meeting on the frontiers - some would say fringes - of knowledge, and several times participants scoffed, saying that the discussion was meandering into philosophy. The meeting touched on some of the most fundamental questions of brain and computer research, revealing how little is known of the mind's mechanisms. The goal of artificial intelligence work is to write programs that simulate thought on digital computers. The meeting's goal was to think about different kinds of computers that might do that better. Among the questions posed at the conference: - How do you get a computer to chuckle at a joke? - What is the memory capacity of the brain? Is there a limit to that capacity? - Are there styles of problem solving that are not digitally computable? - Can computer science shed any light on the mechanisms of biological science? Can computer science problems be addressed by biological science mechanisms? Proponents of molecular computers argue that it is possible to make such a machine because biological systems perform those processes all the time. Proponents of artificial intelligence have argued for years that the existence of the brain is proof that it is possible to make a small machine that thinks like a brain. It is a powerful argument. Biological systems already exist that compute information in a better way than digital computers do. "There has got to be inspiration growing out of biology", said F. Eugene Yates, the Crump Institutes director. Bacteria use sophisticated chemical processes to transfer information. Can that process be copied? Enzymes work by stereoscopically matching their molecules with other molecules, a decision-making process that occurs thousands of times a second. It would take a binary computer weeks to make even one match. "It's that failure to do a thing that an enzyme does 10,000 times a second that makes us think there must be a better way," Yates said. In the history of science, theoretical progress and technological progress are intertwined. One makes the other possible. It is not surprising, therefore, that thinking about molecular computers has been spurred recently by advances in chemistry and biotechnology that seem to provide both the materials needed and a means for producing it on a commercial scale. "If you could design such a reaction, you could probably get a bacteria to make it," Yates said. Conrad thinks that a functioning machine is 50 years away, and he described it as a "futuristic" development. - - - - End forwarded message - - - - ------------------------------ End of AIList Digest ******************** 3-Nov-83 13:38:43-PST,11566;000000000001 Mail-From: LAWS created at 3-Nov-83 13:26:23 Date: Thursday, November 3, 1983 1:09PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #88 To: AIList@SRI-AI AIList Digest Thursday, 3 Nov 1983 Volume 1 : Issue 88 Today's Topics: Molecular Computers - Comment, Sequential Systems - Theoretical Sufficiency, Humanness - Definition, Writing Analysis - Reference, Lab Report - Prolog and SYLLOG at IBM, Seminars - Translating LISP & Knowledge and Reasoning ---------------------------------------------------------------------- Date: 1 Nov 83 1844 EST From: Dave.Touretzky@CMU-CS-A Subject: Comment on Molecular Computers - - - - Begin forwarded message - - - - Date: Tue, 1 Nov 1983 12:19 EST From: DANNY%MIT-OZ@MIT-MC.ARPA To: Daniel S. Weld Subject: Molecular Computers I was at the Molecular Computer conference. Unfortunately, there has very lttle progress since the Molecular Electronics conference a year ago. The field is too full of people who think analog computation is "more powerful" and who think that Goedel's proof shows that people can always think better than machine. Sigh. --danny ------------------------------ Date: Thursday, 3 November 1983 13:27:10 EST From: Robert.Frederking@CMU-CS-CAD Subject: Parallel vs. Sequential Re: Phillip Kahn's claim that "not ALL parallel computations can be made sequential": I don't believe it, unless you are talking about infinitely many processing elements. The Turing Machine is the most powerful model of computation known, and it is inherently serial (and equivalent to a Tesselation Automaton, which is totally parallel). Any computation that requires all the values at an "instant" can simply run at N times the sampling rate of your sensors: it locks them, reads each one, and makes its decisions after looking at all of them, and then unlocks them to examine the next time slice. If one is talking practically, this might not be possible due to speed considerations, but theoretically it is possible. So while at a theoretical level ALL parallel computations can be simulated sequentially, in practice one often requires parallelism to cope with real-world speeds. ------------------------------ Date: 2 Nov 83 10:52:22 PST (Wednesday) From: Hoffman.es@PARC-MAXC.ARPA Subject: Awareness, Human-ness Sorry it took me a while to track this down. It's something I recalled when reading the discussion of awareness in V1 #80. It's been lightly edited. --Rodney Hoffman **** **** **** **** **** **** **** **** From Richard Rorty's book, "Philosophy and The Mirror of Nature": Personhood is a matter of decision rather than knowledge, an acceptance of another being into fellowship rather than a recognition of a common essence. Knowledge of what pain is like or what red is like is attributed to beings on the basis of their potential membership in the community. Thus babies and the more attractive sorts of animal are credited with "having feelings" rather than (like machines or spiders) "merely responding to stimuli." To say that babies know what heat is like, but not what the motion of molecules is like is just to say that we can fairly readily imagine them opening their mouths and remarking on the former, but not the latter. To say that a gadget that says "red" appropriately *doesn't* know what red is like is to say that we cannot readily imagine continuing a conversation with the gadget. Attribution of pre-linguistic awareness is merely a courtesy extended to potential or imagined fellow-speakers of our language. Moral prohibitions against hurting babies and the better looking sorts of animals are not based on their possessions of feeling. It is, if anything, the other way around. Rationality about denying civil rights to morons or fetuses or robots or aliens or blacks or gays or trees is a myth. The emotions we have toward borderline cases depend on the liveliness of our imagination, and conversely. ------------------------------ Date: 1 November 1983 18:55 EDT From: Herb Lin Subject: writing analysis You might want to take a look at some of the stuff by R. Flesch who is the primary exponent of a system that takes word and sentence and paragraph lengths and turns it into grade-equivalent reading scores. It's somewhat controversial. [E.g., The Art of Readable Writing. Or, "A New Readability Index", J. of Applied Psychology, 1948, 32, 221-233. References to other authors are also given in Cherry and Vesterman's writeup of the STYLE and DICTION systems included in Berkeley Unix. -- KIL] ------------------------------ Date: Monday, 31-Oct-83 11:49:55-GMT From: Bundy HPS (on ERCC DEC-10) Subject: Prolog and SYLLOG at IBM [Reprinted from the Prolog Digest.] Date: 9 Oct 1983 11:43:51-PDT (Sunday) From: Adrian Walker Subject: Prolog question IBM Research Laboratory K51 5600 Cottle Road San Jose CA 95193 USA Telephone: 408-256-6999 ARPANet: Adrian.IBM@Rand-Relay 10th October 83 Alan, In answer to your question about Prolog implementations, we do most of our work using the Waterloo Prolog 1.3 interpreter on an IBM mainframe (3081). Although not a traditional AI environment, this turns out to be pretty good. For instance, the speed of the Interpreter turns out to be about the same as that of compiled DEC-10 Prolog (running on a DEC-10). As for environment, the system delivered by Waterloo is pretty much stand alone, but there are several good environments built in Prolog on top of it. A valuable feature of Waterloo Prolog 1.3 is a 'system' predicate, which can call anything on the system, E.g. a full screen editor. The work on extracting explanations of 'yes' and 'no' answers from Prolog, which I reported at IJCAI, was done in Waterloo Prolog. We have also implemented a syllogistic system called SYLLOG, and several expert system types of applications. An English language question answerer written by Antonio Porto and me, produces instantaneous answers, even when the 3081 has 250 users. As far as I know, Waterloo Prolog only runs under the VM operating system (not yet under MVS, the other major IBM OS for mainframes). It is available, for a moderate academic licence fee, from Sandra Ward, Department of Computing Services, University of Waterloo, Waterloo, Ontario, Canada. We use it with IBM 3279 colour terminals, which adds variety to a long day at the screen, and can also be useful ! Best wishes, -- Adrian Walker Walker, A. (1981). 'SYLLOG: A Knowledge Based Data Management System,' Report No. 034. Computer Science Department, New York University, New York. Walker, A. (1982). 'Automatic Generation of Explanations of Results from Knowledge Bases,' RJ3481. Computer Science Department, IBM Research Laboratory, San Jose, California. Walker, A. (1983a). 'Data Bases, Expert Systems, and PROLOG,' RJ3870. Computer Science Department, IBM Research Laboratory, San Jose, California. (To appear as a book chapter) Walker, A. (1983b). 'Syllog: An Approach to Prolog for Non-Programmers.' RJ3950, IBM Research Laboratory, San Jose, Cal1fornia. (To appear as a book chapter) Walker, A. (1983c). 'Prolog/EX1: An Inference Engine which Explains both Yes and No Answers.' RJ3771, IBM Research Laboratory, San Jose, Calofornia. (Proc. IJCAI 83) Walker, A. and Porto, A. (1983). 'KBO1, A Knowledge Based Garden Store Assistant.' RJ3928, IBM Research Laboratory, San Jose, California. (In Proc Portugal Workshop, 1983.) ------------------------------ Date: Mon 31 Oct 83 22:57:03-CST From: John Hartman Subject: Fri. Grad Lunch - Understanding and Translating LISP [Reprinted from the UTEXAS-20 bboard.] GRADUATE BROWN BAG LUNCH - Friday 11/4/83, PAI 5.60 at noon: I will talk about how programming knowledge contributes to understanding programs and translating between high level languages. The problems of translating between LISP and MIRROR (= HLAMBDA) will be introduced. Then we'll look at the translation of A* (Best First Search) and see some examples of how recognizing programming cliches contributes to the result. I'll try to keep it fairly short with the hope of getting critical questions and discussion. Old blurb: I am investigating how a library of standard programming constructs may be used to assist understanding and translating LISP programs. A programmer reads a program differently than a compiler because she has knowledge about computational concepts such as "fail/succeed loop" and can recognize them by knowing standard implementations. This recognition benefits program reasoning by creating useful abstractions and connections between program syntax and the domain. The value of cliche recognition is being tested for the problem of high level translation. Rich and Temin's MIRROR language is designed to give a very explicit, static expression of program information useful for automatically answering questions about the program. I am building an advisor for LISP to MIRROR translation which will exploit recognition to extract implicit program information and guide transformation. ------------------------------ Date: Wed, 2 Nov 83 09:17 PST From: Moshe Vardi Subject: Knowledge Seminar [Forwarded by Yoni Malachi .] We are planning to start at IBM San Jose a research seminar on theoretical aspects of reasoning about knowledge, such as reasoning with incomplete information, reasoning in the presence of inconsistencies, and reasoning about changes of belief. The first few meetings are intended to be introductory lectures on various attempts at formalizing the problem, such as modal logic, nonmonotonic logic, and relevance logic. There is a lack of good research in this area, and the hope is that after a few introductory lectures, the format of the meetings will shift into a more research-oriented style. The first meeting is tentatively scheduled for Friday, Nov. 18, at 1:30, with future meetings also to be held on Friday afternoon, but this may change if there are a lot of conflicts. The first meeting will be partly organizational in nature, but there will also be a talk by Joe Halpern on "Applying modal logic to reason about knowledge and likelihood". For further details contact: Joe Halpern [halpern.ibm-sj@rand-relay, (408) 256-4701] Yoram Moses [yom@sail, (415) 497-1517] Moshe Vardi [vardi@su-hnv, (408) 256-4936] 03-Nov-83 0016 MYV Knowledge Seminar We may have a problem with Nov. 18. The response from Stanford to the announcement is overwhelming, but have a room only for 25 people. We may have to postpone the seminar. To be added to the mailing list contact Moshe Vardi (MYV@sail,vardi@su-hnv) ------------------------------ End of AIList Digest ******************** 3-Nov-83 17:04:40-PST,15168;000000000001 Mail-From: LAWS created at 3-Nov-83 17:03:25 Date: Thursday, November 3, 1983 4:59PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #89 To: AIList@SRI-AI AIList Digest Friday, 4 Nov 1983 Volume 1 : Issue 89 Today's Topics: Intelligence - Definition & Measurement & Necessity for Definition ---------------------------------------------------------------------- Date: Tue, 1 Nov 83 13:39:24 PST From: Philip Kahn Subject: Definition of Intelligence When it comes down to it, isn't intelligence the ability to recognize space-time relationships? The nice thing about this definition is that it recognizes that ants, programs, and humans all possess varying degrees of intelligence (that is, varying degrees in their ability to recognize space-time relationships). This implies that intelligence is only correlative, and only indirectly related to physical environmental interaction. ------------------------------ Date: Tue, 1 Nov 1983 22:22 EST From: SLOAN%MIT-OZ@MIT-MC.ARPA Subject: Slow intelligence/chess ... Suppose you played chess at strength 2000 given 5 seconds per move, 2010 given 5 minutes, and 2050 given as much time as you desired... An excellent point. Unfortunately wrong. This is a common error, made primarily by 1500 players and promoters of chess toys. Chess ratings measure PERFORMANCE at TOURNAMENT TIME CONTROLS (generally ranging between 1.5 to 3 moves per minute). To speak of "strength 2000 at 5 seconds per move" or "2500 given as much time as desired" is absolutely meaningless. That is why there are two domestic rating systems, one for over-the-board play and another for postal chess. Both involve time limits, the limits are very different, and the ratings are not comparable. There is probably some correlation, but the set of skills involved are incomparable. This is entirely in keeping with the view that intelligence is coupled with the environment, and involves a speed factor (you must respond in "real-time" - whatever that happens to mean.) It also speaks to the question of "loop-avoidance": in the real world, you can't step in the same stream twice; you must muddle through, ready or not. To me, this suggests that all intelligent behavior consists of generating crude, but feasible solutions to problems very quickly (so as to be ready with a response) and then incrementally improving the solution as time permits. In an ever changing environment, it is better to respond inadequately than to ponder moot points. -Ken Sloan ------------------------------ Date: Tue, 1 Nov 1983 10:15:54 EST From: AXLER.Upenn-1100@Rand-Relay (David M. Axler - MSCF Applications Mgr.) Subject: Turing Test Re-visited I see that the Turing Test has (not unexpectedly) crept back into the discussions of intelligence (1:85). I've wondered a bit as to whether the TT shouldn't be extended a bit; to wit, the challenge it poses should not only include the ability to "pass" the test, but also the ability to act as a judge for the test. Examining the latter should give us all sorts of clues as to what preconceived notions we're imposing when we try to develop a machine or program that satisfies only Turing's original problem Dave Axler ------------------------------ Date: Wed, 2 Nov 1983 10:10 EST From: MINSKY%MIT-OZ@MIT-MC.ARPA Subject: Parallelism & Consciousness What I meant is that defining intelligence seems as pointless as defining "life" and then arguing whether viruses are alive instead of asking how they work and solve the problems that appear to us to be the interesting ones. Instead of defining so hard, one should look to see what there is. For example, about the loop-detecting thing, it is clear that in full generality one can't detect all Turing machine loops. But we all know intelligent people who appear to be caught, to some extent, in thought patterns that appear rather looplike. That paper of mine on jokes proposes that to be intelligent enough to keep out of simple loops, the problem is solved by a variety of heuristic loop detectors, etc. Of course, this will often deflect one from behaviors that aren't loops and which might lead to something good if pursued. That's life. I guess my complaint is that I think it is unproductive to be so concerned with defining "intelligence" to the point that you even discuss whether "it" is time-scale invariant, rather than, say, how many computrons it takes to solve some class of problems. We want to understand problem-solvers, all right. But I think that the word "intelligence" is a social one that accumulates all sorts of things that one person admires when observed in others and doesn't understand how to do. No doubt, this can be narrowed down, with great effort, e.g., by excluding physical; skills (probably wrongly, in a sense) and so forth. But it seemed to me that the discussion here in AILIST was going nowwhere toward understand intelligence, even in that sense. In other words, it seems strange to me that there is no public discussion of substantive issues in the field... ------------------------------ Date: Wed, 2 Nov 1983 10:21 EST From: MINSKY%MIT-OZ@MIT-MC.ARPA Subject: Intelligence and Competition The ability to cope with a CHANGE in the environment marks intelligence. See, this is what's usually called adaptiveness. This is why you don't get anywhere defining intelligence -- until you have a clear idea to define. Why be enslaved to the fact that people use a word, unless you're sure it isn't a social accumulation. ------------------------------ Date: 2 Nov 1983 23:44-PST From: ISAACSON@USC-ISI Subject: Re: Parallelism & Consciousness From Minsky: ...I think that the word "intelligence" is a social one accumulates all sorts of things that one person admires observed in others and doesn't understand how to do... In other words, it seems strange to me that there is no public discussion of substantive issues in the field... Exactly... I agree on both counts. My purpose is to help crystallize a few basic topics, worthy of serious discussion, that relate to those elusive epiphenomena that we tend to lump under that loose characterization: "Intelligence". I read both your LM and Jokes papers and consider them seminal in that general direction. I think, though, that your ideas there need, and certainly deserve, further elucidation. In fact, I was hoping that you would be willing to state some of your key points to this audience. More than this. Recently I've been attracted to Doug Hofstadter's ideas on subcognition and think that attention should be paid to them as well. As a matter of fact, I see certain affinities between you two and would like to see a good discussion that centers on LM, Jokes, and Subcognition as Computation. I think that, in combination, some of the most promising ideas for AI are awaiting full germination in those papers. ------------------------------ Date: Thu, 3 Nov 1983 13:17 EST From: BATALI%MIT-OZ@MIT-MC.ARPA Subject: Inscrutable Intelligence From Minsky: ...I think that the word "intelligence" is a social one that accumulates all sorts of things that one person admires when observed in others and doesn't understand how to do... This seems like an extremely negative and defeatist thing to say. What does it leave us in AI to do, but to ignore the very notion we are supposedly trying to understand? What will motivate one line of research rather than another, what can we use to judge the quality of a piece of research, if we have no idea what it is we are after? It seems to me that one plausible approach to AI is to present an arguable account of what intelligence is about, and then to show that some mechanism is intelligent according to that account. The account, the "definition", of intelligence may not be intuitive to everyone at first. But the performance of the mechanisms constructed in accord with the account will constitute evidence that the account is correct. (This is where the Turing test comes in, not as a definition of intelligence, but as evidence for its presence.) ------------------------------ Date: Tue 1 Nov 83 13:10:32-EST From: SUNDAR@MIT-OZ Subject: parallelism and conciousness [Forwarded by RickL%MIT-OZ@MIT-MC.] [...] It seems evident from the recent conversations that the meaning of intelligence is much more than mere 'survivability' or 'adaptability'. Almost all the views expressed however took for granted the concept of "time"-which,seems to me is 'a priori'(in the Kantian sense). What do you think of a view of that says :intelligence is the ability of an organism that enables it to preserve,propagate and manipulate these 'a priori'concepts. The motivation for doing so could be a simple pleasure,pain mechanism (which again I feel are concepts not adequately understood).It would seem that while the pain mechanism would help cut down large search spaces when the organism comes up against such problems,the pleasure mechanism would help in learning,and in the acquisition of new 'a priori' wisdom. Clearly in the case of organisms that multiply by fission (where the line of division between parent and child is not exactly clear)the structure of the organism may be preserved .In such cases it would seem that the organism survives seemingly forever . However it would not be considered intelligent by the definition proposed above . The questions that seem interesting to me therefore are: 1 How do humans acquire the concept of 'time'? 2 'Change' seem to be measured in terms of time (adaptation,survival etc are all the presence or absense of change) but 'time' itself seems to be meaningless without 'change'! 3 How do humans decide that an organism is 'intelligent ' or not? Seems to me that most of the people in the AIList made judgements (the amoeba , desert tortoise, cockroach examples )which should mean that they either knew what intelligence was or wasn't-but it still isn't exactly clear after all the smoke's cleared. Any comments on the above ideas? As a relative novice to the field of AI I'd appreciate your opinions. Thanks. --Sundar-- ------------------------------ Date: Thu, 3 Nov 1983 16:42 EST From: MINSKY%MIT-OZ@MIT-MC.ARPA Subject: Inscrutable Intelligence Sure. I agree you want an account of what intelligence is "about". When I complained about making a "definition" I meant one of those useless compact thingies in dictionaries. But I don't agree that you need this for scientific motivation. Batali: do you really think Biologists need definitions of Life for such purposes? Finally, I simply don't think this is a compact phenomenon. Any such "account", if brief, will be very partial and incomplete. To expect a test to show that "the account is correct" depends on the nature of the partial theory. In a nutshell, I still don't see any use at all for such definition, and it will lead to calling all sorts of partial things "intelligence". The kinds of accounts to confirm are things like partial theories that need their own names, like heuristic search method credit-assignment scheme knowledge-representation scheme, etc. As in biology, we simply are much too far along to be so childish as to say "this program is intelligent" and "this one is not". How often do you see a biologist do an experiment and then announce "See, this is the secret of Life". No. He says, "this shows that enzyme FOO is involved in degrading substrate BAR". ------------------------------ Date: 3 Nov 1983 14:45-PST From: ISAACSON@USC-ISI Subject: Re: Inscrutable Intelligence I think that your message was really addressed to Minsky, who already replied. I also think that the most one can hope for are confirmations of "partial theories" relating, respectively, to various aspects underlying phenomena of "intelligence". Note that I say "phenomena" (plural). Namely, we may have on our hands a broad spectrum of "intelligences", each one of which the manifestation of somewhat *different* mix of underlying ingredients. In fact, for some time now I feel that AI should really stand for the study of Artificial Intelligences (plural) and not merely Artificial Intelligence (singular). ------------------------------ Date: Thu, 3 Nov 1983 19:29 EST From: BATALI%MIT-OZ@MIT-MC.ARPA Subject: Inscrutable Intelligence From: MINSKY%MIT-OZ at MIT-MC.ARPA do you really think Biologists need definitions of Life for such purposes? No, but if anyone was were claiming to be building "Artificial Life", that person WOULD need some way to evaluate research. Remember, we're not just trying to find out things about intelligence, we're not just trying to see what it does -- like the biochemist who discovers enzyme FOO -- we're trying to BUILD intelligences. And that means that we must have some relatively precise notion of what we're trying to build. Finally, I simply don't think this is a compact phenomenon. Any such "account", if brief, will be very partial and incomplete. To expect a test to show that "the account is correct" depends on the nature of the partial theory. In a nutshell, I still don't see any use at all for such definition, and it will lead to calling all sorts of partial things "intelligence". If the account is partial and incomplete, and leads to calling partial things intelligence, then the account must be improved or rejected. I'm not claiming that an account must be short, just that we need one. The kinds of accounts to confirm are things like partial theories that need their own names, like heuristic search method credit-assignment scheme knowledge-representation scheme, etc. But why are these thing interesting? Why is heuristic search better than "blind" search? Why need we assign credit? Etc? My answer: because such things are the "right" thing to do for a program to be intelligent. This answer appeals to a pre-theoretic conception of what intelligence is. A more precise notion would help us assess the relevance of these and other methods to AI. One potential reason to make a more precise "definition" of intelligence is that such a definition might actually be useful in making a program intelligent. If we could say "do that" to a program while pointing to the definition, and if it "did that", we would have an intelligent program. But I am far too optimistic. (Perhaps "childishly" so). ------------------------------ End of AIList Digest ******************** 4-Nov-83 22:25:08-PST,11809;000000000001 Mail-From: LAWS created at 4-Nov-83 22:05:10 Date: Friday, November 4, 1983 9:43PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #90 To: AIList@SRI-AI AIList Digest Saturday, 5 Nov 1983 Volume 1 : Issue 90 Today's Topics: Intelligence, Looping Problem ---------------------------------------------------------------------- Date: Thu, 3 Nov 1983 23:46 EST From: MINSKY%MIT-OZ@MIT-MC.ARPA Subject: Inscrutable Intelligence One potential reason to make a more precise "definition" of intelligence is that such a definition might actually be useful in making a program intelligent. If we could say "do that" to a program while pointing to the definition, and if it "did that", we would have an intelligent program. But I am far too optimistic. I think so. You keep repeating how good it would be to have a good definition of intelligence and I keep saying it would be as useless as the biologists' search for the definition of "life". Evidently we're talking past each other so it's time to quit. Last word: my reason for making the argument was that I have seen absolutely no shred of good ideas in this forum, apparently because of this definitional orientation. I admit the possibility that some good mathematical insight could emerge from such discussions. But I am personally sure it won't, in this particular area. ------------------------------ Date: Friday, 4 November 1983, 01:17-EST From: jcma@MIT-MC Subject: Inscrutable Intelligence [Reply to Minsky.] BOTTOM LINE: Have you heard of OPERATIONAL DEFINITIONS? You are correct in pointing out that we need not have the ultimate definition of intelligence. But, it certainly seems useful for the practical purposes of investigating the phenomena of intelligence (whether natural or artificial) to have at least an initial approximation, an operational definition. Some people, (e.g., Winston), have proposed "people-like behavior" as their operational definition for intelligence. Perhaps you can suggest an incremental improvement over that rather vague definition. If artficial intelligence can't come up with an operational definition of intellgence, no matter how crude, it tends to undermine the credibility of the discipline and encourage the view that AI researchers are flakey. Moreover, it makes it very difficult to determine the degree to which a program exhibits "intelligence." If you were being asked to spend $millions on a field of inquiry, wouldn't you find it strange (bordering on absurd) that the principle proponents couldn't render an operational definition of the object of investigation? p.s. I can't imagine that psychology has no operational definition of intelligence (in fact, what is it?). So, if worst comes to worst, AI can just borrow psychology's definition and improve on it. ------------------------------ Date: Fri, 4 Nov 1983 09:57 EST From: Dan Carnese Subject: Inscrutable Intelligence There's a wonderful quote from Wittgenstein that goes something like: One of the most fundamental sources of philosophical bewilderment is to have a substantive but be unable to find the thing that corresponds to it. Perhaps the conclusion from all this is that AI is an unfortunate name for the enterprise, since no clear definitions for I are available. That shouldn't make it seem any less flakey than, say, "operations research" or "management science" or "industrial engineering" etc. etc. People outside a research area care little what it is called; what it has done and is likely to do is paramount. Trying to find the ultimate definition for field-naming terms is a wonderful, stimulating philosophical enterprise. However, one can make an empirical argument that this activity has little impact on technical progress. ------------------------------ Date: 4 Nov 1983 8:01-PST From: fc%usc-cse%USC-ECL@SRI-NIC Subject: Re: AIList Digest V1 #89 This discussion on intelligence is starting to get very boring. I think if you want a theoretical basis, you are going to have to forget about defining intelligence and work on a higher level. Perhaps finding representational schemes to represent intelligence would be a more productive line of pursuit. There are such schemes in existence. As far as I can tell, the people in this discussion have either scorned them, or have never seen them. Perhaps you should go to the library for a while and look at what all the great philosophers have said about the nature of intelligence rather than rehashing all of their arguments in a light and incomplete manner. Fred ------------------------------ Date: 3 Nov 83 0:46:16-PST (Thu) From: hplabs!hp-pcd!orstcs!hakanson @ Ucb-Vax Subject: Re: Parallelism & Consciousness - (nf) Article-I.D.: hp-pcd.2284 No, no, no. I understood the point as meaning that the faster intelligence is merely MORE intelligent than the slower intelligence. Who's to say that an amoeba is not intelligent? It might be. But we certainly can agree that most of us are more intelligent than an amoeba, probably because we are "faster" and can react more quickly to our environment. And some super-fast intelligent machine coming along does NOT make us UNintelligent, it just makes it more intelligent than we are. (According to the previous view that faster = more intelligent, which I don't necessarily subscribe to.) Marion Hakanson {hp-pcd,teklabs}!orstcs!hakanson (Usenet) hakanson@{oregon-state,orstcs} (CSnet) ------------------------------ Date: 31 Oct 83 13:18:58-PST (Mon) From: decvax!duke!unc!mcnc!ecsvax!unbent @ Ucb-Vax Subject: re: transcendental recursion [& reply] Article-I.D.: ecsvax.1457 i'm also new on this net, but this item seemed like a good one to get my feet wet with. if we're going to pursue the topic of consciousness vs intelligence, i think it's important not to get confused about consciousness vs *self*-consciousness at the beginning. there's a perfectly clear sense in which any *sentient* being is "conscious"--i.e., conscious *of* changes in its environment. but i have yet to see any good reason for supposing that cats, rats, bats, etc. are *self*-conscious, e.g., conscious of their own states of consciousness. "introspective" or "self- monitoring" capacity goes along with self-consciousness, but i see no particular reason to suppose that it has anything special to do with *consciousness* per se. as long as i'm sticking my neck out, let me throw in a cautionary note about confusing intelligence and adaptability. cockroaches are as adaptable as all get out, but not terribly intelligent; and we all know some very intelligent folks who can't adapt to novelties at all. --jay rosenberg (escvax!unbent) [I can't go along with the cockroach claim. They are a successful species, but probably haven't changed much in millions of years. Individual cockroaches are elusive, but can they solve mazes or learn tricks? As for the "intelligent folks": I previously stated my preference for power tests over timed aptitude tests -- I happen to be rather slow to change channels myself. If these people are unable to adapt even given time, on what basis can we say that they are intelligent? If they excel in particular areas (e.g. idiot savants), we can qualify them as intelligent within those specialties, just as we reduce our expectations for symbolic algebra programs. If they reached states of high competence through early learning, then lost the ability to learn or adapt further, I will only grant that they >>were<< intelligent. -- KIL] ------------------------------ Date: 3 Nov 83 0:46:00-PST (Thu) From: hplabs!hp-pcd!orstcs!hakanson @ Ucb-Vax Subject: Re: Semi-Summary of Halting Problem Disc [& Comment] A couple weeks ago, I heard Marvin Minsky speak up at Seattle. Among other things, he discussed this kind of "loop detection" in an AI program. He mentioned that he has a paper just being published, which he calls his "Joke Paper," which discusses the applications of humor to AI. According to Minsky, humor will be a necessary part of any intelligent system. If I understood correctly, he believes that there is (will be) a kind of a "censor" which recognizes "bad situations" that the intelligent entity has gotten itself into. This censor can then learn to recognize the precursors of this bad situation if it starts to occur again, and can intervene. This then is the reason why a joke isn't funny if you've heard it before. And it is funny the first time because it's "absurd," the laughter being a kind of alarm mechanism. Naturally, this doesn't really help with a particular implementation, but I believe that I agree with the intuitions presented. It seems to agree with the way I believe *I* think, anyway. I hope I haven't misrepresented Minsky's ideas, and to be sure, you should look for his paper. I don't recall him mentioning a title or publisher, but he did say that the only reference he could find on humor was a book by Freud, called "Jokes and the Unconscious." (Gee, I hope his talk wasn't all a joke....) Marion Hakanson {hp-pcd,teklabs}!orstcs!hakanson (Usenet) hakanson@{oregon-state,orstcs} (CSnet) [Minsky has previously mentioned this paper in AIList. You can get a copy by writing to Minsky%MIT-OZ@MIT-MC. -- KIL] ------------------------------ Date: 31 Oct 83 7:52:43-PST (Mon) From: hplabs!hao!seismo!ut-sally!ut-ngp!utastro!nather @ Ucb-Vax Subject: Re: The Halting Problem Article-I.D.: utastro.766 A common characteristic of humans that is not shared by the machines we build and the programs we write is called "boredom." All of us get bored running around the same loop again and again, especially if nothing is seen to change in the process. We get bored and quit. *---> WARNING!!! <---* If we teach our programs to get bored, we will have solved the infinite-looping problem, but we will lose our electronic slaves who now work, uncomplainingly, on the same tedious jobs day in and day out. I'm not sure it's worth the price. Ed Nather ihnp4!{kpno, ut-sally}!utastro!nather ------------------------------ Date: 31 Oct 83 20:03:21-PST (Mon) From: harpo!eagle!hou5h!hou5g!hou5f!hou5e!hou5d!mat @ Ucb-Vax Subject: Re: The Halting Problem Article-I.D.: hou5d.725 If we teach our programs to get bored, we will have solved the infinite-looping problem, but we will lose our electronic slaves who now work, uncomplainingly, on the same tedious jobs day in and day out. I'm not sure it's worth the price. Hmm. I don't usually try to play in this league, but it seems to me that there is a place for everything and every talent. Build one machine that gets bored (in a controlled way, please) to work on Fermat's last Theorem. Build another that doesn't to check tolerances on camshafts or weld hulls. This [solving the looping problem] isn't like destroying one's virginity, you know. Mark Terribile Duke Of deNet ------------------------------ End of AIList Digest ******************** 6-Nov-83 22:59:43-PST,17372;000000000001 Mail-From: LAWS created at 6-Nov-83 22:58:19 Date: Sunday, November 6, 1983 10:51PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #91 To: AIList@SRI-AI AIList Digest Monday, 7 Nov 1983 Volume 1 : Issue 91 Today's Topics: Parallelism, Turing Machines ---------------------------------------------------------------------- Date: 1 Nov 83 22:39:06-PST (Tue) From: hplabs!hao!seismo!rlgvax!cvl!umcp-cs!israel @ Ucb-Vax Subject: Re: Parallelism and Conciousness Article-I.D.: umcp-cs.3498 [Initial portion missing. -- KIL] a processing unit that we can currently build. If you mean 'at the exact same time', then I defy you to show me a case where this is necessary. The statement "No algorithm is inherently parallel", just means that the algoritm itself (as opposed to the engineering of putting it into practice) does not necessarily have to be done in parallel. Any parallel algorithm that you give me, I can write a sequential algorithm that does the same thing. Now, if you assume a finite number of processors for the parallel algorithm, then the question of whether the sequential algorithm will work under time constraints is dependent on the speed of the processor worked on. I don't know if there has been any work done on theoretical limits of the speed of a processor (Does anyone know? is this a meaningful question?), but if we assume none (a very chancy assumption at best), then any parallel algorithm can be done sequentially in practice. If you allow an infinite number of processors for the parallel algorithm, then the sequential version of the algorithm can't ever work in practice. But can the parallel version? What do we run it on? Can you picture an infinitely parallel computer which has robots with shovels with it, and when the computer needs an unallocated processor and has none, then the robots dig up the appropriate minerals and construct the processor. Of course, it doesn't need to be said that if the system notices that the demand for processors is faster than the robots' processor production output, then the robots make more robots to help them with the raw materials gathering and the construction. :-) -- ^-^ Bruce ^-^ University of Maryland, Computer Science {rlgvax,seismo}!umcp-cs!israel (Usenet) israel.umcp-cs@CSNet-Relay (Arpanet) ------------------------------ Date: 31 Oct 83 19:55:44-PST (Mon) From: pur-ee!uiucdcs!uicsl!dinitz @ Ucb-Vax Subject: Re: Parallelism and Conciousness - (nf) Article-I.D.: uiucdcs.3572 I see no reason why consciousness should be inherently parallel. But it turns out that the only examples of conscious entities (i.e. those which nearly everyone agrees are conscious) rely heavily on parallelism at several levels. This is NOT to say that they derive their consciousness from parallelism, only that there is a high corelation between the two. There are good reasons why natural selection would favor parallelism. Besides the usually cited ones (e.g. speed, simplicity) is the fact that the world goes by very quickly, and carries a high information content. That makes it desirable and advantageous for a conscious entity to be aware of several things at once. This strongly suggests parallelism (although a truly original species might get away with timesharing). Pushing in the other direction, I should note that it is not necessary to bring the full power of the human intellect to bear against ALL of our environment at once. Hence the phenomenon of attention. It suffices to have weaker processes in charge of uninteresting phenomena in the environment, as long as these have the ability to enlist more of the organism's information processing power when the situation becomes interesting enough to demand it. (This too could be finessed with a clever timesharing scheme, but I know of no animal that does it that way.) Once again, none of this entails a connection causal connection between parallelism and consciousness. It just seems to have worked out that nature liked it that way (in the possible world in which we live). Rick Dinitz ...!uiucdcs!uicsl!dinitz ------------------------------ Date: 1 Nov 83 11:53:58-PST (Tue) From: hplabs!hao!seismo!rochester!blenko @ Ucb-Vax Subject: Re: Parallelism & Consciousness Article-I.D.: rocheste.3648 Interesting to see this discussion taking place among people (apparently) committed to an information-processing model for intelligence. I would be satisfied with the discovery of mechanisms that duplicate the information-processing functions associated with intelligence. The issue of real-time performance seems to be independent of functional performance (not from an engineering point of view, of course; ever tell one of your hardware friends to "just turn up the clock"?). The fact that evolutionary processes act on both the information-processing and performance characteristics of a system may argue for the (evolutionary) superiority of one mechanism over another; it does not provide prescriptive information for developing functional mechanisms, however, which is the task we are currently faced with. Tom ------------------------------ Date: 1 Nov 83 19:01:59-PST (Tue) From: hplabs!hao!seismo!rlgvax!cvl!umcp-cs!speaker @ Ucb-Vax Subject: Re: Parallelism and Conciousness Article-I.D.: umcp-cs.3523 No algorithm is inherently parallel. The algorithms you are thinking about occur in the serial world of the Turing machine. Turing machines, remember, have only only one input. Consider what happens to your general purpose turing machine when it must compute on more than one input and simultaneously! So existence in the real world may require parallelism. How do you define simultaneously? If you mean within a very short period of time, then that requirement is based on the maximum speed of a processing unit that we can currently build. If you mean 'at the exact same time', then I defy you to show me a case where this is necessary. A CHALLENGE!!! Grrrrrrrr...... Okay, let's say we have two discrete inputs that must be monitored by a Turing machine. Signals may come in over these inputs simultaneously. How do you propose to monitor both discretes at the same time? You can't monitor them as one input because your Turing machine is allowed only one state at a time on its read/write head. Remember that the states of the inputs run as fast as those of the Turing machine. You can solve this problem by building two Turing machines, each of which may look at the discretes. I don't have to appeal to practical speeds of processors. We're talking pure theory here. -- - Speaker-To-Stuffed-Animals speaker@umcp-cs speaker.umcp-cs@CSnet-Relay ------------------------------ Date: 1 Nov 83 18:41:10-PST (Tue) From: hplabs!hao!seismo!rlgvax!cvl!umcp-cs!speaker @ Ucb-Vax Subject: Infinite loops and Turing machines... Article-I.D.: umcp-cs.3521 One of the things I did in my undergrad theory class was to prove that a multiple-tape Turing machine is equivalent to one with a single tape (several tapes were very handy for programming). Also, we showed that a TM with a 2-dimensional tape infinite in both x and y was also equivalent to a single-tape TM. On the other hand, the question of a machine with an infinite number of read heads was left open... Aha! I knew someone would come up with this one! Consider that when we talk of simultaneous events... we speak of simultaneous events that occur within one Turing machine state and outside of the Turing machine itself. Can a one-tape Turing machine read the input of 7 discrete sources at once? A 7 tape machine with 7 heads could! The reason that they are not equivelent is that we have allowed for external states (events) outside of the machine states of the Turing machine itself. -- - Speaker-To-Stuffed-Animals speaker@umcp-cs speaker.umcp-cs@CSnet-Relay ------------------------------ Date: 1 Nov 83 16:56:19-PST (Tue) From: hplabs!hao!seismo!philabs!linus!security!genrad!mit-eddie!rlh @ Ucb-Vax Subject: Re: Parallelism and Conciousness Article-I.D.: mit-eddi.885 requirement is based on the maximum speed of a processing unit that we can currently build. If you mean 'at the exact same time', then I defy you to show me a case where this is necessary. The statement "No algorithm is inherently parallel", just means that the algorithm itself (as opposed to the engineering of putting it into practice) does not necessarily have to be done in parallel. Any parallel algorithm that you give me, I can write a sequential algorithm that does the same thing. Consider the retina, and its processing algorithm. It is certainly true that once the raw information has been collected and in some way band-limited, it can be processed in either fashion; but one part of the algorithm must necessarily be implemented in parallel. To get the photon efficiencies that are needed for dark-adapted vision (part of the specifications for the algorithm) one must have some continuous, distributed attention to the light field. If I match the spatial and temporal resolution of the retina, call it several thousand by several thousand by some milliseconds, by sequentially scanning with a single receptor, I can only catch one in several-squared million photons, not the order of one in ten that our own retina achieves. ------------------------------ Date: 2 Nov 83 19:44:21-PST (Wed) From: pur-ee!uiucdcs!uicsl!preece @ Ucb-Vax Subject: Re: Parallelism and Conciousness - (nf) Article-I.D.: uiucdcs.3633 There is a significant difference between saying "No algorithm is inherently parallel" and saying "Any algorithm can be carried out without parallelism." There are many algorithms that are inherently parallel. Many (perhaps all) of them can be SIMULATED without true parallel processing. I would, however, support the contention that computational models of natural processes need not follow the same implementations, and that a serial simulation of a parallel process can produce the same result. scott preece ihnp4!uiucdcs!uicsl!preece ------------------------------ Date: 2 Nov 83 15:22:20-PST (Wed) From: hplabs!hao!seismo!philabs!linus!security!genrad!grkermit!masscom p!kobold!tjt @ Ucb-Vax Subject: Re: Parallelism and Conciousness Article-I.D.: kobold.191 Gawd!! Real-time processing with a Turing machine?! Pure theory indeed! Turing machines are models for *abstract* computation. You get to write an initial string on the tape(s) and start up the machine: it does not monitor external inputs changing asynchronously. You can define your *own* machine which is just like a Turing machine, except that it *does* monitor external inputs changing asynchronously (Speaker machines anyone :-). Also, if you want to talk *pure theory*, I could just enlarge my input alphabet on a single input to encode all possible simultaneous values at multiple inputs. -- Tom Teixeira, Massachusetts Computer Corporation. Littleton MA ...!{harpo,decvax,ucbcad,tektronix}!masscomp!tjt (617) 486-9581 ------------------------------ Date: 2 Nov 83 16:28:10-PST (Wed) From: hplabs!hao!seismo!philabs!linus!security!genrad!grkermit!masscom p!kobold!tjt @ Ucb-Vax Subject: Re: Parallelism and Conciousness Article-I.D.: kobold.192 In regards to the statement No algorithm is inherently parallel. which has been justified by the ability to execute any "parallel" program on a single sequential processor. The difference between parallel and sequential algorithms is one of *expressive* power rather than *computational* power. After all, if it's just computational power you want, why aren't you all programming Turing machines? The real question is what is the additional *expressive* power of parallel programs. The additional expressive power of parallel programming languages is a result of not requiring the programmer to serialize steps of his computation when he is uncertain whether either one will terminate. -- Tom Teixeira, Massachusetts Computer Corporation. Littleton MA ...!{harpo,decvax,ucbcad,tektronix}!masscomp!tjt (617) 486-9581 ------------------------------ Date: 4 Nov 83 8:13:22-PST (Fri) From: hplabs!hao!seismo!ut-sally!ut-ngp!utastro!nather @ Ucb-Vax Subject: Our Parallel Eyeballs Article-I.D.: utastro.784 Consider the retina, and its processing algorithm. [...] There seems to be a misconception here. It's not clear to me that "parallel processing" includes simple signal accumulation. Astronomers use area detectors that simply accumulate the charge deposited by photons arriving on an array of photosensitive diodes; after the needed "exposure" the charge image is read out (sequentially) for display, further processing, etc. If the light level is high, readout can be repeated every few milliseconds, or, in some devices, proceed continuously, allowing each pixel to accumulate photons between readouts, which reset the charge to zero. I note in passing that we tend to think sequentially (our self-awareness center seems to be serial) but operate in parallel (our heart beats along, and body chemistry gets its signals even when we're chewing gum). We have, for the most part, built computers in our own (self)image: serial. We're encountering real physical limits in serial computing (the finite speed of light) and clearly must turn to parallel operations to go much faster. How we learn to "think in parallel" is not clear, but people who do the logic design of computers try to get as many operations into one clock cycle as possible, and maybe that's the place to start. Ed Nather ihnp4!{ut-sally,kpno}!utastro!nather ------------------------------ Date: 3 Nov 83 9:39:07-PST (Thu) From: decvax!microsoft!uw-beaver!ubc-visi!majka @ Ucb-Vax Subject: Get off the Turing Machines Article-I.D.: ubc-visi.513 From: Marc Majka A Turing machine is a theoretical model of computation. points out that all this noise about "simultaneous events" is OUTSIDE of the notion of a Turing machine. Turing machines are a theoretical formulation which gives theoreticians a formal system in which to consider problems in computability, decidability, the "hardness" of classes of functions, and etc. They don't really care whether set membership in a class 0 grammer is decidable in less than 14.2 seconds. The unit of time is the state transition, or "move" (as Turing called it). If you want to discuss time (in seconds or meters), you are free to invent a new model of computation which includes that element. You are then free to prove theorems about it and attempt to prove it equivalent to other models of computation. Please do this FORMALLY and post (or publish) your results. Otherwise, invoking Turing machines is a silly and meaningless exercise. Marc Majka ------------------------------ Date: 3 Nov 83 19:47:04-PST (Thu) From: pur-ee!uiucdcs!uicsl!preece @ Ucb-Vax Subject: Re: Parallelism and Conciousness - (nf) Article-I.D.: uiucdcs.3677 Arguments based on speed of processing aren't acceptable. The question of whether parallel processing is required has to be in the context of arbitrarily fast processors. Thus you can't talk about simultaneous inputs changing state at processor speed (unless you're considering the interesting case where the input is directly monitoring the processor itself and therefore intrinsically as fast as the processor; in that case you can't cope, but I'm not sure it's an interesting case with respect to consciousness). Consideration of the retina, on the other hand, brings up the basic question of what is a parallel processor. Is an input latch (allowing delayed polling) or a multi-input averager a parallel process or just part of the plumbing? We can also, of course, group the input bits and assume an arbitrarily fast processor dealing with the bits 64 (or 128 or 1 million) at a time. I don't think I'd be willing to say that intelligence or consciousness can't be slow. On the other hand, I don't think there's too much point to this argument, since it's pretty clear that producing a given level of performance will be easier with parallel processing. scott preece ihnp4!uiucdcs!uicsl!preece ------------------------------ End of AIList Digest ******************** 6-Nov-83 23:14:43-PST,12728;000000000001 Mail-From: LAWS created at 6-Nov-83 23:13:27 Date: Sunday, November 6, 1983 11:06PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #92 To: AIList@SRI-AI AIList Digest Monday, 7 Nov 1983 Volume 1 : Issue 92 Today's Topics: Halting Problem, Metaphysics, Intelligence ---------------------------------------------------------------------- Date: 31 Oct 83 19:13:28-PST (Mon) From: harpo!floyd!clyde!akgua!psuvax!simon @ Ucb-Vax Subject: Re: Semi-Summary of Halting Problem Discussion Article-I.D.: psuvax.335 About halting: it is unclear what is meant precisely by "can a program of length n decide whether programs of length <= n will halt". First, the input to the smaller programs is not specified in the question. Assuming that it is a unique input for each program, known a priori (for example, the index of the program), then the answer is obviously YES for the following restriction: the deciding program has size 2**n and decides on smaller programs (there are a few constants that are neglected too). There are less than 2*2**n programs of length <=n. For each represent halting on the specific input the test is to apply to by 1, looping by 0. The resulting string is essentially the program needed - it clearly exists. Getting hold of it is another matter - it is also obvious that this cannot be done in a uniform manner for every n because of the halting problem. At the cost of more sophisticated coding, and tremendous expenditure of time, a similar construction can be made to work for programs of length O(n). If the input is not fixed, the question is obviously hopeless - there are very small universal programs. As a practical matter it is not the halting proble that is relevant, but its subrecursive analogues. janos simon ------------------------------ Date: 3 Nov 83 13:03:22-PST (Thu) From: harpo!eagle!mhuxl!mhuxm!pyuxi!pyuxss!aaw @ Ucb-Vax Subject: Re: Halting Problem Discussion Article-I.D.: pyuxss.195 A point missing in this discussion is that the halting problem is equivalent to the question: Can a method be formulated to attempt to solve ANY problem which can determine if it is not getting closer to the solution so the meta-halters (not the clothing) can't be more than disguised time limits etc. for the general problem, since they CAN NOT MAKE INFERENCES ABOUT THE PROCESS they are to halt Aaron Werman pyuxi!pyuxss!aaw ------------------------------ Date: 9 Nov 83 21:05:28-EST (Wed) From: pur-ee!uiucdcs!uokvax!andree @ Ucb-Vax Subject: Re: re: awareness - (nf) Article-I.D.: uiucdcs.3586 Robert - If I understand correctly, your reasons for preferring dualism (or physicalism) to functionalism are: 1) It seems more intuitively obvious. 2) You are worried about legal/ethical implications of functionalism. I find that somewhat amusing, as those are EXACTLY my reasons for prefering functionalism to either dualism or physicalism. The legal implications of differentiating between groups by arbitrarily denying `souls' to one is well-known; it usually leads to slavery. Trying to find the ultimate definition for field-naming terms is a wonderful, stimulating philosophical enterprise. I think you missed the point all together. The idea is that *OPERATIONAL DEFINITIONS* are known to be useful and are found in all mature disciplines (e.g., physics). The fact that AI doesn't have an operation definition of intelligence simply points up the fact that the field of inquiry is not yet a discipline. It is a proto-discipline precisely because key issues remain vague and undefined and because there is no paradigm (in the Khunian sense of the term, not popular vulgarizations). That means that it is not possible to specify criteria for certification in the field, not to mention the requisite curriculum for the field. This all means that there is lots of work to be done before AI can enter the normal science phase. However, one can make an empirical argument that this activity has little impact on technical progress. Let's see your empirical argument. I haven't noticed any intelligent machines running around the AI lab lately. I certainly haven't noticed any that can carry on any sort of reasonable conversation. Have you? So, where is all this technical progress regarding understanding intelligence? Make sure you don't fall into the trap of thinking that intelligent machines are here today (Douglas Hofstadter debunks this position in his "Artificial Intelligence: Subcognition as Computation," CS Dept., Indiana U., Nov. 1982). ------------------------------ Date: 5 November 1983 15:38 EST From: Steven A. Swernofsky Subject: Turing test in everyday life Have you ever gotten one of those phone calls from people who are trying to sell you a magazine subscription? Those people sound *awfully* like computers! They have a canned speech, with canned places to wait for human (customer) response, and they seem to have a canned answer to anything you say. They are also *boring*! I know the entity at the other end of the line is not a computer (because they recognize my voice -- someone correct me if this is not a good test) but we might ask: how good would a computer program have to be to fool someone into thinking that it is human, in this limited case? I suspect you wouldn't have to do much, since the customer doesn't expect much from the salescreature who phones. Perhaps there is a lesson here. -- Steve [There is a system, in use, that can recognize affirmative and negative replies to its questions. It also stores a recording of your responses and can play the recording back to you before ending the conversation. The system is used for selling (e.g., record albums) and for dunning, and is effective partly because it is perceived as "mechanical". People listen to it because of the novelty, it can be programmed to make negative responses very difficult, and the playback of your own replies is very effective. -- KIL] ------------------------------ Date: 1 Nov 83 13:41:53-PST (Tue) From: hplabs!hao!seismo!uwvax!reid @ Ucb-Vax Subject: Slow Intelligence Article-I.D.: uwvax.1129 When people's intelligence is evaluated, at least subjectively, it is common to hear such things as "He is brilliant but never applies himself," or "She is very intelligent, but can never seem to get anything accomplished due to her short attention span." This seems to imply to me that intelligence is sort of like voltage--it is potential. Another analogy might be a weight-lifter, in the sense that no one doubts her ability to do amazing physical things, based on her appearance, but she needn't prove it on a regular basis.... I'm not at all sure that people's working definition of intelligence has anything at all to do with either time or sur- vival. Glenn Reid ..seismo!uwvax!reid (reid@uwisc.ARPA) ------------------------------ Date: 2 Nov 83 8:08:19-PST (Wed) From: harpo!eagle!mhuxl!ulysses!unc!mcnc!ecsvax!unbent @ Ucb-Vax Subject: intelligence and adaptability Article-I.D.: ecsvax.1466 Just two quick remarks from a philosopher: 1. It ain't just what you do; it's how you do it. Chameleons *adapt* to changing environments very quickly--in a way that furthers their goal of eating lots of flies. But what they're doing isn't manifesting *intelligence*. 2. There's adapting and adapting. I would have thought that one of the best evidences of *our* intelligence is not our ability to adapt to new environments, but rather our ability to adapt new environments to *us*. We don't change when our environment changes. We build little portable environments which suit *us* (houses, spaceships), and take them along. ------------------------------ Date: 3 Nov 83 7:51:42-PST (Thu) From: decvax!tektronix!ucbcad!notes @ Ucb-Vax Subject: What about physical identity? - (nf) Article-I.D.: ucbcad.645 It's surprising to me that people are still speaking in terms of machine intelligence unconnected with a notion of a physical host that must interact with the real world. This is treated as a trivial problem at most (I think Ken Laws said that one could attach any kind of sensing device, and hence (??) set any kind of goal for a machine). So why does Hubert Dreyfus treat this problem as one whose solution is a *necessary*, though not sufficient, condition for machine intelligence? But is it a solved problem? I don't think so--nowhere near, from what I can tell. Nor is it getting the attention it requires for solution. How many robots have been built that can infer their own physical limits and capabilities? My favorite example is the oft-quoted SHRDLU conversation; the following exchange has passed for years without comment: -> Put the block on top of the pyramid -> I can't. -> Why not? -> I don't know. (That's not verbatim.) Note that in human babies, fear of falling seems to be hardwired. It will still attempt, when old enough, to do things like put a block on top of a pyramid--but it certainly doesn't seem to need an explanation for why it should not bother after the first few tries. (And at that age, it couldn't understand the explanation anyway!) SHRDLU would have to be taken down, and given another "rule". SHRDLU had no sense of what it is to fall down. It had an arm, and an eye, but only a rather contrived "sense" of its own physical identity. It is this sense that Dreyfus sees as necessary. --- Michael Turner (ucbvax!ucbesvax.turner) ------------------------------ Date: 4 Nov 83 5:57:48-PST (Fri) From: ihnp4!ihuxn!ruffwork @ Ucb-Vax Subject: RE:intelligence and adaptability Article-I.D.: ihuxn.400 I would tend to agree that it's not how a being adapts to its environment, but how it changes the local environment to better suit itself. Also, I would have to say that adapting the environment would only aid in ranking the intelligence of a being if that action was a voluntary decision. There are many instances of creatures that alter their surroundings (water spiders come to mind), but could they decide not to ??? I doubt it. ...!iham1!ruffwork ------------------------------ Date: 4 Nov 83 15:36:33-PST (Fri) From: harpo!eagle!hou5h!hou5a!hou5d!mat @ Ucb-Vax Subject: Re: RE:intelligence and adaptability Article-I.D.: hou5d.732 Man is the toolmaker and the principle tooluser of all the living things that we know of. What does this mean? Consider driving a car or skating. When I do this, I have managed to incorporate an external system into my own control system with its myriad of pathways both forward and backward. This takes place at a level below that which usually is considered to constitute intelligent thought. On the other hand, we can adopt external things into our thought-model of the world in a way which no other creature seems to be capable of. Is there any causal relationship here? Mark Terribile DOdN ------------------------------ Date: 6 Nov 1983 20:54-PST From: fc%usc-cse%USC-ECL@SRI-NIC Subject: Re: AIList Digest V1 #90 Irwin Marin's course in AI started out by asking us to define the term 'Natural Stupidity'. I guess artificial intelligence must be anything both unnatural and unstupid. We had a few naturally stupid examples to work with, so we got a definition quite quickly. Naturally stupid types were unable to adapt, unable to find new representations, and made of flesh and bone. Artificially intelligent types were machines designed to adapt their responses and seek out more accurate representations of their environment and themselves. Perhaps this would be a good 'working' definition. At any rate, definitions are only 'working' if you work with them. If you can work with this one I suggest you go to it and stop playing with definitions. FC ------------------------------ End of AIList Digest ******************** 7-Nov-83 13:20:30-PST,15167;000000000001 Mail-From: LAWS created at 7-Nov-83 13:19:17 Date: Monday, November 7, 1983 1:11PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #93 To: AIList@SRI-AI AIList Digest Tuesday, 8 Nov 1983 Volume 1 : Issue 93 Today's Topics: Implementations - Lisp for MV8000, Expert Systems - Troubleshooting & Switching Systems, Alert - IEEE Spectrum, Fifth Generation - Stalking The Gigalip, Intelligence - Theoretical Speed, Humor - Freud Reference, Metadiscussion - Wittgenstein Quote, Seminars - Knowledge Representation & Logic Programming, Conferences - AAAI-84 Call for Papers ---------------------------------------------------------------------- Date: Tue, 1 Nov 83 16:51:42 EST From: Michael Fischer Subject: Lisp for MV8000 The University of New Haven is looking for any version of Lisp that runs on a Data General MV8000, or for a portable Lisp written in Fortran or Pascal that could be brought up in a short time. Please reply to me by electronic mail and I will bring it to their attention, or contact Alice Fischer directly at (203) 932-7069. -- Michael Fischer ------------------------------ Date: 5 Nov 83 21:31:57-EST (Sat) From: decvax!microsoft!uw-beaver!tektronix!tekig1!sal @ Ucb-Vax Subject: Expert systems for troubleshooting Article-I.D.: tekig1.1442 I am in the process of evaluating the feasibility of developing expert systems for troubleshooting instruments and functionally complete circuit boards. If anyone has had any experience in this field or has seen a similar system, please get in touch with me either through the net or call me at 503-627-3678 during 8:00am - 6:00pm PST. Thanks. Salahuddin Faruqui Tektronix, Inc. Beaverton, OR 97007. ------------------------------ Date: 4 Nov 83 17:20:42-PST (Fri) From: ihnp4!ihuxl!pvp @ Ucb-Vax Subject: Looking for a rules based expert system. Article-I.D.: ihuxl.707 I am interested in obtaining a working version of a rule based expert system, something on the order of RITA, ROSIE, or EMYCIN. I am interested in the knowledge and inference control structure, not an actual knowledge base. The application would be in the area of switching system maintenance and operation. I am in the 5ESS(tm) project, and so prefer a Unix based product, but I would be willing to convert a different type if necessary. An internal BTL product would be desirable, but if anyone knows about a commercially available system, I would be interested in evaluating it. Thanks in advance for your help. Philip Polli BTL Naperville IX 1F-474 (312) 979-0834 ihuxl!pvp ------------------------------ Date: Mon 7 Nov 83 09:50:29-PST From: Ken Laws Subject: IEEE Spectrum Alert The November issue of IEEE Spectrum is devoted to the 5th Generation. In addition to the main survey (which includes some very detailed tables about sources of funding), there are: A review of Feigenbaum and McCorduck's book, by Mark Stefik. A glossary (p. 39) of about 25 AI and CS terms, taken from Gevarter's Overview of AI and Robotics for NASA. Announcement (p. 126) of The Artificial Intelligence Report, a newsletter for people interested in AI but not engaged in research. It will begin in January; no price is given. Contact Artificial Intelligence Publications, 95 First St., Los Altos, CA 94022, (415) 949-2324. Announcement (p. 126) of a tour of Japan for those interested in the 5th Generation effort. Brief discussion (p. 126) of Art and Computers: The First Artificial- Intelligence Coloring Book, a set of line drawings by an artist-taught rule-based system. An interesting parable (p. 12) for those who would educate the public about AI or any other topic. -- Ken Laws ------------------------------ Date: 5-Nov-83 10:41:44-CST (Sat) From: Overbeek@ANL-MCS (Overbeek) Subject: Stalking The Gigalip [Reprinted from the Prolog Digest.] E. W. Lusk and I recently wrote a short note concerning attempts to produce high-speed Prolog machines. I apologize for perhaps restating the obvious in the introduction. In any event we solicit comments. Stalking the Gigalip Ewing Lusk Ross A. Overbeek Mathematics and Computer Science Division Argonne National Laboratory Argonne, Illinois 60439 1. Introduction The Japanese have recently established the goal of pro- ducing a machine capable of producing between 10 million and 1 billion logical inferences per second (where a logical inference corresponds to a Prolog procedure invocation). The motivating belief is that logic programming unifies many significant areas of computer science, and that expert sys- tems based on logic programming will be the dominant appli- cation of computers in the 1990s. A number of countries have at least considered attempting to compete with the Japanese in the race to attain a machine capable of such execution rates. The United States funding agencies have definitely indicated a strong desire to compete with the Japanese in the creation of such a logic engine, as well as in the competition to produce supercomputers that can deliver at least two orders of magnitude improvement (meas- ured in megaflops) over current machines. Our goal in writ- ing this short note is to offer some opinions on how to go about creating a machine that could execute a gigalip. It is certainly true that the entire goal of creating such a machine should be subjected to severe criticism. Indeed, we feel that it is probably the case that a majority of people in the AI research community feel that it offers (at best) a misguided effort. Rather than entering this debate, we shall concentrate solely on discussing an approach to the goal. In our opinion a significant component of many of the proposed responses by researchers in the United States is based on the unstated assumption that the goal itself is not worth pursuing, and that the benefits will accrue from addi- tional funding to areas in AI that only minimally impinge on the stated objective. [ This paper is available on {SU-SCORE} as: PS:ANL-LPHunting.Txt There is a limited supply of hard copies that can be mailed to those with read-only access to this newsletter -ed ] ------------------------------ Date: Monday, 7 November 1983 12:03:23 EST From: Robert.Frederking@CMU-CS-CAD Subject: Intelligence; theoretical speed Not to stir this up again, but around here, some people like the definition that intelligence is "knowledge brought to bear to solve problems". This indicates that you need knowledge, ways of applying it, and a concept of a "problem", which implies goals. One problem with measuring human "IQ"s is that you almost always end up measuring (at least partly) how much knowledge someone has, and what culture they're part of, as well as the pure problem solving capabilities (if any such critter exists). As for the theoretical speed of processing, the speed of light is a theoretical limit on the propagation of information (!), not just matter, so the maximum theoretical cycle speed of a processor with a one foot long information path (mighty small) is a nanosecond (not too fast!). So the question is, what is the theoretical limit on the physical size of a processor? (Or, how do you build a transistor out of three atoms?) ------------------------------ Date: 4 Nov 83 7:01:30-PST (Fri) From: harpo!eagle!mhuxl!mhuxm!pyuxi!pyuxss!aaw @ Ucb-Vax Subject: Humor Article-I.D.: pyuxss.196 [Semi-Summary of Halting Problem Disc] must have been some kind of joke. Sigmunds' book is a real layman thing, and in it he asserts that the joke a: where are you going? b: MINSKY a: you said "minsky" so I'd think you are going to "pinsky". I happen to know you are going to "minsky" so whats the use in lying? is funny. aaron werman pyuxi!pyuxss!aaw ------------------------------ Date: 05 Nov 83 1231 PST From: Jussi Ketonen Subject: Inscrutable Intelligence On useless discussions - one more quote by Wittgenstein: Wovon man nicht sprachen kann, darueber muss man schweigen. ------------------------------ Date: 05 Nov 83 0910 PST Date: Fri, 4 Nov 83 19:28 PST From: Moshe Vardi Subject: Knowledge Seminar Due to the overwhelming response to my announcement and the need to find a bigger room, the first meeting is postponed to Dec. 9, 10:00am. Moshe Vardi ------------------------------ Date: Thu, 3 Nov 1983 22:50 EST From: HEWITT%MIT-OZ@MIT-MC.ARPA Subject: SEMINAR [Forwarded by SASW@MIT-MC.] Date: Thursday, November 10, l983 3:30 P.M. Place: NE43 8th floor Playroom Title: "Some Fundamental Limitations of Logic Programming" Speaker: Carl Hewitt Logic Programming has been proposed by some as the universal programming paradigm for the future. In this seminar I will discuss some of the history of the ideas behind Logic Programming and assess its current status. Since many of the problems with current Logic Programming Languages such as Prolog will be solved, it is not fair to base a critique of Logic Programming by focusing on the particular limitations of languages like Prolog. Instead I will focus discussion on limitations which are inherent in the enterprise of attempting to use logic as a programming language. ------------------------------ Date: Thu 3 Nov 83 10:44:08-PST From: Ron Brachman Subject: AAAI-84 Call for Papers CALL FOR PAPERS AAAI-84 The 1984 National Conference on Artificial Intelligence Sponsored by the American Association for Artificial Intelligence (in cooperation with the Association for Computing Machinery) University of Texas, Austin, Texas August 6-10, 1984 AAAI-84 is the fourth national conference sponsored by the American Association for Artificial Intelligence. The purpose of the conference is to promote scientific research of the highest caliber in Artificial Intelligence (AI), by bringing together researchers in the field and by providing a published record of the conference. TOPICS OF INTEREST Authors are invited to submit papers on substantial, original, and previously unreported research in any aspect of AI, including the following: AI and Education Knowledge Representation (including Intelligent CAI) Learning AI Architectures and Languages Methodology Automated Reasoning (including technology transfer) (including automatic program- Natural Language ming, automatic theorem-proving, (including generation, commonsense reasoning, planning, understanding) problem-solving, qualitative Perception (including speech, vision) reasoning, search) Philosophical and Scientific Cognitive Modelling Foundations Expert Systems Robotics REQUIREMENTS FOR SUBMISSION Timetable: Authors should submit five (5) complete copies of their papers (hard copy only---we cannot accept on-line files) to the AAAI office (address below) no later than April 2, 1984. Papers received after this date will be returned unopened. Notification of acceptance or rejection will be mailed to the first author (or designated alternative) by May 4, 1984. Title page: Each copy of the paper should have a title page (separate from the body of the paper) containing the title of the paper, the complete names and addresses of all authors, and one topic from the above list (and subtopic, where applicable). Paper body: The authors' names should not appear in the body of the paper. The body of the paper must include the paper's title and an abstract. This part of the paper must be no longer than thirteen (13) pages, including figures but not including bibliography. Pages must be no larger than 8-1/2" by 11", double-spaced (i.e., no more than twenty-eight (28) lines per page), with text no smaller than standard pica type (i.e., at least 12 pt. type). Any submission that does not conform to these requirements will not be reviewed. The publishers will allocate four pages in the conference proceedings for each accepted paper, and will provide additional pages at a cost to the authors of $100.00 per page over the four page limit. Review criteria: Each paper will be stringently reviewed by experts in the area specified as the topic of the paper. Acceptance will be based on originality and significance of the reported research, as well as quality of the presentation of the ideas. Proposals, surveys, system descriptions, and incremental refinements to previously published work are not appropriate for inclusion in the conference. Applications clearly demonstrating the power of established techniques, as well as thoughtful critiques and comparisons of previously published material will be considered, provided that they point the way to new research in the field and are substantive scientific contributions in their own right. Submit papers and Submit program suggestions general inquiries to: and inquiries to: American Association for Ronald J. Brachman Artificial Intelligence AAAI-84 Program Chairman 445 Burgess Drive Fairchild Laboratory for Menlo Park, CA 94025 Artificial Intelligence Research (415) 328-3123 4001 Miranda Ave., MS 30-888 AAAI-Office@SUMEX Palo Alto, CA 94304 Brachman@SRI-KL ------------------------------ End of AIList Digest ******************** 9-Nov-83 13:44:17-PST,15644;000000000001 Mail-From: LAWS created at 9-Nov-83 13:41:24 Date: Wednesday, November 9, 1983 1:34PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #94 To: AIList@SRI-AI AIList Digest Wednesday, 9 Nov 1983 Volume 1 : Issue 94 Today's Topics: Metaphysics - Functionalism vs Dualism, Ethics - Implications of Consciousness, Alert - Turing Biography, Theory - Parallel vs. Sequential & Ultimate Speed, Intelligence - Operational Definitions ---------------------------------------------------------------------- Date: Mon 7 Nov 83 18:30:07-PST From: WYLAND@SRI-KL.ARPA Subject: Functionalism vs Dualism in consciousness The argument of functionalism versus dualism is unresolvable because the models are based on different, complementry paradigms: * The functionalism model is based on the reductionist approach, the approach of modern science, which explains phenomena by logically relating them to controlled, repeatable, publically verifiable experiments. The explanations about falling bodies and chemical reactions are in this catagory. * The dualism model is based on the miraculous approach, which explains phenomena as singular events, which are by definition not controlled, not repeatable, not verifiable, and not public - i.e., the events are observed by a specific individual or group. The existance of UFO's, parapsychology, and the existance of externalized consciosness (i.e. soul) is in this catagory. These two paradigms are the basis of the argument of Science versus Religion, and are not resolvable EITHER WAY. The reductionist model, based on the philosophy of Parminides and others, assumes a constant, unchanging universe which we discover through observation. Such a universe is, by definition, repeatable and totally predictable: the concept that we could know the total future if we knew the position and velocity of all particles derives from this. The success of Science at predicting the future is used as an argument for this paradigm. The miraculous model assumes the reality of change, as put forth by Heraclitus and others. It allows reality to be changed by outside forces, which may or may not be knowable and/or predictable. Changes caused by outside forces are, by definition, singular events not caused by the normal chains of causality. Our personal consciousness and (by extension, perhaps) the existance of life in the universe are singular events (as far as we know), and the basic axioms of any reductionist model of the universe are, by definition, unexplainable because they must come from outside the system. The argument of functionalism versus dualism is not resolvable in a final sense, but there are some working rules we can use after considering both paradigms. Any definition of intellegence, consciousness (as opposed to Consciousness), etc. has to be based on the reductionist model: it is the only way we can explain things in such a manner that we can predict results and prove theories. On the other hand, the concept that all sources of consciousness are mechanical is a religious position: a catagorical assumption about reality. It was not that long ago that science said that stones do not fall from the sky; all it would take to make UFOs accepted as fact would be for one to land and set up shop as a merchant dealing in rugs and spices from Aldebaran and Vega. ------------------------------ Date: Tuesday, 8 November 1983 14:24:55 EST From: Robert.Frederking@CMU-CS-CAD Subject: Ethics and Definitions of Consciousness Actually, I believe you'll find that slavery has existed both with and without believing that the slave had a soul. In many ancient societies slaves were of identically the same stock as yourself, they had just run into serious economic difficulties. As I recall, slavery of the blacks in the U.S. wasn't justified by their not having souls, but by claiming they were better off (or similar drivel). The fact that denying other people had souls was used at some time to justify it doesn't bother me, since all kinds of other rationalizations have been used. Now we are approaching the time when we will have intelligent mechanical slaves. Are you advocating that it should be illegal to own robots that can pass the Turing (or other similar) test? I think that a very important thing to consider is that we can probably make a robot really enjoy being a slave, by setting up the appropriate top-level goals. Should this be illegal? I think not. Suppose we reach the point where we can alter fetuses (see "Brave New World" by Aldous Huxley) to the point where they *really* enjoy being slaves to whoever buys them. Should this be illegal? I think so. What if we build fetuses from scratch? Harder to say, but I suspect this should be illegal. The most conservative (small "c") approach to the problem is to grant human rights to anything that *might* qualify as intelligent. I think this would be a mistake, unless you allow biological organisms a distinction as outlined above. The next most conservative approach seems to me to leave the situation where it is today: if it is physically an independent human life, it has legal rights. ------------------------------ Date: 8 Nov 1983 09:26-EST From: Jon.Webb@CMU-CS-IUS.ARPA Subject: parallel vs. sequential Parallel and sequential machines are not equivalent, even in abstract models. For example, an absract parallel machine can generate truly random numbers by starting two processes at the same time, which are identical except that one sends the main processor a "0" and the other sends a "1". The main processor accepts the first number it receives. A Turing machine can generate only pseudo-random numbers. However, I do not believe a parallel machine is more powerful (in the formal sense) than a Turing machine with a true random-number generator. I don't know of a proof of this; but it sounds like something that work has been done on. Jon ------------------------------ Date: Tuesday, 8-Nov-83 18:33:07-GMT From: O'KEEFE HPS (on ERCC DEC-10) Reply-to: okeefe.r.a. Subject: Ultimate limit on computing speed -------- There was a short letter about this in CACM about 6 or 7 years ago. I haven't got the reference, but the argument goes something like this. 1. In order to compute, you need a device with at least two states that can change from one state to another. 2. Information theory (or quantum mechanics or something, I don't remember which) shows that any state change must be accompanied by a transfer of at least so much energy (a definite figure was given). 3. Energy contributes to the stress-energy tensor just like mass and momentum, so the device must be at least so big or it will undergo gravitational collapse (again, a definite figure). 4. It takes light so long to cross the diameter of the device, and this is the shortest possible delay before we can definitely say that the device is in its new state. 5. Therefore any physically realisable device (assuming the validity of general relativity, quantum mechanics, information theory ...) cannot switch faster than (again a definite figure). I think the final figure was 10^-43 seconds, but it's been a long time since I read the letter. I have found the discussion of "what is intelligence" boring, confused, and unhelpful. If people feel unhappy working in AI because we don't have an agreed definition of the I part (come to that, do we *really* have an agreed definition of the A part either? if we come across a planet inhabited by metallic creatures with CMOS brains that were produced by natural processes, should their study belong to AI or xenobiology, and does it matter?) why not just change the name of the field, say to "Epistemics And Robotics". I don't give a tinker's curse whether AI ever produces "intelligent" machines; there are tasks that I would like to see computers doing in the service of humanity that require the representation and appropriate deployment of large amounts of knowledge. I would be just as happy calling this AI, MI, or EAR. I think some of the contributors to this group are suffering from physics envy, and don't realise what an operational definition is. It is a definition which tells you how to MEASURE something. Thus length is operationally defined by saying "do such and such. Now, length is the thing that you just measured." Of course there are problems here: no amount of operational definition will justify any connection between "length-measured-by-this-foot-rule-six-years-ago" and "length-measured- by-laser-interferometer-yesterday". The basic irrelevance is that an operational definition of say light (what your light meter measures) doesn't tell you one little thing about how to MAKE some light. If we had an operational definition of intelligence (in fact we have quite a few, and like all operational definitions, nothing to connect them) there is no reason to expect that to help us MAKE something intelligent. ------------------------------ Date: 7 Nov 83 20:50:48 PST (Monday) From: Hoffman.es@PARC-MAXC.ARPA Subject: Turing biography Finally, there is a major biography of Alan Turing! Alan Turing: The Enigma by Andrew Hodges $22.50 Simon & Schuster ISBN 0-671-49207-1 The timing is right: His war-time work on the Enigma has now been de-classified. His rather open homosexuality can be discussed in other than damning terms these days. His mother passed away in 1976. (She maintained that his death in 1954 was not suicide, but an accident, and she never mentioned his sexuality nor his 1952 arrest.) And, of course, the popular press is full of stories on AI, and they always bring up the Turing Test. The book is 529 pages, plus photographs, some diagrams, an author's note and extensive bibliographic footnotes. Doug Hofstadter's review of the book will appear in the New York Times Book Review on November 13. --Rodney Hoffman ------------------------------ Date: Mon, 7 Nov 83 15:40:46 CST From: Robert.S.Kelley Subject: Operational definitions of intelligence p.s. I can't imagine that psychology has no operational definition of intelligence (in fact, what is it?). So, if worst comes to worst, AI can just borrow psychology's definition and improve on it. Probably the most generally accepted definition of intelligence in psychology comes from Abraham Maslow's remark (here paraphrased) that "Intelligence is that quality which best distinguishes such persons as Albert Einstein and Marie Curie from the inhabitants of a home for the mentally retarded." A poorer definition is that intelligence is what IQ tests measure. In fact psychologists have sought without success for a more precise definition of intelligence (or even learning) for over 100 years. Rusty Kelley (kelleyr.rice@RAND-RELAY) ------------------------------ Date: 7 Nov 83 10:17:05-PST (Mon) From: harpo!eagle!mhuxl!ulysses!unc!mcnc!ecsvax!unbent @ Ucb-Vax Subject: Inscrutable Intelligence Article-I.D.: ecsvax.1488 I sympathize with the longing for an "operational definition" of 'intelligence'--especially since you've got to write *something* on grant applications to justify all those hardware costs. (That's not a problem we philosophers have. Sigh!) But I don't see any reason to suppose that you're ever going to *get* one, nor, in the end, that you really *need* one. You're probably not going to get one because "intelligence" is one of those "open textury", "clustery" kinds of notions. That is, we know it when we see it (most of the time), but there are no necessary and sufficient conditions that one can give in advance which instances of it must satisfy. (This isn't an uncommon phenomenon. As my colleague Paul Ziff once pointed out, when we say "A cheetah can outrun a man", we can recognize that races between men and *lame* cheetahs, *hobbled* cheetahs, *three-legged* cheetahs, cheetahs *running on ice*, etc. don't count as counterexamples to the claim even if the man wins--when such cases are brought up. But we can't give an exhaustive list of spurious counterexamples *in advance*.) Why not rest content with saying that the object of the game is to get computers to be able to do some of the things that *we* can do--e.g., recognize patterns, get a high score on the Miller Analogies Test, carry on an interesting conversation? What one would like to say, I know, is "do some of the things we do *the way we do them*--but the problem there is that we have no very good idea *how* we do them. Maybe if we can get a computer to do some of them, we'll get some ideas about us--although I'm skeptical about that, too. --Jay Rosenberg (ecsvax!unbent) ------------------------------ Date: Tue, 8 Nov 83 09:37:00 EST From: ihnp4!houxa!rem@UCLA-LOCUS THE MUELLER MEASURE If an AI could be built to answer all questions we ask it to assure us that it is ideally human (the Turing Test), it ought to be smart enough to figure out questions to ask itself that would prove that it is indeed artificial. Put another way: If an AI could make humans think it is smarter than a human by answering all questions posed to it in a Turing-like manner, it still is dumber than a human because it could not ask questions of a human to make us answer the questions so that it satisfies its desire for us to make it think we are more artificial than it is. Again: If we build an AI so smart it can fool other people by answering all questions in the Turing fashion, can we build a computer, anti-Turing-like, that could make us answer questions to fool other machines into believing we are artificial? Robert E. Mueller, Bell Labs, Holmdel, New Jersey houxa!rem ------------------------------ Date: 9 November 1983 03:41 EST From: Steven A. Swernofsky Subject: Turing test in everyday life . . . I know the entity at the other end of the line is not a computer (because they recognize my voice -- someone correct me if this is not a good test) but we might ask: how good would a computer program have to be to fool someone into thinking that it is human, in this limited case? [There is a system, in use, that can recognize affirmative and negative replies to its questions. . . . -- KIL] No, I always test these callers by interrupting to ask them questions, by restating what they said to me, and by avoiding "yes/no" responses. I appears to me that the extremely limited domain, and the utter lack of expertise which people expect from the caller, would make it very easy to simulate a real person. Does the fact of a limited domain "disguise" the intelligence of the caller, or does it imply that intelligence means a lot less in a limited domain? -- Steve ------------------------------ End of AIList Digest ******************** 9-Nov-83 17:18:10-PST,17376;000000000001 Mail-From: LAWS created at 9-Nov-83 17:14:54 Date: Wednesday, November 9, 1983 5:08PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #95 To: AIList@SRI-AI AIList Digest Thursday, 10 Nov 1983 Volume 1 : Issue 95 Today's Topics: Alert - Hacker's Dictionary, Conference - Robotic Intelligence and Productivity, Tutorial - Machine Translation, Report - AISNE meeting ---------------------------------------------------------------------- Date: 8 Nov 1983 1215:19-EST From: Lawrence Osterman Subject: Guy Steele's [Reprinted from the CMU-C bboard.] New book is now out. The Hacker's Dictionary, Available in the CMU Bookstore right now. The cost is 5.95 (6.31 after taxes) and its well worth getting (It includes (among other things) The COMPLETE INTERCAL character set (ask anyone in 15-312 last fall), Trash 80,N, Moby, and many others (El Camino Bignum?)) Larry [According to another message, the CMU bookstore immediately sold out. -- KIL] ------------------------------ Date: 7 Nov 1983 1127-PST From: MEDIONI@USC-ECLC Subject: Conference announcement ****** CONFERENCE ANNOUCEMENT ****** ROBOTIC INTELLIGENCE AND PRODUCTIVITY CONFERENCE WAYNE STATE UNIVERSITY, DETROIT, MICHIGAN NOVEMBER 18-19, 1983 For more information and advance program, please contact: Dr Pepe Siy (313) 577-3841 (313) 577-3920 - Messages or Dr Singh (313) 577-3840 ------------------------------ Date: Tue 8 Nov 83 10:06:34-CST From: Jonathan Slocum Subject: Tutorial Announcement [The following is copied from a circular, with the author's encouragement. Square brackets delimit my personal insertions, for clarification. -- JS] THE INSTITUT DALLE MOLLE POUR LES ETUDES SEMANTIQUES ET COGNITIVES DE L'UNIVERSITE DE GENEVE ("ISSCO") is to hold a Tutorial on MACHINE TRANSLATION from Monday 2nd April to Friday 6th, 1984, in Lugano, Switzerland The attraction of Machine Translation as an application domain for computers has long been recognized, but pioneers in the field seriously underestimated the complexity of the problem. As a result, early systems were severely limited. The design of more recent systems takes into account the interdisciplinary nature of the task, recognizing that MT involves the construction of a complete system for the collection, representation, and strategic deployment of a specialised kind of linguistic knowledge. This demands contribution from the fields of both theoretical and computational linguistics, conputer science, and expert system design. The aim of this tutorial is to convey the state of the art by allowing experts in different aspects of MT to present their particular points of view. Sessions covering the historical development of MT and its possible future evolution will also be included to provide a tutorial which should be relevant to all concerned with the relationship between natural language and computer science. The Tutorial will take place in the Palazzo dei Congressi or the Villa Heleneum, both set in parkland on the shore of Lake Lugano, which is perhaps the most attractive among the lakes of the Swiss/Italian Alps. Situated to the south of the Alpine massif, Spring is early and warm. Participants will be accommodated in nearby hotels. Registration will take place on the Sunday evening preceding the Tutorial. COSTS: Fees for registration submitted by January 31, 1984, will be 120 Swiss franks for students, 220 Swiss franks for academic participants, and 320 Swiss franks for others. After this date the fees will increase by 50 Swiss franks for all participants. The fees cover tuition, handouts, coffee, etc. Hotel accommodation varies between 30 and 150 Swiss franks per night [booking form available, see below]. It may be possible to arrange cheaper [private] accommodation for students. FOR FURTHER INFORMATION [incl. booking forms, etc.] (in advance of the Tutorial) please contact ISSCO, 54 route des Acacias, CH-1227 Geneva; or telephone [41 for Switzerland] (22 for Geneva) 20-93-33 (University of Geneva), extension ("interne") 21-16 ("vingt-et-un-seize"). The University switchboard is closed daily from 12 to 1:30 Swiss time. [Switzerland is six (6) hours ahead of EST, thus 9 hours ahead of PST.] ------------------------------ Date: Tue 8 Nov 83 10:59:12-CST From: Jonathan Slocum Subject: Tutorial Program PROVISIONAL PROGRAMME Each session is scheduled to include a 50-minute lecture followed by a 20-minute discussion period. Most evenings are left free, but rooms will be made available for informal discussion, poster sessions, etc. Sun. 1st 5 p.m. to 9 p.m. Registration Mon. 2nd 9:30 Introductory session M. King [ISSCO] 11:20 A non-conformist's view of the G. Sampson [Lancaster] state of the art 2:30 Pre-history of Machine Translation B. Buchmann [ISSCO] 4:20 SYSTRAN P. Wheeler [Commission of the European Communities] Tue. 3rd 9:30 An overview of post-65 developments E. Ananiadou [ISSCO] S. Warwick [ISSCO] 11:20 Software for MT I: background J.L. Couchard [ISSCO] D. Petitpierre [ISSCO] 2:30 SUSY D. MAAS [Saarbruecken] 4:20 TAUM Meteo and TAUM Aviation P. Isabelle [Montreal] Wed. 4th 9:30 Linguistic representations in A. De Roeck [Essex] syntax based MT systems 11:00 AI approaches to MT P. Shann [ISSCO] 12:00 New developments in Linguistics E. Wehrli [UCLA] and possible implications for MT 3:00 Optional excursion Thu. 5th 9:30 GETA C. Boitet [Grenoble] 11:20 ROSETTA J. Landsbergen [Philips] 2:30 Software for MT II: R. Johnson [Manchester] some recent developments M. Rosner [ISSCO] 4:20 Creating an environment for A. Melby [Brigham Young] the translator Fri. 5th 9:30 METAL J. Slocum [Texas] 11:20 EUROTRA M. King [ISSCO] 2:30 New projects in France C. Boitet [Grenoble] 4:20 MT - the future A. Zampoli [Pisa] 5:30 Closing session There will be a 1/2 hour coffee break between sessions. The lunch break is from 12:30 to 2:30. ------------------------------ Date: Mon, 7 Nov 83 14:01 EST From: Visions Subject: Report on AISNE meeting (long message) BRIEF REPORT ON FIFTH ANNUAL CONFERENCE OF THE AI SOCIETY OF NEW ENGLAND Held at Brown University, Providence, Rhode Island, 4th-5th November 1983. Programme Chairman: Drew McDermott (Yale) Local Arrangements Chairman: Eugene Charniak (Brown) Friday, 4th November 8:00PM Long talk by Harry Pople (Pittsburgh), "Where is the expertise in expert systems?" Comments and insights about the general state of work in expert systems. INTERNIST: history, structure, and example. 9:30PM "Intense intellectual colloquy and tippling" [Quoted from programme] LATE Faculty and students at Brown very hospitably billeted us visitors in their homes. Saturday, 5th November 10:00AM Panel discussion, Ruven Brooks (ITT), Harry Pople (Pittsburgh), Ramesh Patil (MIT), Paul Cohen (UMass), "Feasible and infeasible expert-systems applications". [Unabashedly selective and incoherent notes:] RB: Expert systems have to be relevant, and appropriate, and feasible. There are by-products of building expert systems, for example, the encouragement of the formalization of the problem domain. HP: Historically, considering DENDRAL and MOLGEN, say, users have ultimately made greater use of the tools and infrastructure set up by the designers than of the top-level capabilities of the expert system itself. The necessity of taking into account the needs of the users. RP: What is an expert system? Is MACSYMA no more than a 1000-key pocket calculator? Comparison of expert systems against real experts. Expert systems that actually work -- narrow domains in which hypotheses can easily be verified. What if the job of identifying the applicability of an expert system is a harder problem than the one the expert system itself solves? In the domains of medical diagnosis: enormous space of diagnoses, especially if multiple disorders are considered. Needed: reasoning about: 3D space, anatomy; time; multiple disorders, causality; demography; physiology; processes. HP: A strategic issue in research: small-scale, tractable problems that don't scale up. Is there an analogue of Blocksworld? PC: Infeasible (but not too infeasible) problems are fit material for research; feasible problems for development. The importance of theoretical issues in choosing an application area for research. An animated, general discussion followed. 11:30AM Short talks: Richard Brown (Mitre), Automatic programming. Use of knowledge about programming and knowledge about the specific application domain. Ken Wasserman (Columbia), "Representing complex physical objects". For use in a system that digests patent abstracts. Uses frame-like represent- ation, giving parts, subparts, and the relationships between them. Paul Barth (Schlumberger-Doll), Automatic programming for drilling-log interpretation, based on a taxonomy of knowledge sources, activities, and corresponding transformation and selection operations. Malcolm Cook (UMass), Narrative summarization. Goal orientations of the characters and the interactions between them. "Affect state map". Extract recognizable patterns of interaction called "plot units". Summary based on how these plot units are linked together. From this summary structure a natural-language summary of the original can be generated. 12:30PM Lunch, during which Brown's teaching lab, equipped with 55 Apollos, was demonstrated. 2:00PM Panel discussion, Drew McDermott (Yale), Randy Ellis (UMass), Tomas Lozano-Perez (MIT), Mallory Selfridge (UConn), "AI and Robotics". DMcD contemplated the effect that the realization of a walking, talking, perceiving robot would have on AI. He remarked how current robotics work does entail a lot of AI, but that there is necessary, robotics- -specific, ground-work (like matrices, a code-word for "much mathematics"). All the other panelists had a similar view of this inter-relation between robotics and AI. The other panelists then sketched robotics work being done at their respective institutions. RE: Integration of vision and touch, using a reasonable world model, some simple planning, and feedback during the process. Cartesian robot, gripper, Ken Overton's tactile array sensor (force images), controllable camera, Salisbury hand. Need for AI in robotics, especially object representation and search. Learning -- a big future issue for a robot that actually moves about in the world. Problems of implementing algorithms in real time. For getting started in robotics: kinematics, materials science, control theory, AI techniques, but how much of each depends on what you want to do in robotics. TL-P: A comparatively lengthy talk on "Automatic synthesis of fine motion strategies", best exemplified by the problem of putting a peg into a hole. Given the inherent uncertainty in all postions and motions, the best strategy (which we probably all do intuitively) is to aim the peg just to one side of the hole, sliding it across into the hole when it hits, grazing the far side of the hole as it goes down. A method for generating such a strategy automatically, using a formalism based on configuration spaces, generalized dampers, and friction cones. MS: Plans for commanding a robot in natural language, and for describing things to it, and for teaching it how to do things by showing it examples (from which the robot builds an abstract description, usable in other situations). A small, but adequate robotics facility. Afterwards, an open discussion, during which was stressed how important it is that the various far-flung branches of AI be more aware of each other, and not become insular. Regarding robotics research, all panelists agreed strongly that it was absolutely necessary to work with real robot hardware; software simulations could not hope to capture all the pernickety richness of the world, motion, forces, friction, slippage, uncertainty, materials, bending, spatial location, at least not in any computationally practical way. No substitute for reality! 3:30PM More short talks Jim Hendler (Brown), an overview of things going on at Brown, and in the works. Natural language (story comprehension). FRAIL (frame-based knowledge representation). NASL (problem solving). An electronic repair manual, which generates instructions for repairs as needed from an internal model, hooked up with a graphics and 3D modelling system. And in the works: expert systems, probabilistic reasoning, logic programming, problem solving, parallel computation (in particular marker-passing and BOLTZMANN-style machines). Brown is looking for a new AI faculty member. [Not a job ad, just a report of one!] David Miller (Yale), "Uncertain planning through uncertain territory". How to get from A to B if your controls and sensors are unreliable. Find a path to your goal, along the path select checkpoints (landmarks), adjust the path to go within eye-shot of the checkpoints, then off you go, running demons to watch out for checkpoints and raise alarms if they don't appear when expected. This means you're lost. Then you generate hypotheses about where you are now (using your map), and what might have gone wrong to get you there (based on a self-model). Verify one (some? all?) of these hypotheses by looking around. Patch your plan to get back to an appro- priate checkpoint. Verify the whole process by getting back on the beaten track. Apparently there's a real Hero robot that cruises about a room doing this. Bud Crawley (GTE) described what was going on at GTE Labs in AI. Know- ledge-based systems. Natural-language front-end for data bases. Distributed intelligence. Machine learning. Bill Taylor (Gould Inc.), gave an idea of what applied AI research means to his company, which (in his division) makes digital controllers for running machines out on the factory floor. Currently, an expert system for repairing these controllers in the field. [I'm not sure how far along in being realized this was, I think very little.] For the future, a big, smart system that would assist a human operator in managing the hundreds of such controllers out on the floor of a decent sized factory. Graeme Hirst (Brown, soon Toronto), "Artificial Digestion". Artificial Intelligence attempts to model a very poorly understood system, the human cognitive system. Much more immediate and substantial results could be obtained by modelling a much better understood system, the human digestive system. Examples of the behavior of a working prototype system on simulated food input, drawn from a number of illustrative food-domains, including a four-star French restaurant and a garbage pail. Applications of AD: automatic restaurant reviewing, automatic test-marketing of new food products, and vicarious eating for the diet-conscious and orally impaired. [Forget about expert systems; this is the hot new area for the 80's!] 4:30PM AISNE Business Meeting (Yes, some of us stayed till the end!) Next year's meeting will held at Boston University. The position of programme chairman is still open. A Final Remark: All the above is based on my own notes of the conference. At the very least it reflects my own interests and pre-occupations. Considering the disorganized state of my notes, and the late hour I'm typing this, a lot of the above may be just wrong. My apologies to anyone I've misrepresented; by all means correct me. I hope the general interest of this report to the AI community outweighs all these failings. LJK =========================================================================== ------------------------------ End of AIList Digest ******************** 14-Nov-83 08:55:06-PST,10369;000000000001 Mail-From: LAWS created at 14-Nov-83 08:54:25 Date: Monday, November 14, 1983 8:48AM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #96 To: AIList@SRI-AI AIList Digest Monday, 14 Nov 1983 Volume 1 : Issue 96 Today's Topics: Theory - Parallel Systems, Looping Problem in Literature, Intelligence ---------------------------------------------------------------------- Date: 8 Nov 83 23:03:04-PST (Tue) From: pur-ee!uiucdcs!uokvax!andree @ Ucb-Vax Subject: Re: Infinite loops and Turing machines.. - (nf) Article-I.D.: uiucdcs.3712 /***** uokvax:net.ai / umcp-cs!speaker / 9:41 pm Nov 1, 1983 */ Aha! I knew someone would come up with this one! Consider that when we talk of simultaneous events... we speak of simultaneous events that occur within one Turing machine state and outside of the Turing machine itself. Can a one-tape Turing machine read the input of 7 discrete sources at once? A 7 tape machine with 7 heads could! /* ---------- */ But I can do it with a one-tape, one-head turing machine. Let's assume that each of your 7 discrete sources can always be represeted in n bits. Thus, the total state of all seven sources can be represented in 7*n bits. My one-tape turing machine has 2 ** (7*n) symbols, so it can handle your 7 sources, each possible state of all 7 being one symbol of input. One of the things I did in an undergraduate theory course was show that an n-symbol turing machine is no more powerful than a two-symbol turing machine for any finite (countable?) n. You just loose speed. Subject: parallel vs. sequential An excellent treatise on how some parallel machines are more powerful than all sequential machines can be found in Will Clinger's doctoral dissertation "Foundations of Actor Semantics" which can be obtained by sending $7 to Publications Office MIT Artificial Intelligence Laboratory 545 Technology Square Cambridge, Mass. 02139 requesting Technical Report 633 dated May 1981. ------------------------------ Date: Fri 11 Nov 83 17:12:08-PST From: Wilkins Subject: parallelism and turing machines Regarding the "argument" that parallel algorithms cannot be run serially because a Turing machine cannot react to things that happen faster than the time it needs to change states: clearly, you need to go back to whoever sold you the Turing machine for this purpose and get a turbocharger for it. Seriously, I second the motion to move towards more useful discussions. ------------------------------ Date: 9 Nov 83 19:28:21-PST (Wed) From: ihnp4!cbosgd!mhuxl!ulysses!unc!mcnc!ncsu!uvacs!mac @ Ucb-Vax Subject: the halting problem in history Article-I.D.: uvacs.1048 If there were any 'subroutines' in the brain that could not halt... I'm sure they would have been found and bred out of the species long ago. I have yet to see anyone die from an infinite loop. (umcp-cs.3451) There is such. It is caused by seeing an object called the Zahir. One was a Persian astrolabe, which was cast into the sea lest men forget the world. Another was a certain tiger. Around 1900 it was a coin in Buenos Aires. Details in "The Zahir", J.L.Borges. ------------------------------ Date: 8 Nov 83 16:38:29-PST (Tue) From: decvax!wivax!linus!vaxine!wjh12!foxvax1!brunix!rayssd!asa @ Ucb-Vax Subject: Re: Inscrutable Intelligence Article-I.D.: rayssd.233 The problem with a psychological definition of intelligence is in finding some way to make it different from what animals do, and cover all of the complex things that huumans can do. It used to be measured by written test. This was grossly unfair, so visual tests were added. These tend to be grossly unfair because of cultural bias. Dolphins can do very "intelligent" things, based on types of "intelligent behavior". The best definition might be based on the rate at which learning occurs, as some have suggested, but that is also an oversimplification. The ability to deduce cause and effect, and to predict effects is obviously also important. My own feeling is that it has something to do with the ability to build a model of yourself and modify yourself accordingly. It may be that "I conceive" (not "I think"), or "I conceive and act", or "I conceive of conceiving" may be as close as we can get. ------------------------------ Date: 8 Nov 83 23:02:53-PST (Tue) From: pur-ee!uiucdcs!uokvax!rigney @ Ucb-Vax Subject: Re: Parallelism & Consciousness - (nf) Article-I.D.: uiucdcs.3711 Perhaps something on the order of "Intelligence enhances survivability through modification of the environment" is in order. By modification something other than the mere changes brought about by living is indicated (i.e. Rise in CO2 levels, etc. doesn't count). Thus, if Turtles were intelligent, they would kill the baby rabbits, but they would also attempt to modify the highway to present less of a hazard. Problems with this viewpoint: 1) It may be confusing Technology with Intelligence. Still, tool making ability has always been a good sign. 2) Making the distinction between Intelligent modifications and the effect of just being there. Since "conscious modification" lands us in a bigger pit of worms than we're in now, perhaps a distinction should be drawn between reactive behavior (reacting and/or adapting to changes) and active behavior (initiating changes). Initiative is therefore a factor. 3) Monkeys make tools(Antsticks), Dolphins don't. Is this an indication of intelligence, or just a side-effect of Monkeys having hands and Dolphins not? In other words, does Intelligence go away if the organism doesn't have the means of modifying its environment? Perhaps "potential" ability qualifies. Or we shouldn't consider specific instances (Is a man trapped in a desert still intelligent, even if he has no way to modify his environment.) Does this mean that if you had a computer with AI, and stripped its peripherals, it would lose intelligence? Are human autistics intelligent? Or are we only considering species, and not representatives of species? In the hopes that this has added fuel to the discussion, Carl ..!ctvax!uokvax!rigney ..!duke!uok!uokvax!rigney ------------------------------ Date: 8 Nov 83 20:51:15-PST (Tue) From: pur-ee!uiucdcs!uicsl!dinitz @ Ucb-Vax Subject: Re: RE:intelligence and adaptability - (nf) Article-I.D.: uiucdcs.3746 Actually, SHRDLU had neither hand nor eye -- only simulations of them. That's a far cry from the real thing. ------------------------------ Date: 9 Nov 83 16:20:10-PST (Wed) From: ihnp4!houxm!mhuxl!ulysses!unc!mcnc!ncsu!uvacs!mac @ Ucb-Vax Subject: inscrutable intelligence Article-I.D.: uvacs.1047 Regarding inscrutability of intelligence [sri-arpa.13363]: Actually, it's typical that a discipline can't define its basic object of study. Ever heard a satisfactory definition of mathematics (it's not just the consequences of set theory) or philosophy.? What is physics? Disciplines are distinguished from each other for historical and methodological reasons. When they can define their subject precisely it is because they have been superseded by the discipline that defines their terms. It's usually not important (or possible) to define e.g. intelligence precisely. We know it in humans. This is where the IQ tests run into trouble. AI seems to be about behavior in computers that would be called intelligent in humans. Whether the machines are or are not intelligent (or, for that matter, conscious) is of little interest and no import. In this I guess I agree with Rorty [sri-arpa.13322]. Rorty is willing to grant consciousness to thermostats if it's of any help. (Best definition of formal mathematics I know: "The science where you don't know what you're talking about or whether what you're saying is true".) A. Colvin mac@virginia ------------------------------ Date: 12 Nov 83 0:37:48-PST (Sat) From: decvax!genrad!security!linus!utzoo!utcsstat!laura @ Ucb-Vax Subject: Re: Parallelism & Consciousness - (nf) Article-I.D.: utcsstat.1420 The other problem with the "turtles should be killing baby rabbits" definition of intelligence is that it seems to imply that killing (or at least surviving) is an indication of intelligence. i would rather not believe this, unless there is compelling evidence that the 2 are related. So far I have not seen the evidence. Laura Creighton utcsstat!laura ------------------------------ Date: 20 Nov 83 0:24:46-EST (Sun) From: pur-ee!uiucdcs!trsvax!karl @ Ucb-Vax Subject: Re: Slow Intelligence - (nf) Article-I.D.: uiucdcs.3789 " .... I'm not at all sure that people's working definition of intelligence has anything at all to do with either time or survival. " Glenn Reid I'm not sure that people's working definition of intelligence has anything at all to do with ANYTHING AT ALL. The quoted statement implies that peoples' working definition of intelligence is different - it is subjective to each individual. I would like to claim that each individual's working definition of intelligence is sub- ject to change also. What we are working with here is conceptual.. not a tangible ob- ject which we can spot at an instance. If the object is concep- tual, and therefore subjective, then it seems that we can (and probably will) change it's definition as our collective experi- ences teach us differently. Karl T. Braun ...ctvax!trsvax!karl ------------------------------ End of AIList Digest ******************** 14-Nov-83 09:08:18-PST,15676;000000000001 Mail-From: LAWS created at 14-Nov-83 09:06:03 Date: Monday, November 14, 1983 8:59AM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #97 To: AIList@SRI-AI AIList Digest Monday, 14 Nov 1983 Volume 1 : Issue 97 Today's Topics: Pattern Recognition - Vector Fields, Psychology - Defense, Ethics - AI Responsibilities, Seminars - NRL & Logic Specifications & Deductive Belief ---------------------------------------------------------------------- Date: Sun, 13 Nov 83 19:25:40 PST From: Philip Kahn Subject: Need references in field of spatial pattern recognition This letter to AI-LIST is a request for references from all of you out there that are heavily into spatial pattern recognition. First let me explain my approach, then I'll hit you with my request. Optical flow and linear contrast edges have been getting a lot of attention recently. Utilizing this approach, I view a line as an ordered set of [image] elements; that is, a line is comprised of a finite ordered set of elements. Each element of a line is treated as a directed line (a vector with direction and magnitude). Here's what I am trying to define: with such a definition of a line, it should be possible to create mappings between lines to form fairly abstract ideas of similarity between lines. Since objects are viewed as a particular arrangement of lines, this analysis would suffice in identifying objects as being alike. Some examples, the two lines possessing the most similarities (i.e., MAX ( LINE1 .intersection. LINE2 ) ) may be one criterion of comparison. I'm looking for any references you might have on this area. This INCLUDES: 1) physiology/biology/neuroanatomy articals dealing with functional mappings from the ganglion to any level of cortical processing. 2) fuzzy set theory. This includes ordered set theory and any and all applications of set theory to pattern recognition. 3) any other pertinent references I would greatly appreciate any references you might provide. After a week or two, I will compile the references and put them on the AI-LIST so that we all can use them. Viva la effort! Philip Kahn [My correspondence with Philip indicates that he is already familiar with much of the recent literature on optic flow. He has found little, however, on the subject of pattern recognition in vector fields. Can anyone help? -- KIL] ------------------------------ Date: Sun, 13 Nov 1983 22:42 EST From: HEWITT%MIT-OZ@MIT-MC.ARPA Subject: Rational Psychology [and Reply] Date: 28 Sep 83 10:32:35-PDT (Wed) To: AIList at MIT-MC From: decvax!duke!unc!mcnc!ncsu!fostel @ Ucb-Vax Subject: RE: Rational Psychology [and Reply] ... Is psychology rational? Someone said that all sciences are rational, a moot point, but not that relevant unless one wishes to consider Psychology a science. I do not. This does not mean that psychologists are in any way inferior to chemists or to REAL scientists like those who study physics. But I do think there .... ----GaryFostel---- This is an old submission, but having just read it I felt compelled to reply. I happen to be a Computer Scientist, but I think Psychologists, especially Experimental Psychologists, are better scientists than the average Computer "Scientist". At least they have been trained in the scientific method, a skill most Computer Scientists lack. Just because Psychologist, by and large, cannot defend themselves on this list is no reason to make idle attacks with only very superficial knowledge on the subject. Fanya Montalvo ------------------------------ Date: Sun 13 Nov 83 13:14:06-PST From: David Rogers Subject: just a reminder... Artificial intelligence promises to alter the world in enormous ways during our lifetime; I believe it's crucial for all of us to look forward to the effects our our work, both individually and collectively, to make sure that it will be to the benefit of all peoples in the world. It seems to be tiresome to people to remind them of the incredible effect that AI will have in our lifetimes, yet the profound mature of the changes to the world made by a small group of researchers makes it crucial that we don't treat our efforts casually. For example, the military applications of AI will dwarf that of the atomic bomb, but even more important is the fact that the atomic bomb is a primarily military device, while AI will impact the world as much (if not more) in non-military domains. Physics in the early part of this century was at the cutting edge of knowledge, similar to the current place of AI. The culmination of their work in the atomic bomb changed their field immensely and irrevocably; even on a personal level, researchers in physics found their lives greatly impacted, often shattered. Many of the top researchers left the field. During our lifetimes I think we will see a similar transformation, with the "fun and games" of these heady years turning into a deadly seriousness, I think we will also see top researchers leaving the field, once we start to see some of our effects on the world. It is imperative for all workers in this field to formulate and share a moral outlook on what we do, and hope to do, to the world. I would suggest we have, at the minimum, a three part responsibility. First, we must make ourselves aware of the human impact of our work, both short and long term. Second, we must use this knowledge to guide the course of our research, both individually and collectively, rather than simply flowing into whatever area the grants are flowing into. Third and most importantly, we must be spokespeople and consciences to the world, forcing others to be informed of what we are doing and its effects. Researches who still cling to "value-free" science should not be working in AI. I will suggest a few areas we should be thinking about: - Use of AI for offensive military use vs. legitimate defense needs. While the line is vague, a good offense is surely not always the best defense. - Will the work cause a centralization of power, or cause a decentralization of power? Building massive centers of power in this age increases the risk of humans dominated by machine. - Is the work offering tools to extend the grasp of humans, or tools to control humans? - Will people have access to the information generated by the work, or will the benefits of information access be restricted to a few? Finally, will the work add insights into ourselves a human beings, or will it simply feed our drives, reflecting our base nature back at ourselves? In the movie "Tron" an actor says "Our spirit remains in each and every program we wrote"; what IS our spirit? David ------------------------------ Date: 8 Nov 1983 09:44:28-PST From: Elaine Marsh Subject: AI Seminar Schedule [I am passing this along because it is the first mention of this seminar series in AIList and will give interested readers the chance to sign up for the mailing list. I will not continue to carry these seminar notices because they do not include abstracts. -- KIL] U.S. Navy Center for Applied Research in Artificial Intelligence Naval Research Laboratory - Code 7510 Washington, DC 20375 WEEKLY SEMINAR SERIES 14 Nov. 1983 Dr. Jagdish Chandra, Director Mathematical Sciences Division Army Research Office, Durham, NC "Mathematical Sciences Activities Relating to AI and Its Applications at the Army Research Office" 21 Nov. 1983 Professor Laveen Kanal Department of Computer Science University of Maryland, College Park, MD "New Insights into Relationships among Heuristic Search, Dynamic Programming, and Branch & Bound Procedures" 28 Nov. 1983 Dr. William Gale Bell Labs Murray Hill, NJ "An Expert System for Regression Analysis: Applying A.I. Ideas in Statistics" 5 Dec. 1983 Professor Ronald Cole Department of Computer Science Carnegie-Mellon University, Pittsburgh, PA "What's New in Speech Recognition?" 12 Dec. 1983 Professor Robert Haralick Department of Electrical Engineering Virginia Polytechnic Institute, Blacksburg, VA "Application of AI Techniques to the Interpretation of LANDSAT Scenes over Mountainous Areas" Our meeting are usually held Monday mornings at 10:00 a.m. in the Conference Room of the Navy Center for Applied Research in Artificial Intelligence (Bldg. 256) located on Bolling Air Force Base, off I-295, in the South East quadrant of Washington, DC. Coffee will be available starting at 9:45 a.m. If you would like to speak, or be added to our mailing list, or would just like more information contact Elaine Marsh at marsh@nrl-aic [(202)767-2382]. ------------------------------ Date: Mon 7 Nov 83 15:20:15-PST From: Sharon Bergman Subject: Ph.D. Oral [Reprinted from the SU-SCORE bboard.] Ph.D. Oral COMPILING LOGIC SPECIFICATIONS FOR PROGRAMMING ENVIRONMENTS November 16, 1983 2:30 p.m., Location to be announced Stephen J. Westfold A major problem in building large programming systems is in keeping track of the numerous details concerning consistency relations between objects in the domain of the system. The approach taken in this thesis is to encourage the user to specify a system using very-high-level, well-factored logic descriptions of the domain, and have the system compile these into efficient procedures that automatically maintain the relations described. The approach is demonstrated by using it in the programming environment of the CHI Knowledge-based Programming system. Its uses include describing and implementing the database manager, the dataflow analyzer, the project management component and the system's compiler itself. It is particularly convenient for developing knowledge representation schemes, for example for such things as property inheritance and automatic maintenance of inverse property links. The problem description using logic assertions is treated as a program such as in PROLOG except that there is a separation of the assertions that describe the problem from assertions that describe how they are to be used. This factorization allows the use of more general logical forms than Horn clauses as well as encouraging the user to think separately about the problem and the implementation. The use of logic assertions is specified at a level natural to the user, describing implementation issues such as whether relations are stored or computed, that some assertions should be used to compute a certain function, that others should be treated as constraints to maintain the consistency of several interdependent stored relations, and whether assertions should be used at compile- or execution-time. Compilation consists of using assertions to instantiate particular procedural rule schemas, each one of which corresponds to a specialized deduction, and then compiling the resulting rules to LISP. The rule language is a convenient intermediate between the logic assertion language and the implementation language in that it has both a logic interpretation and a well-defined procedural interpretation. Most of the optimization is done at the logic level. ------------------------------ Date: Fri 11 Nov 83 09:56:17-PST From: Sharon Bergman Subject: Ph.D. Oral [Reprinted from the SU-SCORE bboard.] Ph.D. Oral Tuesday, Nov. 15, 1983, 2:30 p.m. Bldg. 170 (history corner), conference room A DEDUCTIVE MODEL OF BELIEF Kurt Konolige Reasoning about knowledge and belief of computer and human agents is assuming increasing importance in Artificial Intelligence systems in the areas of natural language understanding, planning, and knowledge representation in general. Current formal models of belief that form the basis for most of these systems are derivatives of possible- world semantics for belief. However,, this model suffers from epistemological and heuristic inadequacies. Epistemologically, it assumes that agents know all the consequences of their belief. This assumption is clearly inaccurate, because it doesn't take into account resource limitations on an agent's reasoning ability. For example, if an agent knows the rules of chess, it then follows in the possible- world model that he knows whether white has a winning strategy or not. On the heuristic side, proposed mechanical deduction procedures have been first-order axiomatizations of the possible-world belief. A more natural model of belief is a deductions model: an agent has a set of initial beliefs about the world in some internal language, and a deduction process for deriving some (but not necessarily all) logical consequences of these beliefs. Within this model, it is possible to account for resource limitations of an agent's deduction process; for example, one can model a situation in which an agent knows the rules of chess but does not have the computational resources to search the complete game tree before making a move. This thesis is an investigation of Gentzen-type formalization of the deductive model of belief. Several important original results are proven. Among these are soundness and completeness theorems for a deductive belief logic; a corespondence result that shows the possible- worlds model is a special case of the deduction model; and a model analog ot Herbrand's Theorem for the belief logic. Several other topics of knowledge and belief are explored in the thesis from the viewpoint of the deduction model, including a theory of introspection about self-beliefs, and a theory of circumscriptive ignorance, in which facts an agent doesn't know are formalized by limiting or circumscribing the information available to him. Here it is! ------------------------------ End of AIList Digest ******************** 15-Nov-83 10:31:37-PST,15081;000000000001 Mail-From: LAWS created at 15-Nov-83 10:28:54 Date: Tuesday, November 15, 1983 10:21AM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #98 To: AIList@SRI-AI AIList Digest Tuesday, 15 Nov 1983 Volume 1 : Issue 98 Today's Topics: Intelligence - Definitions & Metadiscussion, Looping Problem, Architecture - Parallelism vs. Novel Architecture, Pattern Recognition - Optic Flow & Forced Matching, Ethics & AI, Review - Biography of Turing ---------------------------------------------------------------------- Date: 14 Nov 1983 15:03-PST From: fc%usc-cse%USC-ECL@SRI-NIC Subject: Re: AIList Digest V1 #96 An intelligent race is one with a winner, not one that keeps on rehashing the first 5 yards till nobody wants to watch it anymore. FC ------------------------------ Date: 14 Nov 83 10:22:29-PST (Mon) From: ihnp4!houxm!mhuxl!ulysses!unc!mcnc!ncsu!fostel @ Ucb-Vax Subject: Intelligence and Killing Article-I.D.: ncsu.2396 Someone wondered if there was evidence that intelligence was related to the killing off of other animals. Presumably that person is prepared to refute the apparant similtaneous claims of man as the most intelligent and the most deadly animal. Personally, I might vote dolphins as more intelligent, but I bet they do their share of killing too. They eat things. ----GaryFostel---- ------------------------------ Date: 14 Nov 83 14:01:55-PST (Mon) From: ihnp4!ihuxv!portegys @ Ucb-Vax Subject: Behavioristic definition of intelligence Article-I.D.: ihuxv.584 What is the purpose of knowing whether something is intelligent? Or has a soul? Or has consciousness? I think one of the reasons is that it makes it easier to deal with it. If a creature is understood to be a human being, we all know something about how to behave toward it. And if a machine exhibits intelligence, the quintessential quality of human beings, we also will know what to do. One of the things that this implies is that we really should not worry too much about whether a machine is intelligent until one gets here. The definition of it will be in part determined by how we behave toward it. Right now, I don't feel very confused about how to act in the presence of a computer running an AI program. Tom Portegys, Bell Labs IH, ihuxv!portegys ------------------------------ Date: 12 Nov 83 19:38:02-PST (Sat) From: decvax!decwrl!flairvax!kissell @ Ucb-Vax Subject: Re: the halting problem in history Article-I.D.: flairvax.267 "...If there were any subroutines in the brain that did not halt..." It seems to me that there are likely large numbers of subroutines in the brain that aren't *supposed* to halt. Like breathing. Nothing wrong with that; the brain is not a metaphor for a single-instruction-stream processor. I've often suspected, though, that some pathological states, depression, obsession, addiction, etcetera can be modeled as infinite loops "executed" by a portion of the brain, and thus why "shock" treatments sometimes have beneficial effects on depression; a brutal "reset" of the whole "system". ------------------------------ Date: Tue, 15 Nov 83 07:58 PST From: "Glasser Alan"@LLL-MFE.ARPA Subject: parallelism vs. novel architecture There has been a lot of discussion in this group recently about the role of parallelism in artificial intelligence. If I'm not mistaken, this discussion began in response to a message I sent in, reviving a discussion of a year ago in Human-Nets. My original message raised the question of whether there might exist some crucial, hidden, architectural mechanism, analogous to DNA in genetics, which would greatly clarify the workings of intelligence. Recent discussions have centered on the role of parallelism alone. I think this misses the point. While parallelism can certainly speed things up, it is not the kind of fundamental departure from past practices which I had in mind. Perhaps a better example would be Turing's and von Neumann's concept of the stored-program computer, replacing earlier attempts at hard-wired computers. This was a fundamental break- through, without which nothing like today's computers could be practical. Perhaps true intelligence, of the biological sort, requires some structural mechanism which has yet to be imagined. While it's true that a serial Turing machine can do anything in principle, it may be thoroughly impractical to program it to be truly intelligent, both because of problems of speed and because of the basic awkwardness of the architecture. What is hopelessly cumbersome in this architecture may be trivial in the right one. I know this sounds pretty vague, but I don't think it's meaningless. ------------------------------ Date: Mon 14 Nov 83 17:59:07-PST From: David E.T. Foulser Subject: Re: AIList Digest V1 #97 There is a paper by Kruskal on multi-dimensional scaling that might be of interest to the user interested in vision processing. I'm not too clear on what he's doing, so this could be off-base. Dave Foulser ------------------------------ Date: Mon 14 Nov 83 22:24:45-MST From: Stanley T. Shebs Subject: Pattern Matchers Thanks for the replies about loop detection; some food for thought in there... My next puzzle is about pattern matchers. Has anyone looked carefully at the notion of a "non-failing" pattern matcher? By that I mean one that never or almost never rejects things as non-matching. Consider a database of assertions (or whatever) and the matcher as a search function which takes a pattern as argument. If something in the db matches the pattern, then it is returned. At this point, the caller can either accept or reject the item from the db. If rejected, the matcher would be called again, to find something else matching, and so forth. So far nothing unusual. The matcher will eventually signal utter failure, and that there is nothing satisfactory in the database. My idea is to have the matcher constructed in such a way that it will return things until the database is entirely scanned, even if the given pattern is a very simple and rigid one. In other words, the matcher never gives up - it will always try to find the most tenuous excuse to return a match. Applications I have in mind: NLP for garbled and/or incomplete sentences, and creative thinking (what does a snake with a tail in its mouth have to do with benzene? talk about tenuous connections!). The idea seems related to fuzzy logic (an area I am sadly ignorant of), but other than that, there seems to be no work on the idea (perhaps it's a stupid one?). There seem to be two main problems - organizing the database in such a way that the matcher can easily progress from exact matches to extremely remote ones (can almost talk about a metric space of assertions!), and setting up the matcher's caller so as not to thrash too badly (example: a parser may have trouble deciding whether a sentence is grammatically incorrect or a word's misspelling looks like another word, if the word analyzer has a nonfailing matcher). Does anybody know anything about this? Is there a fatal flaw somewhere? Stan Shebs BTW, a frame-based system can be characterized as a semantic net (if you're willing to mung concepts!), and a semantic net can be mapped into an undirected graph, which *is* a metric space. ------------------------------ Date: 14 November 1983 1359-PST (Monday) From: crummer at AEROSPACE (Charlie Crummer) Subject: Ethics and AI Research Dave Rogers brought up the subject of ethics in AI research. I agree with him that we must continually evaluate the projects we are asked to work on. Unfortunately, like the example he gave of physicists working on the bombs, we will not always know what the government has in mind for our work. It may be valid to indict the workers on the Manhattan project because they really did have an idea what was going on but the very early researchers in the field of radioactivity probably did not know how their discoveries would be used. The application of morality must go beyond passively choosing not to work on certain projects. We must become actively involved in the application by our government of the ideas we create. Once an idea or physical effect is discovered it can never be undiscovered. If I choose not to work on a project (which I definitely would if I thought it immoral) that may not make much difference. Someone else will always be waiting to pick up the work. It is sort of like preventing rape by refusing to rape anyone. --Charlie ------------------------------ Date: 14 Nov 83 1306 PST From: Russell Greiner Subject: Biography of Turing [Reprinted from the SU-SCORE bboard.] n055 1247 09 Nov 83 BC-BOOK-REVIEW (UNDATED) By CHRISTOPHER LEHMANN-HAUPT c. 1983 N.Y. Times News Service ALAN TURING: The Enigma. By Andrew Hodges. 587 pages. Illustrated. Simon & Schuster. $22.50. He is remembered variously as the British cryptologist whose so-called ''Enigma'' machine helped to decipher Germany's top-secret World War II code; as the difficult man who both pioneered and impeded the advance of England's computer industry; and as the inventor of a theoretical automaton sometimes called the ''Turing (Editors: umlaut over the u) Machine,'' the umlaut being, according to a glossary published in 1953, ''an unearned and undesirable addition, due, presumably, to an impression that anything so incomprehensible must be Teutonic.'' But this passionately exhaustive biography by Andrew Hodges, an English mathematician, brings Alan Turing very much back to life and offers a less forbidding impression. Look at any of the many verbal snapshots that Hodges offers us in his book - Turing as an eccentrically unruly child who could keep neither his buttons aligned nor the ink in his pen, and who answered his father when asked if he would be good, ''Yes, but sometimes I shall forget!''; or Turing as an intense young man with a breathless high-pitched voice and a hiccuppy laugh - and it is difficult to think of him as a dark umlauted enigma. Yet the mind of the man was an awesome force. By the time he was 24 years old, in 1936, he had conceived as a mathematical abstraction his computing machine and completed the paper ''Computable Numbers,'' which offered it to the world. Thereafter, Hodges points out, his waves of inspiration seemed to flow in five-year intervals - the Naval Enigma in 1940, the design for his Automatic Computing Engine (ACE) in 1945, a theory of structural evolution, or morphogenesis, in 1950. In 1951, he was elected a Fellow of the Royal Society. He was not yet 40. But the next half-decade interval did not bring further revelation. In February 1952, he was arrested, tried, convicted and given a probationary sentence for ''Gross Indecency contrary to Section 11 of the Criminal Law Amendment Act 1885,'' or the practice of male homosexuality, a ''tendency'' he had never denied and in recent years had admitted quite openly. On June 7, 1954, he was found dead in his home near Manchester, a bitten, presumably cyanide-laced apple in his hand. Yet he had not been despondent over his legal problems. He was not in disgrace or financial difficulty. He had plans and ideas; his work was going well. His devoted mother - about whom he had of late been having surprisingly (to him) hostile dreams as the result of a Jungian psychoanalysis - insisted that his death was the accident she had long feared he would suffer from working with dangerous chemicals. The enigma of Alan Mathison Turing began to grow. Andrew Hodges is good at explaining Turing's difficult ideas, particularly the evolution of his theoretical computer and the function of his Enigma machines. He is adept at showing us the originality of Turing's mind, especially the passion for truth (even when it damaged his career) and the insistence on bridging the worlds of the theoretical and practical. The only sections of the biography that grow tedious are those that describe the debates over artificial intelligence - or maybe it's the world's resistance to artificial intelligence that is tedious. Turing's position was straightforward enough: ''The original question, 'Can machines think?' I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.'' On the matter of Turing's suicide, Hodges concedes its incomprehensibility, but then announces with sudden melodrama: ''The board was ready for an end game different from that of Lewis Carroll's, in which Alice captured the Red Queen, and awoke from nightmare. In real life, the Red Queen had escaped to feasting and fun in Moscow. The White Queen would be saved, and Alan Turing sacrificed.'' What does Hodges mean by his portentous reference to cold-war politics? Was Alan Turing a murdered spy? Was he a spy? Was he the victim of some sort of double-cross? No, he was none of the above: the author is merely speculating that as the cold war heated up, it must have become extremely dangerous to be a homosexual in possession of state secrets. Hodges is passionate on the subject of the precariousness of being homosexual; it was partly his participation in the ''gay liberation'' movement that got him interested in Alan Turing in the first place. Indeed, one has to suspect Hodges of an overidentification with Alan Turing, for he goes on at far too great length on Turing's existential vulnerability. Still, word by word and sentence by sentence, he can be exceedingly eloquent on his subject. ''He had clung to the simple amidst the distracting and frightening complexity of the world,'' the author writes of Turing's affinity for the concrete. ''Yet he was not a narrow man,'' Hodges continues. ''Mrs. Turing was right in saying, as she did, that he died while working on a dangerous experiment. It was the experiment called LIFE - a subject largely inducing as much fear and embarrassment for the official scientific world as for her. He had not only thought freely, as best he could, but had eaten of two forbidden fruits, those of the world and of the flesh. They violently disagreed with each other, and in that disagreement lay the final unsolvable problem.'' ------------------------------ End of AIList Digest ******************** 16-Nov-83 14:35:20-PST,18538;000000000001 Mail-From: LAWS created at 16-Nov-83 14:33:27 Date: Wednesday, November 16, 1983 2:25PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #99 To: AIList@SRI-AI AIList Digest Thursday, 17 Nov 1983 Volume 1 : Issue 99 Today's Topics: AI Literature - Comtex, Review - Abacus, Artificial Humanity, Conference - SPIE Call for Papers, Seminar - CRITTER for Critiquing Circuit Designs, Military AI - DARPA Plans (long message) ---------------------------------------------------------------------- Date: Wed 16 Nov 83 10:14:02-PST From: Ken Laws Subject: Comtex The Comtex microfiche series seems to be alive and well, contrary to a rumor printed in an early AIList issue. The ad they sent me offers the Stanford and MIT AI memoranda (over $2,000 each set), and says that the Purdue PRIP [pattern recognition and image processing] technical reports will be next. Also forthcoming are the SRI and Carnegie-Mellon AI reports. -- Ken Laws ------------------------------ Date: Wed 16 Nov 83 10:31:26-PST From: Ken Laws Subject: Abacus I have the first issue of Abacus, the new "soft" computer science magazine edited by Anthony Ralston. It contains a very nice survey or introduction to computer graphics for digital filmmaking and an interesting exploration of how the first electronic digital computer came to be. There is also a superficial article about computer vision which fails to answer its title question, "Why Computers Can't See (Yet)". [It is possibly that I'm being overly harsh since this is my own area of expertise. My feeling, however, is that the question cannot be answered by just pointing out that vision is difficult and that we have dozens of different approaches, none of which works in more than specialized cases. An adequate answer requires a guess at how it is that the human vision system can work in all cases, and why we have not been able to duplicate it.] The magazine also offers various computer-related departments, notably those covering book reviews, the law, personal computing, puzzles, and politics. Humorous anecdotes are solicited for filler material, a la Reader's Digest. There is no AI-related column at present. The magazine has a "padded" feel, particularly since every ad save one is by Springer-Verlag, the publisher. They even ran out of things to advertise and so repeated several full-page ads. No doubt this is a new-issue problem and will quickly disappear. I wish them well. -- Ken Laws ------------------------------ Date: 16 Nov 1983 10:21:32 EST (Wednesday) From: Mark S. Day Subject: Artificial Humanity From: ihnp4!ihuxv!portegys @ Ucb-Vax Subject: Behavioristic definition of intelligence What is the purpose of knowing whether something is intelligent? Or has a soul? Or has consciousness? I think one of the reasons is that it makes it easier to deal with it. If a creature is understood to be a human being, we all know something about how to behave toward it. And if a machine exhibits intelligence, the quintessential quality of human beings, we also will know what to do. Without wishing to flame or start a pointless philosophical discussion, I do not consider intelligence to be the quintessential quality of human beings. Nor do I expect to behave in the same way towards an artificially intelligent program as I would towards a person. Turing tests etc. notwithstanding, I think there is a distinction between "artificial intelligence" and "artificial humanity," and that by and large people are not striving to create "artificial humanity." ------------------------------ Date: Wed 16 Nov 83 09:30:18-PST From: Ken Laws Subject: Artificial Humanity I attended a Stanford lecture by Doug Lenat on Tuesday. He mentioned three interesting bugs that developed in EURISKO, a self-monitoring and self-modifying program. One turned up when EURISKO erroneously claimed to have discovered a new type of flip-flop. The problem was traced to an array indexing error. EURISKO, realizing that it had never in its entire history had a bounds error, had deleted the bounds-checking code. The first bounds error occurred soon after. Another bug cropped up in the "credit assignment" rule base. EURISKO was claiming that a particular rule had been responsible for discovering a great many other interesting rules. It turned out that the gist of the rule was "If the system discovers something interesting, attach my name as the discoverer." The third bug became evident when EURISKO halted at 4:00 one morning waiting for an answer to a question. The system was supposed to know that questions were OK when a person was around, but not at night with no people at hand. People are represented in its knowledge base in the same manner as any other object. EURISKO wanted (i.e., had as a goal) to ask a question. It realized that the reason it could not was that no object in its current environment had the "person" attribute. It therefore declared itself to be a "person", and proceeded to ask the question. Doug says that it was rather difficult to explain to the system why these were not reasonable things to do. -- Ken Laws ------------------------------ Date: Wed 16 Nov 83 10:09:24-PST From: Ken Laws Subject: SPIE Call for Papers SPIE has put out a call for papers for its Technical Symposium East '84 in Arlington, April 29 - May 4. One of the 10 subtopics is Applications of AI, particularly image understanding, expert systems, autonomous navigation, intelligent systems, computer vision, knowledge-based systems, contextual scene analysis, and robotics. Abstracts are due Nov. 21, manuscripts by April 2. For more info, contact SPIE Technical Program Committee P.O. Box 10 Bellingham, Washington 98227-0010 (206) 676-3290, Technical Program Dept. Telex 46-7053 -- Ken Laws ------------------------------ Date: 15 Nov 83 14:19:54 EST From: Smadar Subject: An III talk this Thursday... [Reprinted from the RUTGERS bboard.] Title: CRITTER - A System for 'Critiquing' Circuits Speaker: Van Kelly Date: Thursday, November 17,1983, 1:30-2:30 PM Location: Hill Center, Seventh floor lounge Van kelly, a Ph.D. student in our department, will describe a computer system, CRITTER, for 'critiquing' digital circuit designs. This informal talk based on his current thesis research. Here is an abstract of the talk: CRITTER is an exploratory prototype design aid for comprehensive "critiquing" of digital circuit designs. While originally intended for verifying a circuit's functional correctness and timing safety, it can also be used to estimate design robustness, sensitivity to device parameters, and (to some extent) testability. CRITTER has been built using Artificial Intelligence ("Expert Systems") technology and its reasoning is guided by an extensible collection of electronic knowledge derived from human experts. Also, a new non-procedural representation for both the real-time behavior of circuits and circuit specifications has led to a streamlined circuit modeling formalism based on ordinary mathematical function composition. A version of CRITTER has been tested on circuits with complexities of up to a dozen TTL SSI/MSI packages. A more powerful version is being adapted for use in an automated VLSI design environment. ------------------------------ Date: 16 Nov 83 12:58:07 PST (Wednesday) From: John Larson Subject: AI and the military (long message) Received over the network . . . STRATEGIC COMPUTING PLAN ANNOUNCED; REVOLUTIONARY ADVANCES IN MACHINE INTELLIGENCE TECHNOLOGY TO MEET CRITICAL DEFENSE NEEDS Washington, D.C. (7 Nov. 1983) - - Revolutionary advances in the way computers will be applied to tomorrow's national defense needs were described in a comprehensive "Strategic Computing" plan announced today by the Defense Advanced Research Projects Agency (DARPA). DARPA's plan encompasses the development and application of machine intelligence technology to critical defense problems. The program calls for transcending today's computer capabilities by a "quantum jump." The powerful computers to be developed under the plan will be driven by "expert systems" that mimic the thinking and reasoning processes of humans. The machines will be equipped with sensory and communication modules enabling them to hear, talk, see and act on information and data they develop or receive. This new technology as it emerges during the coming decade will have unprecedented capabilities and promises to greatly increase our national security. Computers are already widely employed in defense, and are relied on to help hold the field against larger forces. But current computers have inflexible program logic, and are limited in their ability to adapt to unanticipated enemy actions in the field. This problem is heightened by the increasing pace and complexity of modern warfare. The new DARPA program will confront this challenge by producing adaptive, intelligent computers specifically aimed at critical military applications. Three initial applications are identified in the DARPA plan. These include autonomous vehicles (unmanned aircraft, submersibles, and land vehicles), expert associates, and large-scale battle management systems. In contrast with current guided missiles and munitions, the new autonomous vehicles will be capable of complex, far-ranging reconnaissance and attack missions, and will exhibit highly adaptive forms of terminal homing. A land vehicle described in the plan will be able to navigate cross-country from one location to another, planning its route from digital terrain data, and updating its plan as its vision and image understanding systems sense and resolve ambiguities between observed and stored terrain data. Its expert local-navigation system will devise schemes to insure concealment and avoid obstacles as the vehicle pursues its mission objectives. A pilot's expert associate will be developed that can interact via speech communications and function as a "mechanized co-pilot". This system will enable a pilot to off-load lower-level instrument monitoring, control, and diagnostic functions, freeing him to focus on high-priority decisions and actions. The associate will be trainable and personalizable to the requirements of specific missions and the methods of an individual pilot. It will heighten pilots' capabilities to act effectively and decisively in high stress combat situations. The machine intelligence technology will also be applied in a carrier battle-group battle management system. This system will aid in the information fusion, option generation, decision making, and event monitoring by the teams of people responsible for managing such large-scale, fast-moving combat situations. The DARPA program will achieve its technical objectives and produce machine intelligence technology by jointly exploiting a wide range of recent scientific advances in artificial intelligence, computer architecture, and microelectronics. Recent advances in artificial intelligence enable the codification in sets of computer "rules" of the thinking processes that people use to reason, plan, and make decisions. For example, a detailed codification of the thought processes and heuristics by which a person finds his way through an unfamiliar city using a map and visual landmarks might be employed as the basis of an experimental expert system for local navigation (for the autonomous land vehicle). Such expert systems are already being successfully employed in medical diagnosis, experiment planning in genetics, mineral exploration, and other areas of complex human expertise. Expert systems can often be decomposed into separate segments that can be processed concurrently. For example, one might search for a result along many paths in parallel, taking the first satisfactory solution and then proceeding on to other tasks. In many expert systems rules simply "lay in wait" - firing only if a specific situation arises. Different parts of such a system could be operated concurrently to watch for the individual contexts in which their rules are to fire. DARPA plans to develop special computers that will exploit opportunities for concurrent processing of expert systems. This approach promises a large increase in the power and intelligence of such systems. Using "coarse-mesh" machines consisting of multiple microprocessors, an increase in power of a factor of one hundred over current systems will be achievable within a few years. By creating special VLSI chip designs containing multiple "fine-mesh" processors, by populating entire silicon wafers with hundreds of such chips, and by using high-bandwidth optoelectronic cables to interconnect groups of wafers, increases of three or four orders of magnitude in symbol processing and rule-firing rates will be achieved as the research program matures. While the program will rely heavily on silicon microelectronics for high-density processing structures, extensive use will also be made of gallium arsenide technology for high-rate signal processing, optoelectronics, and for military applications requiring low-power dissipation and high-immunity to radiation. The expert system technology will enable the DARPA computers to "think smarter." The special architectures for concurrency and the faster, denser VLSI microelectronics will enable them to "think harder and faster." The combination of these approaches promises to be potent indeed. But machines that mimic thinking are not enough by themselves. They must be provided with sensory devices that mimic the functions of eyes and ears. They must have the ability to see their environment, to hear and understand human language, and to respond in kind. Huge computer processing rates will be required to provide effective machine vision and machine understanding of natural language. Recent advances in the architecture of special processor arrays promise to provide the required rates. By patterning many small special processors together on a silicon chip, computer scientists can now produce simple forms of machine vision in a manner analogous to that used in the retina of the eye. Instead of each image pixel being sequentially processed as when using a standard von Neumann computer, the new processor arrays allow thousands of pixels to be processed simultaneously. Each image pixel is processed by just a few transistor switches located close together in a processor cell that communicates over short distances with neighboring cells. The number of transistors required to process each pixel can be perhaps one one-thousandth of that employed in a von Neumann machine, and the short communications distances lead to much faster processing rates per pixel. All these effects multiply the factor of thousands gained by concurrency. The DARPA program plans to provide special vision subsystems that have rates as high as one trillion von Neumann equivalent operations per second as the program matures in the late 1980's. The DARPA Strategic Computing plan calls for the rapid evolution of a set of prototype intelligent computers, and their experimental application in military test-bed environments. The planned activities will lead to a series of demonstrations of increasingly sophisticated machine intelligence technology in the selected applications as the program progresses. DARPA will utilize an extensive infrastructure of computers, computer networks, rapid system prototyping services, and silicon foundries to support these technology explorations. This same infrastructure will also enable the sharing and propagation of successful results among program participants. As experimental intelligent machines are created in the program, some will be added to the computer network resources - further enhancing the capabilities of the research infrastructure. The Strategic Computing program will be coordinated closely with Under Secretary of Defense Research and Engineering, the Military Services, and other Defense Agencies. A number of advisory panels and working groups will also be constituted to assure inter-agency coordination and maintain a dialogue within the scientific community. The program calls for a cooperative effort among American industry, universities, other research institutions, and government. Communication is critical in the management of the program since many of the contibutors will be widely dispersed throughout the U.S. Heavy use will be made of the Defense Department's ARPANET computer network to link participants and to establish a productive research environment. Ms. Lynn Conway, Assistant Director for Strategic Computing in DARPA's Information Processing Techniques Office, will manage the new program. Initial program funding is set at $50M in fiscal 1984. It is proposed at $95M in FY85, and estimated at $600M over the first five years of the program. The successful achievement of the objectives of the Strategic Computing program will lead to the deployment of a new generation of military systems containing machine intelligence technology. These systems promise to provide the United States with important new methods of defense against both massed forces and unconventional threats in the future - methods that can raise the threshold and decrease the likelihood of major conflict. ------------------------------ End of AIList Digest ******************** 20-Nov-83 15:06:03-PST,12574;000000000001 Mail-From: LAWS created at 20-Nov-83 15:05:20 Date: Sunday, November 20, 1983 2:53PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #100 To: AIList@SRI-AI AIList Digest Sunday, 20 Nov 1983 Volume 1 : Issue 100 Today's Topics: Intelligence - Definition & Msc., Looping Problem - The Zahir, Scientific Method - Psychology ---------------------------------------------------------------------- Date: Wed, 16 Nov 1983 10:48:34 EST From: AXLER.Upenn-1100@Rand-Relay (David M. Axler - MSCF Applications Mgr.) Subject: Intelligence and Categorization I think Tom Portegys' comment in 1:98 is very true. Knowing whether or not a thing is intelligent, has a soul, etc., is quite helpful in letting us categorize it. And, without that categorization, we're unable to know how to understand it. Two minor asides that might be relevant in this regard: 1) There's a school of thought in the fields of linguistics, folklore, anthropology, and folklore, which is based on the notion (admittedly arguable) that the only way to truly understand a culture is to first record and understand its native categories, as these structure both its language and its thought, at many levels. (This ties in to the Sapir-Whorf hypothesis that language structures culture, not the reverse...) From what I've read in this area, there is definite validity in this approach. So, if it's reasonable to try and understand a culture in terms of its categories (which may or may not be translatable into our own culture's categories, of course), then it's equally reasonable for us to need to categorize new things so that we can understand them within our existing framework. 2) Back in medieval times, there was a concept known as the "Great Chain of Being", which essentially stated that everything had its place in the scheme of things; at the bottom of the chain were inanimate things, at the top was God, and the various flora and fauna were in-between. This set of categories structured a lot of medieval thinking, and had major influences on Western thought in general, including thought about the nature of intelligence. Though the viewpoint implicit in this theory isn't widely held any more, it's still around in other, more modern, theories, but at a "subconscious" level. As a result, the notion of 'machine intelligence' can be a troubling one, because it implies that the inanimate is being relocated in the chain to a position nearly equal to that of man. I'm ranging a bit far afield here, but this ought to provoke some discussion... Dave Axler ------------------------------ Date: 15 Nov 83 15:11:32-PST (Tue) From: pur-ee!CS-Mordred!Pucc-H.Pucc-I.Pucc-K.ags @ Ucb-Vax Subject: Re: Parallelism & Consciousness - (nf) Article-I.D.: pucc-k.115 Faster = More Intelligent. Now there's an interesting premise... According to relativity theory, clocks (and bodily processes, and everything else) run faster at the top of a mountain or on a plane than they do at sea level. This has been experimentally confirmed. Thus it seems that one can become more intelligent merely by climbing a mountain. Of course the effect is temporary... Maybe this is why we always see cartoons about people climbing mountains to inquire about "the meaning of life" (?) Dave Seaman ..!pur-ee!pucc-k!ags ------------------------------ Date: 17 Nov 83 16:38 EST From: Jim Lynch Subject: Continuing Debate (discussion) on intelligence. I have enjoyed the continuing discussion concerning the definition of intelligence and would only add a few thoughts. 1. I tend to agree with Minsky that intelligence is a social concept, but I believe that it is probably even more of an emotional one. Intelligence seems to fall in the same category with notions such as beauty, goodness, pleasant, etc. These concepts are personal, intensely so, and difficult to describe, especially in any sort of quantitative terms. 2. A good part of the difficulty with defining Artificial Intelligence is due, no doubt, to a lack of a good definition for intelligence. We probablyy cannot define AI until the psychologists define "I". 3. Continuing with 2, the definition probably should not worry us too much. After all, do psychologists worry about "Natural Computation"? Let us let the psychologists worry about what intelligence is, let us worry about how to make it artificial!! (As has been pointed out many times, this is certainly an iterative process and we can surely learn much from each other!). 4. The notion of intelligence seems to be a continuum; it is doubtful that we can define a crisp and fine line dividing the intelligent from the non-intelligent. The current debate has provided enough examples to make this clear. Our job, therefore, is not to make computers intelligent, but to make them more intelligent. Thanks for the opportunity to comment, Jim Lynch, Dahlgren, Virginia ------------------------------ Date: Thu 17 Nov 83 16:07:41-PST From: Ken Laws Subject: Intelligence I had some difficultly refuting a friend's argument that intelligence is "problem solving ability", and that deciding what problems to solve is just one facet or level of intelligence. I realize that this is a vague definition, but does anyone have a refutation? I think we can take for granted that summing the same numbers over and over is not more intelligent than summing them once. Discovering a new method of summing them (e.g., finding a pattern and a formula for taking advantage of it) is intelligent, however. To some extent, then, the novelty of the problem and the methods used in its solution must be taken into account. Suppose that we define intelligence in terms of the problem-solving techniques available in an entity's repertoire. A machine's intelligence could be described much as a pocket calculator's capabilities are: this one has modus ponens, that one can manipulate limits of series. The partial ordering of such capabilities must necessarily be goal- dependent and so should be left to the purchaser. I agree with the AIList reader who defined an intelligent entity as one that builds and refines knowledge structures representing its world. Ability to manipulate and interconvert particular knowledge structures fits well into the capability rating system above. Learning, or ability to remember new techniques so that they need not be rederived, is downplayed in this view of intelligence, although I am sure that it is more than just an efficiency hack. Problem solving speed seems to be orthogonal to the capability dimension, as does motivation to solve problems. -- Ken Laws ------------------------------ Date: 16 Nov 83 4:21:55-PST (Wed) From: harpo!seismo!philabs!linus!utzoo!utcsstat!laura @ Ucb-Vax Subject: KILLING THINGS Article-I.D.: utcsstat.1439 I think that one has to make a distinction between dolphins killing fish to eat, and hypothetical turtles killing rabbits, not to eat, but because they compete for the same land resources. To my mind they are different sorts of killings (though from the point of veiw of the hapless rabbit or fish they may be the same). Dolphins kill sharks that attack the school, though -- I do not think that this 'self-defense' killing is the same as the planned extermination of another species. if you believe that planned extermination is the definition of intelligence then I'll bet you are worried about SETI. On the other hand, I suppose you must not believe that pacifist vegetarian monks qualify as intelligent. Or is intelligence something posessed by a species rather than an individual? Or perhaps you see that eating plants is indeed killing them. Now, we have, defined all animals and plants like the venus fly-trap as intelligent while most plants are not. All the protists that I can think of right now would also be intelligent, though a euglena would be an interesting case. I think that "killing things" is either too general or too specific (depending on your definition of killing and which things you admit to your list of "things") to be a useful guide for intelligence. What about having fun? Perhaps the ability to laugh is the dividing point between man (as a higher intelligence) and animals, who seem to have some appreciation for pleasure (if not fun) as distinct from plants and protists whose joy I have never seen measured. Dolphins seem to have a sense of fun as well, which is (to my mind) a very good thing. What this bodes for Mr. Spock, though, is not nice. And despite megabytes of net.jokes, this 11/70 isn't chuckling. :-) Laura Creighton utzoo!utcsstat!laura ------------------------------ Date: Sun 20 Nov 83 02:24:00-CST From: Aaron Temin Subject: Re: Artificial Humanity I found these errors really interesting. I would think a better rule for Eurisko to have used in the bounds checking case would be to keep the bounds-checking code, but use it less frequently, only when it was about to announce something as interesting, for instance. Then it may have caught the flip-flop error itself, while still gaining speed other times. The "credit assignment bug" makes me think Eurisko is emulating some professors I have heard of.... The person bug doesn't even have to be bug. The rule assumes that if a person is around, then he or she will answer a question typed to a console, perhaps? Rather it should state that if a person is around, Eurisko should ask THAT person the question. Thus if Eurisko is a person, it should have asked itself (not real useful, maybe, but less of a bug, I think). While computer enthusiasts like to speak of all programs in anthropomorphic terms, Eurisko seems like one that might really deserve that. Anyone know of any others? -aaron ------------------------------ Date: 13 Nov 83 10:58:40-PST (Sun) From: ihnp4!houxm!hogpc!houti!ariel!vax135!cornell!uw-beaver!tektronix !ucbcad!notes @ Ucb-Vax Subject: Re: the halting problem in history - (nf) Article-I.D.: ucbcad.775 Halting problem, lethal infinite loops in consciousness, and the Zahir: Borges' "Zahir" story was interesting, but the above comment shows just how successful Borges is in his stylistic approach: by overwhelming the reader with historical references, he lends legitimacy to an idea that might only be his own. Try tracking down some of his references some- time--it's not easy! Many of them are simply made up. Michael Turner (ucbvax!ucbesvax.turner) ------------------------------ Date: 17 Nov 83 13:50:54-PST (Thu) From: ihnp4!houxm!mhuxl!ulysses!unc!mcnc!ncsu!fostel @ Ucb-Vax Subject: I recall Rational Psychology Article-I.D.: ncsu.2407 First, let's not revive the Rational Psychology debate. It died of natural causes, and we should not disturb its immortal soul. However, F Montalvo has said something very unpleasant about me, and I'm not quite mature enough to ignore it. I was not making an idle attack, nor do I do so with superficial knowledge. Further, I have made quite similar statements in the presence of the enemy -- card carrying psychologists. Those psychologists whose egos are secure often agree with the assesment. Proper scientific method is very hard to apply in the face of stunning lack of understanding or hard, testable theories. Most proper experiments are morally unacceptable in the pschological arena. As it is, there are so many controls not done, so many sources of artifact, so much use of statistics to try to ferret out hoped-for correlations, so much unavoidable anthropomorphism. As with scholars such as H. Dumpty, you can define "science" to mean what you like, but I think most psychological work fails the test. One more thing, It's pretty immature to assume that someone who disagrees with you has only superficial knowledge of the subject. (See, I told you I was not very mature ....) ----GaryFostel---- ------------------------------ End of AIList Digest ******************** 20-Nov-83 15:44:02-PST,15439;000000000001 Mail-From: LAWS created at 20-Nov-83 15:39:35 Date: Sunday, November 20, 1983 3:15PM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #101 To: AIList@SRI-AI AIList Digest Monday, 21 Nov 1983 Volume 1 : Issue 101 Today's Topics: Pattern Recognition - Forced Matching, Workstations - VAX, Alert - Computer Vision, Correction - AI Labs in IEEE Spectrum, AI - Challenge, Conferences - Announcements and Calls for Papers ---------------------------------------------------------------------- Date: Wed, 16 Nov 83 10:53 EST From: Tim Finin Subject: pattern matchers From: Stanley T. Shebs Subject: Pattern Matchers ... My next puzzle is about pattern matchers. Has anyone looked carefully at the notion of a "non-failing" pattern matcher? By that I mean one that never or almost never rejects things as non-matching. ... There is a long history of matchers which can be asked to "force" a match. In this mode, the matcher is given two objects and returns a description of what things would have to be true for the two objects to match. Two such matchers come immediately to my mind - see "How can MERLIN Understand?" by Moore and Newell in Gregg (ed), Knowledge and Cognition, 1973, and also "An Overview of KRL, A Knowledge Representation Language" by Bobrow and Winograd (which appeared in the AI Journal, I believe, in 76 or 77). ------------------------------ Date: Fri 18 Nov 83 09:31:38-CST From: CS.DENNEY@UTEXAS-20.ARPA Subject: VAX Workstations I am looking for information on the merits (or lack of) of the VAX Workstation 100 for AI development. ------------------------------ Date: Wed, 16 Nov 83 22:22:03 pst From: weeks%ucbpopuli.CC@Berkeley (Harry Weeks) Subject: Computer Vision. There have been some recent articles in this list on computer vision, some of them queries for information. Although I am not in this field, I read with interest a review article in Nature last week. Since Nature may be off the beaten track for many people in AI (in fact articles impinging on computer science are rare, and this one probably got in because it also falls under neuroscience), I'm bringing the article to the attention of this list. The review is entitled ``Parallel visual computation'' and appears in Vol 306, No 5938 (3-9 November), page 21. The authors are Dana H Ballard, Geoffrey E Hinton and Terrence J Sejnowski. There are 72 references into the literature. Harry Weeks g.weeks@Berkeley ------------------------------ Date: 17 Nov 83 20:25:30-PST (Thu) From: pur-ee!uiucdcs!marcel @ Ucb-Vax Subject: Re: IEEE Spectrum Alert - (nf) Article-I.D.: uiucdcs.3909 For safety's sake, let me add a qualification about the table on sources of funding: it's incorrect. The University of Illinois is represented as having absolutely NO research in 5th-generation AI, not even under OTHER funding. This is false, and will hopefully be rectified in the next issue of the Spectrum. I believe a delegation of our Professors is flying to the coast to have a chat with the Spectrum staff ... If we can be so misrepresented, I wonder how the survey obtained its information. None of our major AI researchers remember any attempts to survey their work. Marcel Schoppers U of Illinois @ Urbana-Champaign ------------------------------ Date: 17 Nov 83 20:25:38-PST (Thu) From: pur-ee!uiucdcs!marcel @ Ucb-Vax Subject: Re: just a reminder... - (nf) Article-I.D.: uiucdcs.3910 I agree [with a previous article]. I myself am becoming increasingly worried about a blithe attitude I sometimes hear: if our technology eliminates some jobs, it will create others. True, but not everyone will be capable of keeping up with the change. Analogously, the Industrial Revolution is now seen as a Good Thing, and its impacts were as profound as those promised by AI. And though it is said that the growth of knowledge can only be advantageous in the long run (Logical Positivist view?), many people became victims of the Revolution. In this respect I very much appreciated an idea that was aired at IJCAI-83, namely that we should be building expert systems in economics to help us plan and control the effects of our research. As for the localization of power, that seems almost inevitable. Does not the US spend enough on cosmetics to cover the combined Gross National Products of 37 African countries? And are we not so concerned about our Almighty Pocket that we simply CANNOT export our excess groceries to a needy country, though the produce rot on our dock? Then we can also keep our technology to ourselves. One very obvious, and in my opinion sorely needed, application of AI is to automating legal, veterinary and medical expertise. Of course the law system and our own doctors will give us hell for this, but on the other hand what kind of service profession is it that will not serve except at high cost? Those most in need cannot afford the price. See for yourself what kind of person makes it through Medical School: those who are most aggressive about beating their fellow students, or those who have the money to buy their way in. It is little wonder that so few of them will help the under-priviledged -- from the start the selection criteria wage against such motivation. Let's send our machines in where our "doctors" will not go! Marcel Schoppers U of Illinois @ Urbana-Champaign ------------------------------ Date: 19 Nov 83 09:22:42 EST (Sat) From: rej@Cornell (Ralph Johnson) Subject: The AI Challenge The recent discussions on AIlist have been boring, so I have another idea for discussion. I see no evidence that that AI is going to make as much of a change on the world as data processing or information retrieval. While research in AI has produced many results in side areas such as computer languages, computer architecture, and programming environments, none of the past promises of AI (automatic language translation, for example) have been fulfilled. Why should I expect anything more in the future? I am a soon-to-graduate PhD candidate at Cornell. Since Cornell puts little emphasis on AI, I decided to learn a little on my own. Most AI literature is hard to read, as very little concrete is said. The best book that I read (best for someone like me, that is) was the three-volume "Handbook on Artificial Intelligence". One interesting observation was that I already knew a large percentage of the algorithms. I did not even think of most of them as being AI algorithms. The searching algorithms (with the exception of alpha beta pruning) are used in many areas, and algorithms that do logical deduction are part of computational mathematics (just my opinion, as I know some consider this hard core AI). Algorithms in areas like computer vision were completely new, but I could see no relationship between those algorithms and algorithms in programs called "expert systems", another hot AI topic. [Agreed, but the gap is narrowing. There have been 1 or 2 dozen good AI/vision dissertations, but the chief link has been that many individuals and research departments interested in one area have also been interested in the other. -- KIL] As for expert systems, I could see no relationship between one expert system and the next. An expert system seems to be a program that uses a lot of problem-related hacks to usually come up with the right answer. Some of the "knowledge representation" schemes (translated "data structures") are nice, but everyone seems to use different ones. I have read several tech reports describing recent expert systems, so I am not totally ignorant. What is all the noise about? Why is so much money being waved around? There seems to be nothing more to expert systems than to other complicated programs. [My own somewhat heretical view is that the "expert system" title legitimizes something that every complicated program has been found to need: hackery. A rule-based system is sufficiently modular that it can be hacked hundreds of times before it is so cumbersome that the basic structures must be rewritten. It is software designed to grow, as opposed to the crystalline gems of the "optimal X" paradigm. The best expert systems, of course, also contain explanatory capabilities, hierarchical inference, constrained natural language interfaces, knowledge base consistency checkers, and other useful features. -- KIL] I know that numerical analysis and compiler writing are well developed fields because there is a standard way of thinking that is associated with each area and because a non-expert can use tools provided by experts to perform computation or write a parser without knowing how the tools work. In fact, a good test of an area within computer science is whether there are tools that a non-expert can use to do things that, ten years ago, only experts could do. Is there anything like this in AI? Are there natural language processors that will do what YACC does for parsing computer languages? There seem to be a number of answers to me: 1) Because of my indoctrination at Cornell, I categorize much of the important results of AI in other areas, thus discounting the achievements of AI. 2) I am even more ignorant than I thought, and you will enlighten me. 3) Although what I have said describes other areas of AI pretty much, yours is an exception. 4) Although what I have said describes past results of AI, major achievements are just around the corner. 5) I am correct. You may be saying to yourself, "Is this guy serious?" Well, sort of. In any case, this should generate more interesting and useful information than trying to define intelligence, so please treat me seriously. Ralph Johnson ------------------------------ Date: Thu 17 Nov 83 16:57:55-PST From: C.S./Math Library Subject: Conference Announcements and Call for Papers [Reprinted from the SU-SCORE bboard.] Image Technology 1984 37th annual conference May 20-24, 1984 Boston, Mass. Jim Clark, papers chairman British Robot Association 7th annual conference 14-17, May 1984 Cambridge, England Conference director-B.R.A. 7, British Robot Association, 28-30 High Street, Kempston, Bedford MK427AJ, England First International Conference on Computers and Applications Beijing, China, June 20-22, 1984 co-sponsored by CIE computer society and IEEE computer society CMG XIV conference on computer evaluation--preliminary agenda December 6-9, 1983 Crystal City, Va. International Symposium on Symbolic and Algebraic Computation EUROSAM 84 Cambridge, England July 9-11, 1984 call for papers M. Mignotte, Centre de Calcul, Universite Louis Pasteur, 7 rue rene Descartes, F67084 Strasvourg, France ACM Computer Science Conference The Future of Computing February 14-16, 1984 Philadelphia, Penn. Aaron Beller, Program Chair, Computer and Information Science Department, Temple University Philadelphia, Penn. 19122 HL ------------------------------ Date: Fri 18 Nov 83 04:00:10-CST From: Werner Uhrig Subject: ***** Call for Papers: LISP and Functional Programming ***** please help spread the word by announcing it on your local machines. thanks --------------- ()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()() () CALL FOR PAPERS () () 1984 ACM SYMPOSIUM ON () () LISP AND FUNCTIONAL PROGRAMMING () () UNIVERSITY OF TEXAS AT AUSTIN, AUGUST 5-8, 1984 () () (Sponsored by the ASSOCIATION FOR COMPUTING MACHINERY) () ()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()() This is the third in a series of biennial conferences on the LISP language and issues related to applicative languages. Especially welcome are papers addressing implementation problems and programming environments. Areas of interest include (but are not restricted to) systems, large implementations, programming environments and support tools, architectures, microcode and hardware implementations, significant language extensions, unusual applications of LISP, program transformations, compilers for applicative languages, lazy evaluation, functional programming, logic programming, combinators, FP, APL, PROLOG, and other languages of a related nature. Please send eleven (11) copies of a detailed summary (not a complete paper) to the program chairman: Guy L. Steele Jr. Tartan Laboratories Incorporated 477 Melwood Avenue Pittsburgh, Pennsylvania 15213 Submissions will be considered by each member of the program committee: Robert Cartwright, Rice William L. Scherlis, Carnegie-Mellon Jerome Chailloux, INRIA Dana Scott, Carnegie-Mellon Daniel P. Friedman, Indiana Guy L. Steele Jr., Tartan Laboratories Richard P. Gabriel, Stanford David Warren, Silogic Incorporated Martin L. Griss, Hewlett-Packard John Williams, IBM Peter Henderson, Stirling Summaries should explain what is new and interesting about the work and what has actually been accomplished. It is important to include specific findings or results and specific comparisons with relevant previous work. The committee will consider the appropriateness, clarity, originality, practicality, significance, and overall quality of each summary. Time does not permit consideration of complete papers or long summaries; a length of eight to twelve double-spaced typed pages is strongly suggested. February 6, 1984 is the deadline for the submission of summaries. Authors will be notified of acceptance or rejection by March 12, 1984. The accepted papers must be typed on special forms and received by the program chairman at the address above by May 14, 1984. Authors of accepted papers will be asked to sign ACM copyright forms. Proceedings will be distributed at the symposium and will later be available from ACM. Local Arrangements Chairman General Chairman Edward A. Schneider Robert S. Boyer Burroughs Corporation University of Texas at Austin Austin Research Center Institute for Computing Science 12201 Technology Blvd. 2100 Main Building Austin, Texas 78727 Austin, Texas 78712 (512) 258-2495 (512) 471-1901 CL.SCHNEIDER@UTEXAS-20.ARPA CL.BOYER@UTEXAS-20.ARPA ------------------------------ End of AIList Digest ******************** 22-Nov-83 11:16:29-PST,18647;000000000001 Mail-From: LAWS created at 22-Nov-83 10:36:14 Date: Tuesday, November 22, 1983 10:31AM From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #102 To: AIList@SRI-AI AIList Digest Tuesday, 22 Nov 1983 Volume 1 : Issue 102 Today's Topics: AI and Society - Expert Systems, Scientific Method - Psychology, Architectures - Need for Novelty, AI - Response to Challenge ---------------------------------------------------------------------- Date: 20 Nov 83 14:50:23-PST (Sun) From: harpo!floyd!clyde!akgua!psuvax!simon @ Ucb-Vax Subject: Re: just a reminder... - (nf) Article-I.D.: psuvax.357 It seems a little dangerous "to send machines where doctors won't go" - you'll get the machines treating the poor, and human experts for the privileged few. Also, expert systems for economics and social science, to help us would be fine, if there was a convincing argument that a)these social sciences are truly helpful for coping with unpredictable technological change, and b) that there is a sufficiently accepted basis of quantifiable knowledge to put in the proposed systems. janos simon ------------------------------ Date: Mon, 21 Nov 1983 15:24 EST From: MONTALVO%MIT-OZ@MIT-MC.ARPA Subject: I recall Rational Psychology Date: 17 Nov 83 13:50:54-PST (Thu) From: ihnp4!houxm!mhuxl!ulysses!unc!mcnc!ncsu!fostel @ Ucb-Vax Subject: I recall Rational Psychology ... Proper scientific method is very hard to apply in the face of stunning lack of understanding or hard, testable theories. Most proper experiments are morally unacceptable in the pschological arena. As it is, there are so many controls not done, so many sources of artifact, so much use of statistics to try to ferret out hoped-for correlations, so much unavoidable anthropomorphism. As with scholars such as H. Dumpty, you can define "science" to mean what you like, but I think most psychological work fails the test. ----GaryFostel---- You don't seem to be aware of Experimental Psychology, which involves subjects' consent, proper controls, hypothesis formation and evaluation, and statistical validation. Most of it involves sensory processes and learning. The studies are very rigorous and must be so to end up in the literature. You may be thinking of Clinical Psychology. If so, please don't lump all of Psychology into the same group. Fanya Montalvo ------------------------------ Date: 19 Nov 83 11:15:50-PST (Sat) From: decvax!tektronix!ucbcad!notes @ Ucb-Vax Subject: Re: parallelism vs. novel architecture - (nf) Article-I.D.: ucbcad.835 Re: parallelism and fundamental discoveries The stored-program concept (Von Neumann machine) was indeed a breakthrough both in the sense of Turing (what is theoretically computable) and in the sense of Von Neuman (what is a practical machine). It is noteworthy, however, that I am typing this message using a text editor with a segment of memory devoted to program, another segment devoted to data, and with an understanding on the part of the operating system that if the editor were to try to alter one of its own instructions, the operating system should treat this as pathological, and abort it. In other words, the vaunted power of being able to write data that can be executed as a program is treated in the most stilted and circumspect manner in the interests of practicality. It has been found to be impractical to write programs that modify their own inner workings. Yet people do this to their own consciousness all the time--in a largely unconscious way. Turing-computability is perhaps a necessary condition for intelligence. (That's been beaten to death here.) What is needed is a sufficient condition. Can that possibly be a single breakthrough or innovation? There is no question that, working from the agenda for AI that was so hubristically layed out in the 50's and 60's, such a breakthrough is long overdue. Who sees any intimation of it now? Perhaps what is needed is a different kind of AI researcher. New ground is hard to break, and harder still when the usual academic tendency is to till old soil until it is exhausted. I find it interesting that many of the new ideas in AI are coming from outside the U.S. AI establishment (MIT, CMU, Stanford, mainly). Logic programming seems largely to be a product of the English-speaking world *apart* from the U.S. Douglas Hofstadter's ideas (though probably too optimistic) are at least a sign that, after all these years, some people find the problem too important to be left to the experts. Tally Ho! Maybe AI needs a nut with the undaunted style of a Nicola Tesla. Some important AI people say that Hofstadter's schemes can't work. This makes me think of the story about the young 19th century physicist, whose paper was reviewed and rejected as meaningless by 50 prominent physicists of the time. The 51st was Maxwell, who had it published immediately. Michael Turner (ucbvax!ucbesvax.turner) ------------------------------ Date: 20 November 1983 2359-PST (Sunday) From: helly at AEROSPACE (John Helly) Subject: Challenge I am responding to Ralph Johnson's recent submittal concerning the content and contribution of work in the field of AI. The following comments should be evaluated in light of the fact that I am currently developing an 'expert system' as a dissertation topic at UCLA. My immediate reaction to Johnson's queries/criticisms of AI is that of hearty agreement. Having read a great deal of AI literature, my personal bias is that there is a great deal of rediscovery of Knuth in the context of new applications. The only things apparently unique are that each new 'discovery' carries with it a novel jargon with very little attempt to connect and build on previous work in the field. This reflects a broader concern I have with Computer Science in general in that, having been previously trained as a biologist, I find very little that I consider scientific in this field. This does not diminish my hope for, and consequently my commitment to, work in this area. Like many things, this commitment is based on my intuition (read faith) that there really is something of value in this field. The only rationale I can offer for such a commitment is the presumption that the lack of progress in AI research is the result of the lack of scientific discipline of AI researchers and computer scientists in general. The AI community looks much more like a heterogeneous population of hackers than that of a disciplined, scientific community. Maybe this is symptomatic of a new field of science going through growing pains but I do not personally believe this is the case. I am unaware of any similar developmental process in the history of science. This all sounds pretty negative, I know. I believe that criticism should always be stated with some possible corrective action, though, and maybe I have some. Computer science curricula should require formal scientific training. Exposure to truly empirical sciences would serve to familiarize students with the value of systematic research, experimental design, hypothesis testing and the like. We should find ways to apply the scientific method to our research rather than collecting a lot of anecdotal information about our 'programming environment' and 'heuristics' and publishing it at first light. Maybe the computer science is basically an engineering discipline (i.e., application-oriented) rather than a science. I believe, however, that in the least computer science, even if misnamed, offers powerful tools for investigating human information processing (i.e, intelligence) if approached scientifically. Properly applied these tools can provide the same benefits they have offered physicists, biologists and medical researchers - insight into mechanisms and techniques for simulating the systems of interest. Much of AI is very slick programming. I'm just not certain that it is anything more than that, at least at present. ------------------------------ Date: Mon 21 Nov 83 14:12:35-PST From: Tom Dietterich Subject: Reply to Ralph Johnson Your recent msg to AILIST was certainly provocative, and I thought I'd try to reply to a couple of the points you made. First, I'm a little appalled at what you portray as the "Cornell" attitude towards AI. I hope things will improve there in the future. Maybe I can contribute a little by trying to persuade you that AI has substance. I'd like to begin by calling attention to the criteria that you are using to evaluate AI. I believe that if you applied these same criteria to other areas of computer science, you would find them lacking also. For example, you say that "While research in AI has produced many results in side areas..., none of the past promises of AI have been fulfilled." If we look at other fields of computer science, we find similar difficulties. Computer science has promised secure, reliable, user-friendly computing facilities, cheap and robust distributed systems, integrated software tools. But what do we have? Well, we have some terrific prototypes in research labs, but the rest of the world is still struggling with miserable computing environments, systems that constantly crash, and distributed systems that end up being extremely expensive and unreliable. The problem with this perspective is that it is not fair to judge a research discipline by the success of its applications. In AI research labs, AI has delivered on many of its early promises. We now have machines with limited visual and manipulative capabilities. And we do have systems that perform automatic language translation (e.g., at Texas). Another difficulty of judging AI is that it is a "residual" discipline. As Avron Barr wrote in the introduction to the AI Handbook, "The realization that the detailed steps of almost all intelligent human activity were unknown marked the beginning of Artificial Intelligence as a separate part of computer science." AI tackles the hardest application problems around: those problems whose solution is not understood. The rest of computer science is primarily concerned with finding optimum points along various solution dimensions such as speed, memory requirements, user interface facilities, etc. We already knew HOW to sort numbers before we had computers. The role of Computer Science was to determine how to sort them quickly and efficiently using a computer. But, we didn't know HOW to understand language (at least not at a detailed level). AI's task has been to find solutions to these kinds of problems. Since AI has tackled the most difficult problems, it is not surprising that it has had only moderate success so far. The bright side of this, however, is that long after we have figured out whether P=NP, AI will still be uncovering fascinating and difficult problems. That's why I study it. You are correct in saying that the AI literature is hard to read. I think there are several reasons for this. First, there is a very large amount of terminology to master in AI. Second, there are great differences in methodology. There is no general agreement within the AI community about what the hard problems are and how they should be addressed (although I think this is changing). Good luck with any further reading that you attempt. Now let me address some of your specific observations about AI. You say "I already knew a large percentage of the algorithms. I did not even think of most of them as being AI algorithms." I would certainly agree. I cite this as evidence that there is a unity to all parts of computer science, including AI. You also say "An expert system seems to be a program that uses a lot of problem-related hacks to usually come up with the right answer." I think you have hit upon the key lesson that AI learned in the seventies: The solution to many of the problems we attack in AI lies NOT in the algorithms but in the knowledge. That lesson reflects itself, not so much in differences in code, but in differences in methodology. Expert systems are different and important because they are built using a programming style that emphasizes flexibility, transparency, and rapid prototyping over efficiency. You say "There seems to be nothing more to expert systems than to other complicated programs". I disagree completely. Expert systems can be built, debugged, and maintained more cheaply than other complicated programs. And hence, they can be targeted at applications for which previous technology was barely adequate. Expert systems (knowledge programming) techniques continue the revolution in programming that was started with higher-level languages and furthered by structured programming and object-oriented programming. Your view of "knowledge representations" as being identical with data structures reveals a fundamental misunderstanding of the knowledge vs. algorithms point. Most AI programs employ very simple data structures (e.g., record structures, graphs, trees). Why, I'll bet there's not a single AI program that uses leftist-trees or binomial queues! But, it is the WAY that these data structures are employed that counts. For example, in many AI systems, we use record structures that we call "schemas" or "frames" to represent domain concepts. This is uninteresting. But what is interesting is that we have learned that certain distinctions are critical, such as the distinction between a subset of a set and an element of a set. Or the distinction between a causal agent of a disease (e.g., a bacterium) and a feature that is helpful in guiding diagnosis (e.g., whether or not the patient has been hospitalized). Much of AI is engaged in finding and cataloging these distinctions and demonstrating their value in simplifying the construction of expert systems. In your message, you gave five possible answers that you expected to receive. I guess mine doesn't fit any of your categories. I think you have been quite perceptive in your analysis of AI. But you are still looking at AI from the "algorithm" point of view. If you shift to the "knowledge" perspective, your criteria for evaluating AI will shift as well, and I think you will find the field to be much more interesting. --Tom Dietterich ------------------------------ Date: 22 Nov 83 11:45:30 EST (Tue) From: rej@Cornell (Ralph Johnson) Subject: Clarifying my "AI Challange" I am sorry to create the mistaken impression that I don't think AI should be done or is worth the money we spend on it. The side effects alone are worth much more than has been spent. I do understand the effects of AI on other areas of CS. Even though going to the moon brought no direct benefit to the US outside of prestige (which, by the way, was enormous), we learned a lot that was very worthwhile. Planetary scientists point out that we would have learned a lot more if we had spent the money directly on planetary exploration, but the moon race captured the hearts of the public and allowed the money to be spent on space instead of bombs. In a similar way, AI provides a common area for some of our brightest people to tackle very hard problems, and consequently learn a great deal. My question, though, is whether AI is really going to change the world any more than the rest of computer science is already doing. Are the great promises of AI going to be fulfilled? I am thankful for the comments on expert systems. Following these lines of reasoning, expert systems are differentiated from other programs more by the programming methodology used than by algorithms or data structures. It is very helpful to have these distinctions pointed out, and has made several ideas clearer to me. The ideas in AI are not really any more difficult than those in other areas of CS, they are just more poorly explained. Several times I have run in to someone who can explain well the work that he/she has been doing, and each time I understand what they are doing. Consequently, I believe that the reason that I see few descriptions of how systems work is because the designers are not sure how they work, or they do not know what is important in explaining how they work, or they do not know that it is important to explain how they work. Are they, in fact, describing how they work, and I just don't notice? What I would like is more examples of systems that work, descriptions of how they work, and of how well they work. Ralph Johnson (rej@cornell, cornell!rej) ------------------------------ Date: Tue 22 Nov 83 09:25:52-PST From: Tom Dietterich Subject: Re: Clarifying my "AI Challange" Ralph, I can think of a couple of reasons why articles describing Expert Systems are difficult to follow. First, these programs are often immense. It would take a book to describe all of the system and how it works. Hence, AI authors try to pick out a few key things that they think were essential in getting the system to work. It is kind of like reading descriptions of operating systems. Second, the lesson that knowledge is more important than algorithm has still not been totally accepted within AI. Many people tend to describe their systems by describing the architecture (ie., the algorithms and data structures) instead of the knowledge. The result is that the reader is left saying "Yes, of course I understand how backward chaining (or an agenda system) works, but I still don't understand how it diagnoses soybean diseases..." The HEARSAY people are particularly guilty of this. Also, Lenat's dissertation includes much more discussion of architecture than of knowledge. It often takes many years before someone publishes a good analysis of the structure of the knowledge underlying the expert performance of the system. A good example is Bill Clancey's work analyzing the MYCIN system. See his most recent AI Journal paper. --Tom ------------------------------ End of AIList Digest ******************** 25-Nov-83 15:36:56-PST,14704;000000000001 Mail-From: LAWS created at 25-Nov-83 09:36:58 Date: Fri Nov 25, 1983 09:29-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #103 To: AIList@SRI-AI AIList Digest Friday, 25 Nov 1983 Volume 1 : Issue 103 Today's Topics: Alert - Neural Network Simulations & Weizenbaum on The Fifth Generation, AI Jargon - Why AI is Hard to Read, AI and Automation - Economic Effects & Reliability, Conference - Logic Programming Symposium ---------------------------------------------------------------------- Date: Sun, 20 Nov 83 18:05 PST From: Allen VanGelder Subject: Those interested in AI might want to read ... [Reprinted from the SU-SCORE bboard.] [Those interested in AI might want to read ...] the article in November *Psychology Today* about Francis Crick and Graeme Michison's neural network simulations. Title is "The Dream Machine", p. 22. ------------------------------ Date: Sun 20 Nov 83 18:50:27-PST From: David Rogers Subject: Re: Those interested in AI might want to read... [Reprinted from the SU-SCORE bboard.] I would guess that the "Psychology Today" article is a simplified form of the Crick & Michelson paper which came out in "Nature" about 2 months ago. Can't comment on the Psychology Today article, but the Nature article was stimulating and provocative. The same issue of Nature has a paper (referred to by Crick) of a simulation which was even better than the Crick paper (sorry, Francis!). ------------------------------ Date: Mon 21 Nov 83 09:58:04-PST From: Benjamin Grosof Subject: Weizenbaum review of "The Fifth Generation": hot stuff! [Reprinted from the SU-SCORE bboard.] The current issue of the NY REview of Books contains a review by Joseph Weizenbaum of MIT (Author of "Computer Power and Human Reason", I think) of Feigenbaum and McCorduck's "The Fifth Generation". Warning: it is scathing and controversial, hence great reading. --Benjamin ------------------------------ Date: Wed 23 Nov 83 14:38:38-PST From: Wilkins Subject: why AI is hard to read There is one reason much AI literature is hard to read. It is common for authors to invent a whole new set of jargon to describe their system, instead of desribing it in some common language (e.g., first order logic) or relating it to previous well-understood systems or principles. In recent years there has been an increased awareness of this problem, and hopefully things are improving and will continue to do so. There are also a lot more submissions now to IJCAI, etc, so higher standards end up being applied. Keep truckin' David Wilkins ------------------------------ Date: 21 Nov 1983 10:54-PST From: dietz%usc-cse%USC-ECL@SRI-NIC Reply-to: dietz%USC-ECL@SRI-NIC Subject: Economic effects of automation Reply to Marcel Schoppers (AIList 1:101): I agree that "computers will eliminate some jobs but create others" is a feeble excuse. There's not much evidence for it. Even if it's true, those whose jobs skills are devalued will be losers. But why should this bother me? I don't buy manufactured goods to employ factory workers, I buy them to gratify my own desires. As a computer scientist I will not be laid off; indeed, automation will increase the demand for computer professionals. I will benefit from the higher quality and lower prices of manufactured goods. Automation is entirely in my interest. I need no excuse to support it. ... I very much appreciated the idea ... that we should be building expert systems in economics to help us plan and control the effects of our research. This sounds like an awful waste of time to me. We have no idea how to predict the economic effects of much of anything except at the most rudimentary levels, and there is no evidence that we will anytime soon (witness the failure of econometrics). There would be no way to test the systems. Building expert systems is not a substitute for understanding. Automating medicine and law: a much better idea is to eliminate or scale back the licensing requirements that allow doctors and lawyers to restrict entry into their fields. This would probably be necessary to get much benefit from expert systems anyway. ------------------------------ Date: 22 Nov 83 11:27:05-PST (Tue) From: decvax!genrad!security!linus!utzoo!dciem!mmt @ Ucb-Vax Subject: Re: just a reminder... - (nf) Article-I.D.: dciem.501 It seems a little dangerous "to send machines where doctors won't go" - you'll get the machines treating the poor, and human experts for the privileged few. If the machines were good enough, I wouldn't mind being underpriveleged. I'd rather be flown into a foggy airport by autopilot than human pilot. Martin Taylor {allegra,linus,ihnp4,uw-beaver,floyd,ubc-vision}!utcsrgv!dciem!mmt ------------------------------ Date: 22 Nov 1983 13:06:13-EST (Tuesday) From: Doug DeGroot Subject: Logic Programming Symposium (long message) [Excerpt from a notice in the Prolog Digest.] 1984 International Symposium on Logic Programming February 6-9, 1984 Atlantic City, New Jersey BALLY'S PARK PLACE CASINO Sponsored by the IEEE Computer Society For more information contact PERIERA@SRI-AI or: Registration - 1984 ISLP Doug DeGroot, Program Chairman IBM Thomas J. Watson Research Center P.O. Box 218 Yorktown Heights, NY 10598 STATUS Conference Tutorial Member, IEEE __ $155 __ $110 Non-member __ $180 __ $125 ____________________________________________________________ Conference Overview Opening Address: Prof. J.A. (Alan) Robinson Syracuse University Guest Speaker: Prof. Alain Colmerauer Univeristy of Aix-Marseille II Marseille, France Keynote Speaker: Dr. Ralph E. Gomory, IBM Vice President & Director of Research, IBM Thomas J. Watson Research Center Tutorial: An Introduction to Prolog Ken Bowen, Syracuse University 35 Papers, 11 Sessions (11 Countries, 4 Continents) Preliminary Conference Program Session 1: Architectures I __________________________ 1. Parallel Prolog Using Stack Segments on Shared-memory Multiprocessors Peter Borgwardt (Univ. Minn) 2. Executing Distributed Prolog Programs on a Broadcast Network David Scott Warren (SUNY Stony Brook, NY) 3. AND Parallel Prolog in Divided Assertion Set Hiroshi Nakagawa (Yokohama Nat'l Univ, Japan) 4. Towards a Pipelined Prolog Processor Evan Tick (Stanford Univ,CA) and David Warren Session 2: Architectures II ___________________________ 1. Implementing Parallel Prolog on a Multiprocessor Machine Naoyuki Tamura and Yukio Kaneda (Kobe Univ, Japan) 2. Control of Activities in the OR-Parallel Token Machine Andrzej Ciepielewski and Seif Haridi (Royal Inst. of Tech, Sweden) 3. Logic Programming Using Parallel Associative Operations Steve Taylor, Andy Lowry, Gerald Maguire, Jr., and Sal Stolfo (Columbia Univ,NY) Session 3: Parallel Language Issues ___________________________________ 1. Negation as Failure and Parallelism Tom Khabaza (Univ. of Sussex, England) 2. A Note on Systems Programming in Concurrent Prolog David Gelertner (Yale Univ,CT) 3. Fair, Biased, and Self-Balancing Merge Operators in Concurrent Prolog Ehud Shaipro (Weizmann Inst. of Tech, Israel) Session 4: Applications in Prolog _________________________________ 1. Editing First-Order Proofs: Programmed Rules vs. Derived Rules Maria Aponte, Jose Fernandez, and Phillipe Roussel (Simon Bolivar Univ, Venezuela) 2. Implementing Parallel Algorithms in Concurrent Prolog: The MAXFLOW Experience Lisa Hellerstein (MIT,MA) and Ehud Shapiro (Weizmann Inst. of Tech, Israel) Session 5: Knowledge Representation and Data Bases __________________________________________________ 1. A Knowledge Assimilation Method for Logic Databases T. Miyachi, S. Kunifuji, H. Kitakami, K. Furukawa, A. Takeuchi, and H. Yokota (ICOT, Japan) 2. Knowledge Representation in Prolog/KR Hideyuki Nakashima (Electrotechnical Laboratory, Japan) 3. A Methodology for Implementation of a Knowledge Acquisition System H. Kitakami, S. Kunifuji, T. Miyachi, and K. Furukawa (ICOT, Japan) Session 6: Logic Programming plus Functional Programming - I ____________________________________________________________ 1. FUNLOG = Functions + Logic: A Computational Model Integrating Functional and Logical Programming P.A. Subrahmanyam and J.-H. You (Univ of Utah) 2. On Implementing Prolog in Functional Programming Mats Carlsson (Uppsala Univ, Sweden) 3. On the Integration of Logic Programming and Functional Programming R. Barbuti, M. Bellia, G. Levi, and M. Martelli (Univ. of Pisa and CNUCE-CNR, Italy) Session 7: Logic Programming plus Functional Programming- II ____________________________________________________________ 1. Stream-Based Execution of Logic Programs Gary Lindstrom and Prakash Panangaden (Univ of Utah) 2. Logic Programming on an FFP Machine Bruce Smith (Univ. of North Carolina at Chapel Hill) 3. Transformation of Logic Programs into Functional Programs Uday S. Reddy (Univ of Utah) Session 8: Logic Programming Implementation Issues __________________________________________________ 1. Efficient Prolog Memory Management for Flexible Control Strategies David Scott Warren (SUNY at Stony Brook, NY) 2. Indexing Prolog Clauses via Superimposed Code Words and Field Encoded Words Michael J. Wise and David M.W. Powers, (Univ of New South Wales, Australia) 3. A Prolog Technology Theorem Prover Mark E. Stickel, (SRI, CA) Session 9: Grammars and Parsing _______________________________ 1. A Bottom-up Parser Based on Predicate Logic: A Survey of the Formalism and Its Implementation Technique K. Uehara, R. Ochitani, O. Kakusho, and J. Toyoda (Osaka Univ, Japan) 2. Natural Language Semantics: A Logic Programming Approach Antonio Porto and Miguel Filgueiras (Univ Nova de Lisboa, Portugal) 3. Definite Clause Translation Grammars Harvey Abramson, (Univ. of British Columbia, Canada) Session 10: Aspects of Logic Programming Languages __________________________________________________ 1. A Primitive for the Control of Logic Programs Kenneth M. Kahn (Uppsala Univ, Sweden) 2. LUCID-style Programming in Logic Derek Brough (Imperial College, England) and Maarten H. van Emden (Univ. of Waterloo, Canada) 3. Semantics of a Logic Programming Language with a Reducibility Predicate Hisao Tamaki (Ibaraki Univ, Japan) 4. Object-Oriented Programming in Prolog Carlo Zaniolo (Bell Labs, New Jersey) Session 11: Theory of Logic Programming _______________________________________ 1. The Occur-check Problem in Prolog David Plaisted (Univ of Illinois) 2. Stepwise Development of Operational and Denotational Semantics for Prolog Neil D. Jones (Datalogisk Inst, Denmark) and Alan Mycroft (Edinburgh Univ, Scotland) ___________________________________________________________ An Introduction to Prolog A Tutorial by Dr. Ken Bowen Outline of the Tutorial - AN OVERVIEW OF PROLOG - Facts, Databases, Queries, and Rules in Prolog - Variables, Matching, and Unification - Search Spaces and Program Execution - Non-determinism and Control of Program Execution - Natural Language Processing with Prolog - Compiler Writing with Prolog - An Overview of Available Prologs Who Should Take the Tutorial The tutorial is intended for both managers and programmers interested in understanding the basics of logic programming and especially the language Prolog. The course will focus on direct applications of Prolog, such as natural language processing and compiler writing, in order to show the power of logic programming. Several different commercially available Prologs will be discussed and compared. About the Instructor Dr. Ken Bowen is a member of the Logic Programming Research Group at Syracuse University in New York, where he is also a Professor in the School of Computer and Information Sciences. He has authored many papers in the field of logic and logic programming. He is considered to be an expert on the Prolog programming language. ------------------------------ End of AIList Digest ******************** 28-Nov-83 09:43:11-PST,14310;000000000001 Mail-From: LAWS created at 28-Nov-83 09:41:25 Date: Mon 28 Nov 1983 09:32-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #104 To: AIList@SRI-AI AIList Digest Monday, 28 Nov 1983 Volume 1 : Issue 104 Today's Topics: Information Retrieval - Request, Programming Languages - Lisp Productivity, AI and Society - Expert Systems, AI Funding - Capitalistic AI, Humor - Problem with Horn Clauses, Seminar - Introspective Problem Solver, Graduate Program - Social Impacts at UC-Irvine ---------------------------------------------------------------------- Date: Sun, 27 Nov 83 11:41 EST From: Ed Fox Subject: Request for machine readable volumes, info. on retrieval projects Please send details of how to obtain any machine readable documents such as books, reference volumes, encyclopedias, dictionaries, journals, etc. These would be utilized for experiments in information retrieval. This is not aimed at large bibliographic databases but rather at finding a few medium to long items that exist both in book form and full text computer tape versions (readable under UNIX or VMS). Information on existing or planned projects for retrieval of passages (e.g., paragraphs or pages) from books, encyclopedias, electronic mail digests, etc. would also be helpful. I look forward to your reply. Thanks in advance, Ed Fox. Dr. Edward A. Fox, Dept. of Computer Science, 562 McBryde Hall, Virginia Polytechnic Institute and State University (VPI&SU or Virginia Tech), Blacksburg, VA 24061; (703)961-5113 or 6931; fox%vpi@csnet-relay via csnet, foxea%vpivm1.bitnet@berkeley via bitnet ------------------------------ Date: 25 Nov 83 22:47:27-PST (Fri) From: pur-ee!uiucdcs!smu!leff @ Ucb-Vax Subject: lisp productivity question - (nf) Article-I.D.: uiucdcs.4149 Is anybody aware of study's on productivity studies for lisp? 1. Can lisp programmers program in lisp at the same number of lines per day,week,month as in 'regular' languages like pascal, pl/1, etc. 2. Has anybody tried to write a fairly large program that normally would be done in lisp in a regular language and compared the number of lines ratio. In APL, a letter to Comm. ACM reported that APL programs took one fifth the number of lines as equivalent programs in regular language and took about twice as long per line to debug. Thus APL improved the productivity to get a function done by about a factor of two. I am curious if anything similar has been done for lisp. [One can, off course, write any APL program body as a single line. I suspect it would not take much longer to write that way, but it would be impossible to modify a week later. Much the same could be said for undocumented and poorly structured Lisp code. -- KIL] ------------------------------ Date: 22 Nov 83 21:01:33-PST (Tue) From: decvax!genrad!grkermit!masscomp!clyde!akgua!psuvax!lewis @ Ucb-Vax Subject: Re:Re: just a reminder... - (nf) Article-I.D.: psuvax.359 Why should it be dangerous to have machines treating the poor? There is no reason to believe that human experts will always be superior to machines; in fact, a carefully designed expert system could embody all the skill of the world's best diagnosticians. In addition, an expert system would never get tired or complain about its pay. On the other hand, perhaps you are worried about the machine lacking 'human' insight or compassion. I don't think anyone is suggesting that these qualities can or should be built into such a system. Perhaps we will see a new generation of medical personnel whose job will be to use the available AI facilities to make the most accurate diagnoses, and help patients interface with the system. This will provide patients with the best medical knowledge available, and still allow personal interaction between patients and technicians. -jim lewis psuvax!lewis ------------------------------ Date: 24 Nov 83 22:46:53-PST (Thu) From: pur-ee!uiucdcs!uokvax!emjej @ Ucb-Vax Subject: Re: just a reminder... - (nf) Article-I.D.: uiucdcs.4127 Re sending machines where doctors won't go: do you really think that it's better that poor people not be treated at all than treated by a machine? That's a bit much for me to swallow. James Jones ------------------------------ Date: 22 Nov 83 19:37:14-PST (Tue) From: pur-ee!uiucdcs!uicsl!Anonymous @ Ucb-Vax Subject: Capitalistic AICapitalistic AI - (nf) Article-I.D.: uiucdcs.4071 Have you had your advisor leave to make megabucks in industry? Seriously, I feel that this is a major problem for AI. There is an extremely limited number of AI professors and a huge demand from venture capitalists to set them up in a new company. Even fresh PhD's are going to be disappearing into industry when they can make several times the money they would in academia. The result is an acute (no make that terminal) shortage of professors to oversee the new research generation. The monetary imbalance can only grow as AI grows. At this university (UI) there are lots (hundreds?) of undergrads who want to study AI; and about 8 professors to teach them. Maybe the federal government ought to recognize that this imbalance hurts our technological competitiveness. What will prevent academic flight? Will IBM, Digital, and WANG support professors or will they start hiring them away? Here are a few things needed to keep the schools strong: 1) Higher salaries for profs in "critical areas." (maybe much higher) 2) Long term funding of research centers. (buildings, equipment, staff) 3) University administration support for capitalizing on the results of research, either through making it easy for a professor to maintain a dual life, or by setting up a university owned company to develop and sell the results of research. ------------------------------ Date: 14 Nov 83 17:26:03-PST (Mon) From: harpo!floyd!clyde!akgua!psuvax!burdvax!sjuvax!bbanerje @ Ucb-Vax Subject: Problem with Horn Clauses. Article-I.D.: sjuvax.140 As a novice to Prolog, I have a problem determining whether a clause is Horn, or non Horn. I understand that a clause of the form : A + ~B + ~C is a Horn Clause, While one of the form : A + B + ~C is non Horn. However, my problem comes when trying to determine if the following Clause is Horn or non-Horn. ------------\ / _ \ /_________ / \__** _# # ** (_ o o _) __________ xx ! xx ! HO HO HO ! xxx \_/xxx __/----------- xxxxxxxxxx Happy Holidays Everyone! -- Binayak Banerjee {bpa!astrovax!burdvax}!sjuvax!bbanerje ------------------------------ Date: 11/23/83 11:48:29 From: AGRE Subject: John Batali at the AI Revolving Seminar 30 November [Forwarded by SASW@MIT-MC] John Batali Trying to build an introspective problem-solver Wednesday 30 November at 4PM 545 Tech Sq 8th floor playroom Abstract: I'm trying to write a program that understands how it works, and uses that understanding to modify and improve its performance. In this talk, I'll describe what I mean by "an introspective problem-solver", discuss why such a thing would be useful, and give some ideas about how one might work. We want to be able to represent how and why some course of action is better than another in certain situations. If we take reasoning to be a kind of action, then we want to be able to represent considerations that might be relevant during the process of reasoning. For this knowledge to be useful the program must be able to reason about itself reasoning, and the program must be able to affect itself by its decisions. A program built on these lines cannot think about every step of its reasoning -- because it would never stop thinking about "how to think about" whatever it is thinking about. On the other hand, we want it to be possible for the program to consider any and all of its reasoning steps. The solution to this dilemma may be a kind of "virtual reasoning" in which a program can exert reasoned control over all aspects of its reasoning process even if it does not explicitly consider each step. This could be implemented by having the program construct general reasoning plans which are then run like programs in specific situations. The program must also be able to modify reasoning plans if they are discovered to be faulty. A program with this ability could then represent itself as an instance of a reasoning plan. Brian Smith's 3-LISP achieves what he calls "reflective" access and causal connection: A 3-LISP program can examine and modify the state of its interpreter as it is running. The technical tricks needed to make this work will also find their place in an introspective problem-solver. My work has involved trying to make sense of these issues, as well as working on a representation of planning and acting that can deal with real world goals and constraints as well as with those of the planning and plan-execution processes. ------------------------------ Date: 25 Nov 1983 1413-PST From: Rob-Kling Subject: Social Impacts Graduate Program at UC-Irvine CORPS ------- A Graduate Program on Computing, Organizations, Policy, and Society at the University of California, Irvine This interdisciplinary program at the University of California, Irvine provides an opportunity for scholars and students to investigate the social dimensions of computerization in a setting which supports reflective and sustained inquiry. The primary educational opportunities are a PhD programs in the Department of Information and Computer Science (ICS) and MS and PhD programs in the Graduate School of Management (GSM). Students in each program can specialize in studying the social dimensions of computing. Several students have recieved graduate degrees from ICS and GSM for studying topics in the CORPS program. The faculty at Irvine have been active in this area, with many interdisciplinary projects, since the early 1970's. The faculty and students in the CORPS program have approached them with methods drawn from the social sciences. The CORPS program focuses upon four related areas of inquiry: 1. Examining the social consequences of different kinds of computerization on social life in organizations and in the larger society. 2. Examining the social dimensions of the work and industrial worlds in which computer technologies are developed, marketed, disseminated, deployed, and sustained. 3. Evaluating the effectiveness of strategies for managing the deployment and use of computer-based technologies. 4. Evaluating and proposing public policies which facilitate the development and use of computing in pro-social ways. Studies of these questions have focussed on complex information systems, computer-based modelling, decision-support systems, the myriad forms of office automation, electronic funds transfer systems, expert systems, instructional computing, personal computers, automated command and control systems, and computing at home. The questions vary from study to study. They have included questions about the effectiveness of these technologies, effective ways to manage them, the social choices that they open or close off, the kind of social and cultural life that develops around them, their political consequences, and their social carrying costs. The CORPS program at Irvine has a distinctive orientation - (i) in focussing on both public and private sectors, (ii) in examining computerization in public life as well as within organizations, (iii) by examining advanced and common computer-based technologies "in vivo" in ordinary settings, and (iv) by employing analytical methods drawn from the social sciences. Organizational Arrangements and Admissions for CORPS The primary faculty in the CORPS program hold appointments in the Department of Information and Computer Science and the Graduate School of Management. Additional faculty in the School of Social Sciences, and the Program on Social Ecology, have collaborated in research or have taught key courses for students in the CORPS program. Research is administered through an interdisciplinary research institute at UCI which is part of the Graduate Division, the Public Policy Research Organization. Students who wish additional information about the CORPS program should write to: Professor Rob Kling (Kling.uci-20b@rand-relay) Department of Information and Computer Science University of California, Irvine Irvine, Ca. 92717 or to: Professor Kenneth Kraemer Graduate School of Management University of California, Irvine Irvine, Ca. 92717 ------------------------------ End of AIList Digest ******************** 28-Nov-83 22:42:32-PST,14027;000000000001 Mail-From: LAWS created at 28-Nov-83 22:41:17 Date: Mon 28 Nov 1983 22:36-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #105 To: AIList@SRI-AI AIList Digest Tuesday, 29 Nov 1983 Volume 1 : Issue 105 Today's Topics: AI - Challenge & Responses & Query ---------------------------------------------------------------------- Date: 21 Nov 1983 12:25-PST From: dietz%usc-cse%USC-ECL@SRI-NIC Reply-to: dietz%USC-ECL@SRI-NIC Subject: Re: The AI Challenge I too am skeptical about expert systems. Their attraction seems to be as a kind of intellectual dustbin into which difficulties can be swept. Have a hard problem that you don't know (or that no one knows) how to solve? Build an expert system for it. Ken Laws' idea of an expert system as a very modular, hackable program is interesting. A theory or methodology on how to hack programs would be interesting and useful, but would become another AI spinoff, I fear. ------------------------------ Date: Wed 23 Nov 83 18:02:11-PST From: Michael Walker Subject: response to response to challenge Tom, I thought you made some good points in your response to Ralph Johnson in the AIList, but one of your claims is unsupported, important, and quite possibly wrong. The claim I refer to is "Expert systems can be built, debugged, and maintained more cheaply than other complicated systems. And hence, they can be targeted at applications for which previous technology was barely adequate." I would be delighted if this could be shown to be true, because I would very much like to show friends/clients in industry how to use AI to solve their problems more cheaply. However, there are no formal studies that compare a system built using AI methods to one built using other methods, and no studies that have attempted to control for other causes of differences in ease of building, debugging, maintaining, etc. such as differences in programmer experience, programming language, use or otherwise of structured programming techniques, etc.. Given the lack of controlled, reproducible tests of the effectiveness of AI methods for program development, we have fallen back on qualitative, intuitive arguments. The same sort of arguments have been and are made for structured programming, application generators, fourth-generation languages, high-level languages, and ADA. While there is some truth in the various claims about improved programmer productivity they have too often been overblown as The Solution To All Our Problems. This is the case with claiming AI is cheaper than any other methods. A much more reasonable statement is that AI methods may turn out to be cheaper / faster / otherwise better than other methods if anyone ever actually builds an effective and economically viable expert system. My own guess is that it is easier to develop AI systems because we have been working in a LISP programming environment that has provided tools like interpreted code, interactive debugging/tracing/editing, masterscope analysis, etc.. These points were made quite nicely in Beau Shiel's recent article in Datamation (Power Tools for Programming, I think was the title). None of these are intrinsic to AI. Many military and industry managers who are supporting AI work are going to be very disillusioned in a few years when AI doesn't deliver what has been promised. Unsupported claims about the efficacy of AI aren't going to help. It could hurt our credibility, and thereby our funding and ability to continue the basic research. Mike Walker WALKER@SUMEX-AIM.ARPA ------------------------------ Date: Fri 25 Nov 83 17:40:44-PST From: Tom Dietterich Subject: Re: response to response to challenge Mike, While I would certainly welcome the kinds of controlled studies that you sketched in your msg, I think my claim is correct and can be supported. Virtually every expert system that has been built has been targeted at tasks that were previously untouched by computing technology. I claim that the reason for this is that the proper programming methodology was needed before these tasks could be addressed. I think the key parts of that methodology are (a) a modular, explicit representation of knowledge, (b) careful separation of this knowledge from the inference engine, and (c) an expert-centered approach in which extensive interviews with experts replace attempts by computer people to impose a normative, mathematical theory on the domain. Since there are virtually no cases where expert systems and "traditional" systems have been built to perform the same task, it is difficult to support this claim. If we look at the history of computers in medicine, however, I think it supports my claim. Before expert systems techniques were available, many people had attempted to build computational tools for physicians. But these tools suffered from the fact that they were often burdened with normative theories and often ignored the clinical aspects of disease diagnosis. I blame these deficiencies on the lack of an "expert-centered" approach. These programs were also difficult to maintain and could not produce explanations because they did not separate domain knowledge from the inference engine. I did not claim anywhere in my msg that expert systems techniques are "The Solution to All Our Problems". Certainly there are problems for which knowledge programming techniques are superior. But there are many more for which they are too expensive, too slow, or simply inappropriate. It would be absurd to write an operating system in EMYCIN, for example! The programming advances that would allow operating systems to be written and debugged easily are still undiscovered. You credit fancy LISP environments for making expert systems easy to write, debug, and maintain. I would certainly agree: The development of good systems for symbolic computing was an essential prerequisite. However, the level of program description and interpretation in EMYCIN is much higher than that provided by the Interlisp system. And the "expert-centered" approach was not developed until Ted Shortliffe's dissertation. You make a very good point in your last paragraph: Many military and industry managers who are supporting AI work are going to be very disillusioned in a few years when AI doesn't deliver what has been promised. Unsupported claims about the efficacy of AI aren't going to help. It could hurt our credibility, and thereby our funding and ability to continue the basic research. AI (at least in Japan) has "promised" speech understanding, language translation, etc. all under the rubric of "knowledge-based systems". Existing expert-systems techniques cannot solve these problems. We need much more research to determine what things CAN be accomplished with existing technology. And we need much more research to continue the development of the technology. (I think these are much more important research topics than comparative studies of expert-systems technology vs. other programming techniques.) But there is no point in minimizing our successes. My original message was in response to an accusation that AI had no merit. I chose what I thought was AI's most solid contribution: an improved programming methodology for a certain class of problems. --Tom ------------------------------ Date: Fri 25 Nov 83 17:52:47-PST From: Tom Dietterich Subject: Re: Clarifying my "AI Challange" Although I've written three messages on this topic already, I guess I've never really addressed Ralph Johnson's main question: My question, though, is whether AI is really going to change the world any more than the rest of computer science is already doing. Are the great promises of AI going to be fulfilled? My answer: I don't know. I view "the great promises" as goals, not promises. If you are a physicalist and believe that human beings are merely complex machines, then AI should in principle succeed. However, I don't know if present AI approaches will turn out to be successful. Who knows? Maybe the human brain is too complex to ever be understood by the human brain. That would be interesting to demonstrate! --Tom ------------------------------ Date: 24 Nov 83 5:00:32-PST (Thu) From: pur-ee!uiucdcs!smu!leff @ Ucb-Vax Subject: Re: The AI Challenge - (nf) Article-I.D.: uiucdcs.4118 There was a recent discussion of an AI project that was done at ONR on determining the cause of a chemical spill in a large chemical plant with various ducts and pipes and manholes, etc. I argued that the thing was just an application of graph algorithms and searching techniques. (That project was what could be done in three days by an AI team as part of a challenge from ONR and quite possibly is not representative.) Theorem proving using resolution is something that someone with just a normal algorithms background would not simply come up with 'as an application of normal algorithms.' Using if-then rules perhaps might be a search of the type you might see an algorithms book. Although, I don't expect the average CS person with a background in algorithms to come up with that application although once it was pointed out it would be quite intuitive. One interesting note is that although most of the AI stuff is done in LISP, a big theorem proving program discussed by Wos at a recent IEEE meeting here was written in PASCAL. It did some very interesting things. One point that was made is that they submitted a paper to a logic journal. Although the journal agreed the results were worth publishing, the "computer stuff" had to go. Continuing on this rambling aside, some people submitted results in mechanical engineering using a symbolic manipulator referencing the use of the program in a footnote. The poor referee conscientiously tried to duplicate the derivations manually. Finally he noticed the reference and sent a letter back saying that they must put symbolic manipulation by computer in the covering. Getting back to the original subject, I had a discussion with someone doing research in daemons. After he explained to me what daemons were, I came to the conclusion they were a fancy name for what you described as a hack. A straightforward application of theorem proving or if-then rule techniques would be inefficient or otherwise infeasable so one puts an exception in to handle a certain kind of a case. What is the difference between that an error handler for zero divides rather than putting a statement everywhere one does a division? Along the subject of hacking, a DATAMATION article on 'Real Programmers Don't Use PASCAL.' in which he complained about the demise of the person who would modify a program on the fly using the switch register, etc. He remarkeed at the end that some of the debugging techniques in LISP AI environments were starting to look like the old style techniques of assembler hackers. ------------------------------ Date: 24 Nov 83 22:29:44-PST (Thu) From: pur-ee!notes @ Ucb-Vax Subject: Re: The AI Challenge - (nf) Article-I.D.: pur-ee.1148 As an aside to this discussion, I'm curious as to just what everyone thinks of when they think of AI. I am a student at Purdue, which has absolutely nothing in the way of courses on what *I* consider AI. I have done a little bit of reading on natural language processing, but other than that, I haven't had much of anything in the way of instruction on this stuff, so maybe I'm way off base here, but when I think of AI, I primarily think of: 1) Natural Language Processing, first and foremost. In this, I include being able to "read" it and understand it, along with being able to "speak" it. 2) Computers "knowing" things - i.e., stuff along the lines of the famous "blocks world", where the "computer" has notions of pyramids, boxes, etc. 3) Computers/programs which can pass the Turing test (I've always thought that ELIZA sort of passes this test, at least in the sense that lots of people actually think the computer understood their problems). 4) Learning programs, like the tic-tac-toe programs that remember that "that" didn't work out, only on a much more grandiose scale. 5) Speech recognition and understanding (see #1). For some reason, I don't think of pattern recognition (like analyzing satellite data) as AI. After all, it seems to me that this stuff is mostly just "if it's trees, if it's a road, etc.", which doesn't really seem like "intelligence". [If it were that easy, I'd be out of a job. -- KIL] What do you think of when I say "Artificial Intelligence"? Note that I'm NOT asking for a definition of AI, I don't think there is one. I just want to know what you consider AI, and what you consider "other" stuff. Another question -- assuming the (very) hypothetical situation where computers and their programs could be made to be "infinitely" intelligent, what is your "dream program" that you'd love to see written, even though it realistically will probably never be possible? Jokingly, I've always said that my dream is to write a "compiler that does what I meant, not what I said". --Dave Curry decvax!pur-ee!davy eevax.davy@purdue ------------------------------ End of AIList Digest ******************** 29-Nov-83 12:59:19-PST,20343;000000000001 Mail-From: LAWS created at 29-Nov-83 12:58:05 Date: Tue 29 Nov 1983 12:50-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #106 To: AIList@SRI-AI AIList Digest Wednesday, 30 Nov 1983 Volume 1 : Issue 106 Today's Topics: Conference - Logic Conference Correction, Intelligence - Definitions, AI - Definitions & Research Methodology & Jargon, Seminar - Naive Physics ---------------------------------------------------------------------- Date: Mon 28 Nov 83 22:32:29-PST From: PEREIRA@SRI-AI.ARPA Subject: Correction The ARPANET address in the announcement of the IEEE 1984 Logic Programming Symposium should be PEREIRA@SRI-AI, not PERIERA@SRI-AI. Fernando Pereira [My apologies. I am the one who inserted Dr. Pereira's name incorrectly. I was attempting to insert information from another version of the same announcement that also reached the AIList mailbox. -- KIL] ------------------------------ Date: 21 Nov 83 6:04:05-PST (Mon) From: decvax!mcvax!enea!ttds!alf @ Ucb-Vax Subject: Re: Behavioristic definition of intelligence Article-I.D.: ttds.137 Doesn't the concept "intelligence" have some characteristics in common with a concept such as "traffic"? It seems obvious that one can measure such entities as "traffic intensity" and the like thereby gaining an indirect understanding of the conditions that determine the "traffic" but it seems very difficult to find a direct measure of "traffic" as such. Some may say that "traffic" and "traffic intensity" are synonymous concepts but I don't agree. The common opinion among psychologists seems to be that "intelligence" is that which is measured by an intelligence test. By measuring a set of problem solving skills and weighing the results together we get a value. Why not call it "intelligence" ? The measure could be applicable to machine intelligence also as soon as (if ever) we teach the machines to pass intelligence tests. It should be quite clear that "intelligence" is not the same as "humanness" which is measured by a Turing test. ------------------------------ Date: Sat, 26 Nov 83 2:09:14 EST From: A B Cooper III Subject: Where wise men fear to tread Being nothing more than an amateur observer on the AI scene, I hesitate to plunge in like a fool. Nevertheless, the roundtable on what constitutes intelligence seems ed to cover many interesting hypotheses: survivability speed of solving problems etc but one. Being married to a professional educator, I've found that the common working definition of intelligence is the ability TO LEARN. The more easily one learns new material, the more intelligent one is said to be. The more quickly one learns new material, the more intelligent one is said to be. One who can learn easily and quickly across a broad spectrum of subjects is said to be more intelligent than one whose abilities are concentrated in one or two areas. One who learns only at an average rate, except for one subject area in which he or she excells far above the norms is thought to be TALENTED rather than INTELLIGENT. It seems to be believed that the most intelligent folks learn easily and rapidly without regard to the level of material. They assimilate the difficult with the easy. Since this discussion was motivated, at least in part, by the desire to understand what an "intelligent" computer program should do, I feel that we should re-visit some of our terminology. In the earlier days of Computer Science, I seem to recall some excitement about machines (computers) that could LEARN. Was this the precursor of AI? I don't know. If we build an EXPERT SYSTEM, have we built an intelligent machine (can it assimilate new knowledge easily and quickly), or have we produced a "dumb" expert? Indeed, aren't many of our AI or knowledge-based or expert systems really something like "dumb" experts? ------------------------ You might find the following interesting: Siegler, Robert S, "How Knowledge Influences Learning," AMERICAN SCIENTIST, v71, Nov-Dec 1983. In this reference, Siegler addresses the questions of how children learn and what they know. He points out that the main criticism of intelligence tests (that they measure 'knowledge' and not 'aptitute') may miss the mark--that knowledge and learning may be linked, in humans anyway, in ways that traditional views have not considered. ------------------------- In any case, should we not be addressing as a primary research objective, how to make our 'expert systems' into better learners? Brint Cooper abc@brl.arpa ------------------------------ Date: 23 Nov 83 11:27:34-PST (Wed) From: dambrosi @ Ucb-Vax Subject: Re: Intelligence Article-I.D.: ucbvax.373 Hume once said that when a discussion or argument seems to be interminable and without discernable progress, it is worthwhile to attempt to produce a concrete visualisation of the concept being argued about. Often, he claimed, this will be IMPOSSIBLE to do, and this will be evidence that the word being argued about is a ringer, and the discussion pointless. In more modern parlance, these concepts are definitionally empty for most of us. I submit the following definition as the best presently available: Intelligence consists of perception of the external environment (e.g. vision), knowledge representation, problem solving, learning, interaction with the external environment (e.g. robotics), and communication with other intelligent agents (e.g. natural language understanding). (note the conjunctive connector) If you can't guess where this comes from, check AAAI83 procedings table of contents. bruce d'ambrosio dambrosi%ucbernie@berkeley ------------------------------ Date: Tuesday, 29 Nov 1983 11:43-PST From: narain@rand-unix Subject: Re: AI Challenge AI is advanced programming. We need to solve complex problems involving reasoning, and judgment. So we develop appropriate computer techniques (mainly software) for that. It is our responsibility to invent techniques that make development of efficient intelligent computer programs easier, debuggable, extendable, modifiable. For this purpose it is only useful to learn whatever we can from traditional computer science and apply it to the AI effort. Tom Dietterich said: >> Your view of "knowledge representations" as being identical with data >> structures reveals a fundamental misunderstanding of the knowledge vs. >> algorithms point. Most AI programs employ very simple data structures >> (e.g., record structures, graphs, trees). Why, I'll bet there's not a >> single AI program that uses leftist-trees or binomial queues! But, it >> is the WAY that these data structures are employed that counts. We at Rand have ROSS (Rule Oriented Simulation System) that has been employed very successfully for developing two large scale simulations (one strategic and one tactical). One implementation of ROSS uses leftist trees for maintaining event queues. Since these queues are in the innermost loop of ROSS's operation, it was only sensible to make them as efficient as possible. We think we are doing AI. Sanjai Narain Rand Corp. ------------------------------ Date: Tue, 29 Nov 83 11:31:54 PST From: Michael Dyer Subject: defining AI, AI research methodology, jargon in AI (long msg) This is in three flaming parts: (I'll probably never get up the steam to respond again, so I'd better get it all out at once.) Part I. "Defining intelligence", "defining AI" and/or "responding to AI challenges" considered harmful: (enough!) Recently, I've started avoiding/ignoring AIList since, for the most part, it's been a endless discussion on "defining A/I" (or, most recently) defending AI. If I spent my time trying to "define/defend" AI or intelligence, I'd get nothing done. Instead, I spend my time trying to figure out how to get computers to achieve some task -- exhibit some behavior -- which might be called intelligent or human-like. If/whenever I'm partially successful, I try to keep track about what's systematic or insightful. Both failure points and partial success points serve as guides for future directions. I don't spend my time trying to "define" intelligence by BS-ing about it. The ENTIRE enterprise of AI is the attempt to define intelligence. Here's a positive suggestion for all you AIList-ers out there: I'd be nice to see more discussion of SPECIFIC programs/cognitive models: their Assumptions, their failures, ways to patch them, etc. -- along with contentful/critical/useful suggestions/reactions. Personally, I find Prolog Digest much more worthwhile. The discussions are sometimes low level, but they almost always address specific issues, with people often offering specific problems, code, algorithms, and analyses of them. I'm afraid AIList has been taken over by people who spend so much time exchanging philosophical discussions that they've chased away others who are very busy doing research and have a low BS tolerance level. Of course, if the BS is reduced, that means that the real AI world will have to make up the slack. But a less frequent digest with real content would be a big improvement. {This won't make me popular, but perhaps part of the problem is that most of the contributors seem to be people who are not actually doing AI, but who are just vaguely interested in it, so their speculations are ill-informed and indulgent. There is a use for this kind of thing, but an AI digest should really be discussing research issues. This gets back to the original problem with this digest -- i.e. that researchers are not using it to address specific research issues which arise in their work.} Anyway, here are some examples of task/domains topic that could be addressed. Each can be considered to be of the form: "How could we get a computer to do X": Model Dear Abby. Understand/engage in an argument. Read an editorial and summarize/answer questions about it. Build a daydreamer Give legal advice. Write a science fiction short story ... {I'm an NLP/Cognitive modeling person -- that's why my list may look bizzare to some people} You researchers in robotics/vision/etc. could discuss, say, how to build a robot that can: climb stairs ... recognize a moving object ... etc. People who participate in this digest are urged to: (1) select a task/domain, (2) propose a SPECIFIC example which represents PROTOTYPICAL problems in that task/domain, (3) explain (if needed) why that specific example is prototypic of a class of problems, (4) propose a (most likely partial) solution (with code, if at that stage), and 4) solicit contentful, critical, useful, helpful reactions. This is the way Prolog.digest is currently functioning, except at the programming language level. AIList could serve a useful purpose if it were composed of ongoing research discussions about SPECIFIC, EXEMPLARY problems, along with approaches, their limitations, etc. If people don't think a particular problem is the right one, then they could argue about THAT. Either way, it would elevate the level of discussion. Most of my students tell me that they no longer read AIList. They're turned off by the constant attempts to "defend or define AI". Part II. Reply to R-Johnson Some of R-Johnson's criticisms of AI seem to stem from viewing AI strictly as a TOOLS-oriented science. {I prefer to refer to STRUCTURE-oriented work (ie content-free) as TOOLS-oriented work and CONTENT-oriented work as DOMAIN or PROCESS-oriented. I'm referring to the distinction that was brought up by Schank in "The Great Debate" with McCarthy at AAAI-83 Wash DC). In general, tools-oriented work seems more popular and accepted than content/domain-oriented work. I think this is because: 1. Tools are domain independent, so everyone can talk about them without having to know a specific domain -- kind of like bathroom humor being more universally communicable than topical-political humor. 2. Tools have nice properties: they're general (see #1 above); they have weak semantics (e.g. 1st order logic, lambda-calculus) so they're clean and relatively easy to understand. 3. No one who works on a tool need be worried about being accused of "ad hocness". 4. Breakthroughs in tools-research happen rarely, but when it does, the people associated with the breakthrough become instantly famous because everyone can use their tool (e.g. Prolog) In contrast, content or domain-oriented research and theories suffer from the following ills: 1. They're "ad hoc" (i.e. referring to THIS specific thing or other). 2. They have very complicated semantics, poorly understood, hard to extend, fragile, etc. etc. However, many of the most interesting problems pop up in trying to solve a specific problem which, if solved, would yield insight into intelligence. Tools, for the most part, are neutral with respect to content-oriented research questions. What does Prolog or Lisp have to say to me about building a "Dear Abby" natural language understanding and personal advice-giving program? Not much. The semantics of lisp or prolog says little about the semantics of the programs which researchers are trying to discover/write in Prolog or Lisp. Tools are tools. You take the best ones off the shelf you can find for the task at hand. I love tools and keep an eye out for tools-developments with as much interest as anyone else. But I don't fool myself into thinking that the availability of a tool will solve my research problems. {Of course no theory is exlusively one or the other. Also, there are LEVELS of tools & content for each theory. This levels aspect causes great confusion.} By and large, AIList discussions (when they get around to something specific) center too much around TOOLS and not PROCESS MODELS (ie SPECIFIC programs, predicates, rules, memory organizations, knowledge constructs, etc.). What distinguishes AI from compilers, OS, networking, or other aspects of CS are the TASKS that AI-ers choose. I want computers that can read "War and Peace" -- what problems have to be solved, and in what order, to achieve this goal? Telling me "use logic" is like telling me to "use lambda calculus" or "use production rules". Part III. Use and abuse of jargon in AI. Someone recently commented in this digest on the abuse of jargon in AI. Since I'm from the Yale school, and since Yale commonly gets accused of this, I'm going to say a few words about jargon. Different jargon for the same tools is BAD policy. Different jargon to distinguish tools from content is GOOD policy. What if Schank had talked about "logic" instead of "Conceptual Dependencies"? What a mistake that would have been! Schank was trying to specify how specific meanings (about human actions) combine during story comprehension. The fact that prolog could be used as a tool to implement Schank's conceptual dependencies is neutral with respect to what Schank was trying to do. At IJCAI-83 I heard a paper (exercise for the reader to find it) which went something like this: The work of Dyer (and others) has too many made-up constructs. There are affects, object primitives, goals, plans, scripts, settings, themes, roles, etc. All this terminology is confusing and unnecessary. But if we look at every knowledge construct as a schema (frame, whatever term you want here), then we can describe the problem much more elegantly. All we have to consider are the problems of: frame activation, frame deactivation, frame instantiation, frame updating, etc. Here, clearly we have a tools/content distinction. Wherever possible I actually implemented everything using something like frames-with-procedural-attachment (ie demons). I did it so that I wouldn't have to change my code all the time. My real interest, however, was at the CONTENT level. Is a setting the same as an emotion? Does the task: "Recall the last 5 restaurant you were at" evoke the same search strategies as "Recall the last 5 times you accomplished x", or "the last 5 times you felt gratitude."? Clearly, some classes of frames are connected up to other classes of frames in different ways. It would be nice if we could discover the relevant classes and it's helpful to give them names (ie jargon). For example, it turns out that many (but not all) emotions can be represented in terms of abstract goal situations. Other emotions fall into a completely different class (e.g. religious awe, admiration). In my program "love" was NOT treated as (at the content level) an affect. When I was at Yale, at least once a year some tools-oriented person would come through and give a talk of the form: "I can represent/implement your Scripts/Conceptual-Dependency/ Themes/MOPs/what-have-you using my tool X" (where X = ATNs, Horn clauses, etc.). I noticed that first-year students usually liked such talks, but the advanced students found them boring and pointless. Why? Because if you're content-oriented you're trying to answer a different set of questions, and discussion of the form: "I can do what you've already published in the literature using Prolog" simply means "consider Prolog as a nice tool" but says nothing at the content level, which is usually where the advanced students are doing their research. I guess I'm done. That'll keep me for a year. -- Michael Dyer ------------------------------ Date: Mon 28 Nov 83 08:59:57-PST From: Doug Lenat Subject: CS Colloq 11/29: John Seely Brown [Reprinted from the SU-SCORE bboard.] Tues, Nov 29, 3:45 MJH refreshments; 4:15 Terman Aud (lecture) A COMPUTATIONAL FRAMEWORK FOR A QUALITATIVE PHYSICS-- Giving computers "common-sense" knowledge about physical mechanisms John Seely Brown Cognitive Sciences Xerox, Palo Alto Research Center Humans appear to use a qualitative causal calculus in reasoning about the behavior of their physical environment. Judging from the kinds of explanations humans give, this calculus is quite different from the classical physics taught in classrooms. This raises questions as to what this (naive) physics is like, how it helps one to reason about the physical world and how to construct a formal calculus that captures this kind of reasoning. An analysis of this calculus along with a system, ENVISION, based on it will be covered. The goals for the qualitative physics are i) to be far simpler than classical physics and yet retain all the important distinctions (e.g., state, oscillation, gain, momentum), ii) to produce causal accounts of physical mechanisms, and (3) to provide a logic for common-sense, causal reasoning for the next generation of expert systems. A new framework for examining causal accounts has been suggested based on using collections of locally interacting processors to represent physical mechanisms. ------------------------------ End of AIList Digest ******************** 1-Dec-83 22:38:53-PST,15226;000000000001 Mail-From: LAWS created at 1-Dec-83 22:36:42 Date: Thu 1 Dec 1983 21:58-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #107 To: AIList@SRI-AI AIList Digest Friday, 2 Dec 1983 Volume 1 : Issue 107 Today's Topics: Programming Languages - Lisp Productivity, Alert - Psychology Today, Learning & Expert Systems, Intelligence - Feedback Model & Categorization, Scientific Method - Psychology, Puzzle - The Lady or the Tiger, Seminars - Commerce Representation & Learning Linguistic Categories ---------------------------------------------------------------------- Date: 27 Nov 83 16:57:39-PST (Sun) From: decvax!tektronix!tekcad!franka @ Ucb-Vax Subject: Re: lisp productivity question - (nf) Article-I.D.: tekcad.145 I don't have any documentation, but I heard once from an attendee at a workshop on design automation that someone had reported a 5:1 productivity improvement in LISP vs. C, PASCAL, etc. From personal experience I know this to be true, also. I once wrote a game program in LISP in two days. I later spent two weeks debugging the same game in a C version (I estimated another factor of 4 for a FORTRAN version). The nice thing about LISP is not that the amount of code written is less (although it is, usually by a factor of 2 to 3), but that its environment (even in the scrungy LISPs) is much easier to debug and modify code in. From the truly menacing, /- -\ but usually underestimated, <-> Frank Adrian (tektronix!tekcad!franka) [A caveat: Lisp is very well suited to the nature of game programs. A fair test would require that data processing and numerical analysis problems be included in the mix of test problems. -- KIL] ------------------------------ Date: Mon, 28 Nov 83 11:03 EST From: Steven Gutfreund Subject: Psychology Today The December issue of Psychology Today (V 17, #12) has some more articles that may be of interest to AI people. The issue is titled "USER FRIENDLY" and talks about technological advances that have made machines easier. The articles of interest are: On Papert, Minsky, and John Anderson page 26 An Article written by McCarthy page 46 An Interview with Alan Kay Page 50 (why they call him the Grand old Man is beyond me, Alan is only 43) - steve ------------------------------ Date: Tue 29 Nov 83 18:36:01-EST From: Albert Boulanger Subject: Learning Expert systems Re: Brint Cooper's remark on non-learning expert systems being "dumb": Yes, some people would agree with you. In fact, Dr. R.S. Michalski's group at the U of Illinois is building an Expert System, ADVISE, that incorporates learning capabilities. Albert Boulanger ABOULANGER@BBNG ------------------------------ Date: Wed, 30 Nov 83 09:07 PST From: NNicoll.ES@PARC-MAXC.ARPA Subject: "Intelligence" I see Intelligence as the sophistication of the deep structure mechanisms that generate both thought and behavior. These structures (per Albus), work as cross-coupled hierarchies of phase-locked loops, generating feedback hypotheses about the stimulus at each level of the hierarchy. These feedback hypotheses are better at predicting and matching the stimulus if the structure holds previous patterns that are similar to the present stimulus. Therefore, intelligence is a function of both the amount of knowledge possible to bring to bear on pattern matching a present problem (inference), and the number of levels in the structure of the hierarchy the organism (be it mechanical or organic) can bring to bear on breaking the stimulus/pattern down into its component parts and generate feedback hypotheses to adjust the organisms response at each level. I feel any structure sufficiently complex to exhibit intelligence, be it a bird-brained idiot whose height of reasoning is "find fish - eat fish", or "Deep Thought" who can break down the structures and reason about a whole world, should be considered intelligent, but with different "amounts" of intelligence, and possibly about different experiences. I do not think there is any "threshold" above which an organism can be considered intelligent and below which they are not. This level would be too arbitrary a structure for anything except very delimited areas. So, lets get on with the pragmatic aspects of this work, creating better slaves to do our scut work for us, our reasoning about single-mode structures too complex for a human brain to assimilate, our tasks in environments too dangerous for organic creatures, and our tasks too repetitious for the safety of the human brain/body structure, and move to a lower priority the re-creation of pseudo-human "intelligence". I think that would require a pseudo-human brain structure (combining both "Emotion" and "Will") that would be interesting only in research on humanity (create a test-bed wherein experiments that are morally unacceptable when performed on organic humans could be entertained). Nick Nicoll ------------------------------ Date: 29 Nov 83 20:47:33-PST (Tue) From: decvax!ittvax!dcdwest!sdcsvax!sdcsla!west @ Ucb-Vax Subject: Re: Intelligence and Categorization Article-I.D.: sdcsla.461 From: AXLER.Upenn-1100@Rand-Relay (David M. Axler - MSCF Applications Mgr.) I think Tom Portegys' comment in 1:98 is very true. Knowing whether or not a thing is intelligent, has a soul, etc., is quite helpful in letting us categorize it. And, without that categorization, we're unable to know how to understand it. Two minor asides that might be relevant in this regard: 1) There's a school of thought in the fields of linguistics, folklore, anthropology, and folklore, which is based on the notion (admittedly arguable) that the only way to truly understand a culture is to first record and understand its native categories, as these structure both its language and its thought, at many levels. (This ties in to the Sapir-Whorf hypothesis that language structures culture, not the reverse...) From what I've read in this area, there is definite validity in this approach. So, if it's reasonable to try and understand a culture in terms of its categories (which may or may not be translatable into our own culture's categories, of course), then it's equally reasonable for us to need to categorize new things so that we can understand them within our existing framework. Deciding whether a thing is or is not intelligent seems to be a hairier problem than "simply" categorizing its behavior and other attributes. As to point #1, trying to understand a culture by looking at how it categorizes does not constitute a validation of the process of categorization (particularly in scientific endeavours). Restated: There is no connection between the fact that anthropologists find that studying a culture's categories is a very powerful tool for aiding understanding, and the conclusion that we need to categorize new things to understand them. I'm not saying that categorization is useless (far from it), but Sapir-Whorf's work has no direct bearing on this subject (in my view). What I am saying is that while deciding to treat something as "intelligent", e.g., a computer chess program, may prove to be the most effective way of dealing with it in "normal life", it doesn't do a thing for understanding the thing. If you choose to classify the chess program as intelligent, what has that told you about the chess program? If you classify it as unintelligent...? I think this reflects more upon the interaction between you and the chess program than upon the structure of the chess program. -- Larry West UC San Diego -- ARPA: west@NPRDC -- UUCP: ucbvax!sdcsvax!sdcsla!west -- or ucbvax:sdcsvax:sdcsla:west ------------------------------ Date: 28 Nov 83 18:53:46-PST (Mon) From: harpo!eagle!mhuxl!ulysses!unc!mcnc!ncsu!fostel @ Ucb-Vax Subject: Rational Psych & Scientific Method Article-I.D.: ncsu.2416 Well, I hope this is the last time .... Again, I have been accused of ignorance; again the accustation is false. Its fortunate only my words can make it into this medium. I would appreciate the termination of this discussion, but will not stand by and be patronized without responding. All sane and rational people, hit the and go on to the next news item please. When I say psychologists do not do very good science I am talking about the exact same thing you are talking about. There is no escape. Those "rigorous" experiments sometime succeed in establishing some "facts", but they are sufficiently encumbered by lack of controls that one often does not know what to make of them. This is not to imply a critisism of psychologists as intellectually inferior to chemists, but the field is just not there yet. Is Linguistics a science? Is teaching a science? Laws (and usually morals) prevent the experiments we need, to do REAL controlled experiments; lack of understanding would probably prevent immediate progress even in the absence of those laws. Its a bit like trying to make a "scientific" study of a silicon wafer with 1850's tools and understanding of electronics. A variety of interesting facts could be established, but it is not clear that they would be very useful. Tack on some I/O systems and you could then perhaps allow the collection of reams of timing and capability data and could try to corrollate the results and try to build theories -- that LOOKS like science. But is it? In my book, to be a science, there must be a process of convergence in which the theories more ever closer to explaining reality, and the experiments become ever more precise. I don't see much convergence in experimental psychology. I see more of a cyclic nature to the theories .... ----GaryFostel---- P.S. There are a few other sciences which do not deserve the title, so don't feel singled out. Computer Science for example. ------------------------------ Date: Tue, 29 Nov 83 11:15 EST From: Chris Moss Subject: The Lady or the Tiger [Reprinted from the Prolog Digest.] Since it's getting near Christmas, here are a few puzzlers to solve in Prolog. They're taken from Raymond Smullyan's delightful little book of the above name. Sexist allusions must be forgiven. There once was a king, who decided to try his prisoners by giving them a logic puzzle. If they solved it they would get off, and get a bride to boot; otherwise ... The first day there were three trials. In all three, the king explained, the prisoner had to open one of two rooms. Each room contained either a lady or a tiger, but it could be that there were tigers or ladies in both rooms. On each room he hung a sign as follows: I II In this room there is a lady In one of these rooms there is and in the other room a lady and in one of these there is a tiger rooms there is a tiger "Is it true, what the signs say ?", asked the prisoner. "One of them is true", replied the king, "but the other one is false" If you were the prisoner, which would you choose (assuming, of course, that you preferred the lady to the tiger) ? ------------------------- For the second and third trials, the king explained that either both statements were true, or both are false. What is the situation ? Signs for Trial 2: I II At least one of these rooms A tiger is in the contains a tiger other room Signs for Trial 3: I II Either a tiger is in this room A lady is in the or a lady is in the other room other room Representing the problems is much more difficult than finding the solutions. The latter two test a sometimes ignored aspect of the [Prolog] language. Have fun ! ------------------------------ Date: 27 Nov 1983 20:42:46-EST From: Mark.Fox at CMU-RI-ISL1 Subject: AI talk [Reprinted from the CMU-AI bboard.] TITLE: Databases and the Logic of Business SPEAKER: Ronald M. Lee, IIASA Austria & LNEC Portugal DATE: Monday, Nov. 28, 1983 PLACE: MS Auditorium, GSIA ABSTRACT: Business firms differentiate themsleves with special products, services, etc. Nevertheless, commercial activity requires certain standardized concepts, e.g., a common temporal framework, currency of exchange, concepts of ownership and contractual obligation. A logical data model, called CANDID, is proposed for modelling these standardized aspects in axiomatic form. The practical value is the transportability of this knowledge across a wide variety of applications. ------------------------------ Date: 30 Nov 83 18:58:27 PST (Wednesday) From: Kluger.PA@PARC-MAXC.ARPA Reply-to: Kluger.PA@PARC-MAXC.ARPA Subject: HP Computer Colloquium 12/1/83 Professor Roman Lopez de Montaras Politecnico Universidade Barcelona A Learning System for Linguistic Categorization of Soft Observations We describe a human-guided feature classification system. A person teaches the denotation of subjective linguistic feature descriptors to the system by reference to examples. The resulting knowledge base of the system is used in the classification phase for interpetation of descriptions. Interpersonal descriptions are communicated via semantic translations of subjective descriptions. The advantage of a subjective linguistic description over more traditional arithmomorphic schemes is their high descriptor-feature consistency. This is due to the relative simplicity of the underlying cognitive process. This result is a high feature resolution for the overall cognitive perception and description processes. At present the system is still being used for categorization of "soft" observations in psychological research, but application in any person-machine system are conceivable. ------------------------------ End of AIList Digest ******************** 2-Dec-83 16:31:26-PST,18493;000000000001 Mail-From: LAWS created at 2-Dec-83 16:28:44 Date: Fri 2 Dec 1983 16:15-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #108 To: AIList@SRI-AI AIList Digest Saturday, 3 Dec 1983 Volume 1 : Issue 108 Today's Topics: Editorial Policy, AI Jargon, AI - Challenge Responses, Expert Systems & Knowledge Representation & Learning ---------------------------------------------------------------------- Date: Fri 2 Dec 83 16:08:01-PST From: Ken Laws Subject: Editorial Policy It has been suggested that the volume on this list is too high and the technical content is too low. Two people have recently written to me suggesting that the digest be converted to a magazine format with perhaps a dozen edited departments that would constitute alternating special issues. I appreciate their offers to serve as editors, but have no desire to change the AIList format. The volume has been high, but that is typical of new lists. I encourage technical contributions, but I do not wish to discourage general-interest discussions. AIList provides a forum for material not appropriate to journals and conferences -- "dumb" questions, requests for information, abstracts of work in progress, opinions and half-baked ideas, etc. I do not find these a waste of time, and attempts to screen any class of "uninteresting" messages will only deprive those who are interested in them. A major strength of AIList is that it helps us develop a common vocabulary for those topics that have not yet reached the textbook stage. If people would like to split off their own sublists, I will be glad to help. That might reduce the number of uninteresting messages each reader is exposed to, although the total volume of material would probably be higher. Narrow lists do tend to die out as their boom and bust cycles gradually lengthen, but AIList could serve as the channel by which members could regroup and recruit new members. The chief disadvantage of separate lists is that we would lose valuable cross-fertilization between disciplines. For the present, I simply ask that members be considerate when composing messages. Be concise, preferably stating your main points in list form for easy reference. Remember that electronic messages tend to seem pugnacious, so that even slight sarcasm may arouse numerous rebuttals and criticisms. It is unnecessary to marshall massive support for every claim since you will have the opportunity to reply to critics. Also, please keep in mind that AIList (under my moderatorship) is primarily concerned with AI and pattern recognition, not psychology, metaphysics, philosophy of science, or any other topic that has its own major following. We welcome any material that advances the progress of intelligent machines, but the hard-core discussions from other disciplines should be directed elsewhere. -- Ken Laws ------------------------------ Date: Tue 29 Nov 83 21:09:12-PST From: David Rogers Subject: Re: Dyer's flame In this life of this list a number of issues, among them intelligence, parallelism and AI, defense of AI, rational psychology, and others have been maligned as "pointless" or whatever. Without getting involved in a debate on "philosophy" vs. "real research", a quick scan of these topics shows them to be far from pointless. I regret that Dyer's students have stopped reading this list; perhaps they should follow his advice of submitting the right type of article to this list. As a side note, I am VERY interested in having people outside of mainstream AI participate in this list; while one sometimes wades through muddled articles of little value, this is more than repaid by the fresh viewpoints and occasional gem that would have been otherwise never been found. Ken Laws has done an excellent job grouping the articles by interest and topic; uninterested readers can then skip reading an entire volume, if the theme is uninteresting. A greater number of articles submitted can only improve this process; the burden is on those unsatisfied with the content of this board to submit them. I would welcome submissions of the kind suggested by Dr. Dyer, and hope that others will follow his advice and try to lead the board to whatever avenue they think is the most interesting. There's room here for all of us... David Rogers DRogers@SUMEX-AIM.ARPA ------------------------------ Date: Tue 29 Nov 83 22:24:14-PST From: PEREIRA@SRI-AI.ARPA Subject: Tools I agree with Michael Dyer's comments on the lack of substantive material in this list and on the importance of dealing with new "real" tasks rather than using old solutions of old problems to show off one's latest tool. However, I feel like adding two comments: 1. Some people (me included) have a limited supply of "writing energy" to write serious technical stuff: papers, proposals and the like. Raving about generalities, however, consumes much less of that energy per line than the serious stuff. The people who are busily writing substantive papers have no energy left to summarize them on the net. 2. Very special tools, in particular fortunate situations ("epiphanies"?!) can bring a new and better level of understanding of a problem, just by virtue of what can be said with the new tool, and how. Going the other direction, we all know that we need to change our tools to suit our problems. The paradigmatic relation between subject and tool is for me the one between classical physics and mathematical analysis, where tool and subject are intimately connected but yet distinct. Nothing of the kind has yet happened in AI (which shouldn't surprise us, seeing at how long it took to develop that other relationship...). Note: Knowing of my involvement with Prolog/logic programming, some reader of this might be tempted to think "Ahah! what he is really driving at is that logic/Horn clauses/Prolog [choose one] is that kind of tool for AI. Let me kill that presumption in the bud, these tool addicts are dangerous!" Gentle reader, save your flame! Only time will show whether anything of the kind is the case, and my private view on the subject is sufficiently complicated (confused?) that if I could disentangle it and write about it clearly I would have a paper rather than a net message... Fernando Pereira ------------------------------ Date: Wed 30 Nov 83 11:58:56-PST From: Wilkins Subject: jargon I understand Dyer's comments on what he calls the tool/content distinction. But it seems to me that the content distinctions he rightly thinks are important can often be expressed in terms of tools, and that it would be clearer to do so. He talked about handling one's last trip to the restaurant differently from the last time one is in love. I agree that this is an important distinction to make. I would like to see the difference expressed in "tools", e.g., "when handling a restaurant trip (or some similar class of events) our system does a chronological search down its list of events, but when looking for love, it does a best first search on its list of personal relationships." This is clearer and communicates more than saying the system has a "love-MOP" and a "restaurant-script". This is only a made up example -- I am not saying Mr. Dyer used the above words or that he does not explain things well. I am just trying to construct a non-personal example of the kind of thing to which I object, but that occurs often in the literature. ------------------------------ Date: Wed, 30 Nov 83 13:47 EST From: Steven Gutfreund Subject: McCarthy and 'mental' states In the December Psychology Today John McCarthy has a short article that raises a fairly contentious point. In his article he talks about how it is not necessarily a bad thing that people attribute "human" or what the calls 'mental' attributes to complex systems. Thus when someone anthropomorphises the actions of his/her car, boat, or terminal, one is engaging in a legitimate form of description of a complex process. Indeed he argues further that while currently most computer programs can still be understood by their underlying mechanistic properties, eventually complex expert systems will only be capable of being described by attributing 'mental' states to them. ---- I think this is the proliferation of jargon and verbiage that Ralph Johnson noted is associated with a large segment of AI work. What has happened is not a discovery or emulation of cognitive processes, but a break-down of certain weak programmers' abilities to describe the mechanical characteristics of their programs. They then resort to arcane languages and to attributing 'mental' characteristics to what are basically fuzzy algorithms that have been applied to poorly formalized or poorly characterized problems. Once the problems are better understood and are given a more precise formal characterization, one no longer needs "AI" techniques. - Steven Gutfreund ------------------------------ Date: 28 Nov 83 23:04:58-PST (Mon) From: pur-ee!uiucdcs!uicsl!Anonymous @ Ucb-Vax Subject: Re: Clarifying my 'AI Challange' - (nf) Article-I.D.: uiucdcs.4190 re: The Great Promises of AI Beware the promises of used car salesmen. The press has stories to sell, and so do the more extravagant people within AI. Remember that many of these people had to work hard to convince grantmakers that AI was worth their money, back in the days before practical applications of expert systems began to pay off. It is important to distinguish the promises of AI from the great fantasies that have been speculated by the media (and some AI researchers) in a fit of science fiction. AI applications will certainly be diverse and widespread (thanks no less to the VLSI people). However, I hope that none of us really believes that machines will possess human general intelligence any time soon. We banter about such stuff hoping that when ideas fly, at least some of them will be good ones. The reality is that nobody sees a clear and brightly lit path from here to super-intelligent robots. Rather we see hundreds of problems to be solved. Each solution should bring our knowledge and the capabilities of our programs incrementally forward. But let's not kid ourselves about the complexity of the problems. As it has already been pointed out, AI is tackling the hard problems -- the ones for which nobody knows any algorithms. ------------------------------ Date: Wed, 30 Nov 83 10:29 PST From: Tong.PA@PARC-MAXC.ARPA Subject: Re: AI Challenge Tom Dietterich: Your view of "knowledge representations" as being identical with data structures reveals a fundamental misunderstanding of the knowledge vs. algorithms point. . .Why, I'll bet there's not a single AI program that uses leftist-trees or binomial queues! Sanjai Narain: We at Rand have ROSS. . .One implementation of ROSS uses leftist trees for maintaining event queues. Since these queues are in the innermost loop of ROSS's operation, it was only sensible to make them as efficient as possible. We think we are doing AI. Sanjai, you take the letter but not the spirit of Tom's reflection. I don't think any AI researcher would object to improving the efficiency of her program, or using traditional computer science knowledge to help. But - look at your own description of ROSS development! Clearly you first conceptualized ROSS ("queues are the innermost loop") and THEN worried about efficiency in implementing your conceptualization ("it was only sensible to make them as efficient as possible"). Traditional computer science can shed much light on implementation issues, but has in practice been of little direct help in the conceptualization phase (except occasionally by analogy and generalization). All branches of computer science share basic interests such as how to represent and use knowledge, but AI differs in the GRAIN SIZE of the knowledge it considers. It would be very desirable to have a unified theory of computer science that provides ideas and tools along the continuum of knowledge grain size; but we are not quite there, yet. Until that time, perceiving the different branches of computer science as contributing useful knowledge to different levels of implementation (e.g. knowledge level, data level, register transfer level, hardware level) is probably the best integration our short term memories can handle. Chris Tong ------------------------------ Date: 28 Nov 83 22:25:35-PST (Mon) From: pur-ee!uiucdcs!marcel @ Ucb-Vax Subject: RJ vs AI: Science vs Engineering? - (nf) Article-I.D.: uiucdcs.4187 In response to Johnson vs AI, and Tom Dietterich's defense: The emergence of the knowledge-based perspective is only the beginning of what AI has achieved and is working on. Obvious corollaries: knowledge acquisition and extraction, representation, inference engines. Some rather impressive results have been obtained here. One with which I am most familiar is work being done at Edinburgh by the Machine Intelligence Research Unit on knowledge extraction via induction from user-supplied examples (the induction program is commercially available). A paper by Shapiro (Alen) & Niblett in Computer Chess 3 describes the beginnings of the work at MIRU. Shapiro has only this month finished his PhD, which effectively demonstrates that human experts, with the aid of such induction programs, can produce knowledge bases that surpass the capabilities of any expert as regards their completeness and consistency. Shapiro synthesized a totally correct knowledge base for part of the King-and-Pawn against King-and-Rook chess endgame, and even that relatively small endgame was so complex that, though it was treated in the chess literature, the descriptions provided by human experts consisted largely of gaps. Impressively, 3 chess novices managed (again with the induction program) to achieve 99% correctness in this normally difficult problem. The issue: even novices are better at articulating knowledge by means of examples than experts are at articulating the actual rules involved, *provided* that the induction program can represent its induced rules in a form intelligible to humans. The long-term goal and motivation for this work is the humanization of technology, namely the construction of systems that not only possess expert competence, but are capable of communicating their reasoning to humans. And we had better get this right, lest we get stuck with machines that run our nuclear plants in ways that are perhaps super-smart but incomprehensible ... until a crisis happens, when suddenly the humans need to understand what the machine has been doing until now. The problem: lack of understanding of human cognitive psychology. More specifically, how are human concepts (even for these relatively easy classification tasks) organized? What are the boundaries of 'intelligibility'? Though we are able to build systems that function, in some ways, like a human expert, we do not know much about what distinguishes brain-computable processes from general algorithms. But we are learning. In fact, I am tempted to define this as one criterion distinguishing knowledge-based AI from other computing: the absolute necessity of having our programs explain their own processing. This is close to demanding that they also process in brain-compatible terms. In any case we will need to know what the limits of our brain-machine are, and in what forms knowledge is most easily apprehensible to it. This brings our end of AI very close to cognitive psychology, and threatens to turn knowledge representation into a hard science -- not just What does a system need, to be able to X? but How does a human brain produce behavior/inference X, and how do we implement that so as preserve maximal man-machine compatibility? Hence the significance of the work by Shapiro, mentioned above: the intelligibility of his representations is crucial to the success of his knowledge-acquisition method, and the whole approach provides some clues on how a humane knowledge representation might be scientifically determined. A computer is merely a necessary weapon in this research. If AI has made little obvious progress it may be because we are too busy trying to produce useful systems before we know how they should work. In my opinion there is too little hard science in AI, but that's understandable given its roots in an engineering discipline (the applications of computers). Artificial intelligence is perhaps the only "application" of computers in which hard science (discovering how to describe the world) is possible. We might do a favor both to ourselves and to psychology if knowledge-based AI adopted this idea. Of course, that would cut down drastically on the number of papers published, because we would have some very hard criteria about what comprised a tangible contribution. Even working programs would not be inherently interesting, no matter what they achieved or how they achieved it, unless they contributed to our understanding of knowledge, its organization and its interpretation. Conversely, working programs would be necessary only to demonstrate the adequacy of the idea being argued, and it would be possible to make very solid contributions without a program (as opposed to the flood of "we are about to write this program" papers in AI). So what are we: science or engineering? If both, let's at least recognize the distinction as being valuable, and let's know what yet another expert system proves beyond its mere existence. Marcel Schoppers U of Illinois @ Urbana-Champaign ------------------------------ End of AIList Digest ******************** 4-Dec-83 23:06:49-PST,13008;000000000001 Mail-From: LAWS created at 4-Dec-83 23:06:01 Date: Sun 4 Dec 1983 22:56-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #109 To: AIList@SRI-AI AIList Digest Monday, 5 Dec 1983 Volume 1 : Issue 109 Today's Topics: Expert Systems & VLSI - Request for Material, Programming Languages - Productivity, Editorial Policy - Anonymous Messages, Bindings - Dr. William A. Woods, Intelligence, Looping Problem, Pattern Recognition - Block Modeling, Seminars - Programs as Predicates & Explainable Expert System ---------------------------------------------------------------------- Date: Sun, 4 Dec 83 17:59:53 PST From: Tulin Mangir Subject: Request for Material I am preparing a tutorial and a current bibliography, for IEEE, of the work in the area of expert system applications to CAD and computer aided testing as well as computer aided processing. Specific emphasis is on LSI/VLSI design, testing and processing. I would like this material to be as complete and as current as we can all make. So, if you have any material in these areas that you would like me to include in the notes, ideas about representation of structure, knowledge, behaviour of digital circuits, etc., references you know of, please send me a msg. Thanks. Tulin Mangir (213) 825-2692 825-4943 (secretary) ------------------------------ Date: 29 Nov 83 22:25:19-PST (Tue) From: sri-unix!decvax!duke!mcnc!marcel@uiucdcs.UUCP (marcel )@CCA Subject: Re: lisp productivity question - (nf) Article-I.D.: uiucdcs.4197 And now a plug from the logic programming people: try prolog for easy debugging. Though it may take a while to get used to its modus operandi, it has one advantage that is shared by no other language I know of: rule-based computing with a clean formalism. Not to mention the ease of implementing concepts such as "for all X satisfying P(X) do ...". The end of cumbersome array traversals and difficult boolean conditions! Well, almost. Not to mention free pattern matching. And I wager that the programs will be even shorter in Prolog, primarily because of these considerations. I have written 100-line Prolog programs which were previously coded as Pascal programs of 2000 lines. Sorry, I just couldn't resist the chance to be obnoxious. ------------------------------ Date: Fri, 2 Dec 83 09:47 EST From: MJackson.Wbst@PARC-MAXC.ARPA Subject: Lisp "productivity" "A caveat: Lisp is very well suited to the nature of game programs. A fair test would require that data processing and numerical analysis problems be included in the mix of test problems." A fair test of what? A fair test of which language yields the greatest productivity when applied to the particular mix of test problems, I would think. Clearly (deepfelt theological convictions to the contrary) there is NO MOST-PRODUCTIVE LANGUAGE. It depends on the problem set; I like structured languages so I do my scientific programming in Ratfor, and when I had to do it in Pascal it was awful, but for a different type of problem Pascal would be just fine. Mark ------------------------------ Date: 30 Nov 83 22:49:51-PST (Wed) From: pur-ee!uiucdcs!uicsl!Anonymous @ Ucb-Vax Subject: Lisp Productivity & Anonymous Messages Article-I.D.: uiucdcs.4245 The most incredible programming environment I have worked with to date is that of InterLisp. The graphics-based trace and break packages on Xerox's InterLisp-D (not to mention the Lisp editor, file package, and the programmer's assistant) is, to say the least, addictive. Ease of debugging has been combined with power to yield an environment in which program development/debugging is easy, fast and productive. I think other languages have a long way to go before someone develops comparable environments for them. Of course, part of this is due to the language (i.e., Lisp) itself, since programs written in Lisp tend to be easy to conceptualize and write, short, and readable. [I will pass this message along to the Arpanet AIList readers, but am bothered by its anonymous authorship. This is hardly an incriminating message, and I see no reason for the author to hide. I do not currently reject anonymous messages out of hand, but I will certainly screen them strictly. -- KIL] ------------------------------ Date: Thu 1 Dec 83 07:37:04-PST From: C.S./Math Library Subject: Press Release RE: Dr. William A. Woods [Reprinted from the SU-SCORE bboard.] As of September 16, Chief Scientist directing all research in AI and related technologies for Applied Expert Systems, Inc., Five Cambridge Center, Cambridge, Mass 02142 (617)492-7322 net address Woods@BBND (same as before) HL ------------------------------ Date: Fri, 2 Dec 83 09:57:14 PST From: Adolfo Di-Mare Subject: a new definition of intelligence You're intelligence is directly proportional to the time it takes you to bounce back after you're replaced by an computer. As I'm not an economist, I won't argue on how intelligent we are... Put in another way, is an expert that builds a machine that substitutes him/er intelligent? If s/he is not, is the machine? Adolfo /// ------------------------------ Date: 1 Dec 83 20:37:31-PST (Thu) From: decvax!bbncca!jsol @ Ucb-Vax Subject: Re: Halting Problem Discussion Article-I.D.: bbncca.365 Can a method be formulated for deciding whether or not your are on the right track? Yes. It's call interaction. Ask someone you feel you can trust about whether or not you are getting anywhere, and to offer any advice to help you get where you want to go. Students do it all the time, they come to their teachers and ask them to help them. Looping programs could decide that they have looped for as long as they care to and reality check them. An algorithm to do this is available if anyone wants it (read that to mean I will produce one). -- [--JSol--] JSol@Usc-Eclc/JSol@Bbncca (Arpa) JSol@Usc-Eclb/JSol@Bnl (Milnet) {decvax, wjh12, linus}!bbncca!jsol ------------------------------ From: Bibbero.PMSDMKT Reply-to: Bibbero.PMSDMKT Subject: Big Brother and Block Modeling, Warning [Reprinted from the Human-Nets Digest.] [This application of pattern recognition seems to warrant mention, but comments on the desirability of such analysis should be directed to Human-Nets@RUTGERS. -- KIL] The New York Times (Nov 20, Sunday Business Section) carries a warning from two Yale professors against a new management technique that can be misused to snoop on personnel through sophisticted mathematical analysis of communications, including computer network usage. Professors Scott Boorman, a Yale sociologist, and Paul Levitt, research mathematician at Yale and Harvard (economics) who authored the article also invented the technique some years ago. Briefly, it consists of computer-intensive analysis of personnel communications to divide them into groups or "blocks" depending on who they communicate with, whom they copy on messages, who they phone and who's calls don't they return. Blocks of people so identified can be classified as dissidents, potential traitors or "Young Turks" about to split off their own company, company loyalists, promotion candidates and so forth. "Guilt by association" is built into the system since members of the same block may not even know each other but merely copy the same person on memos. The existence of an informal organization as a powerful directing force in corporations, over and above the formal organization chart, has been recognized for a long time. The block analysis method permits and "x-ray" penetration of these informal organizations through use of computer on-line analysis which may act, per the authors, as "judge and jury." The increasing usage of electronic mail, voice storage and forward systems, local networks and the like make clandestine automation of this kind of snooping simple, powerful, and almost inevitable. The authors cite as misusage evidence the high degree of interest in the method by iron curtain government agencies. An early success (late 60's) was also demonstrated in a Catholic monastery where it averted organizational collapse by identifying members as loyalists, "Young Turks," and outcasts. Currently, interest is high in U.S. corporations, particularily the internal audit departments seeking to identify dissidents. As the authors warn, this revolution in computers and information systems bring us closer to George Orwell's state of Oceania. ------------------------------ Date: 1 Dec 1983 1629-EST From: ELIZA at MIT-XX Subject: Seminar Announcement [Reprinted from the MIT-AI bboard.] Date: Wednesday, December 7th, l983 Time: Refreshments 3:30 P.M. Seminar 3:45 P.M. Place: NE43-512A (545 Technology Square, Cambridge) PROGRAMS ARE PREDICATES C. A. R. Hoare Oxford University A program is identified with the strongest predicate which describes every observation that might be made of a mechanism which executes the program. A programming language is a set of programs expressed in a limited notation, which ensures that they are implementable with adequate efficiency, and that they enjoy desirable algebraic properties. A specification S is a predicate expressed in arbitrary mathematical notation. A program P meets this specification if P ==> S . Thus a calculus for the derivation of correct programs is an immediate corollary of the definition of the language. These theses are illustrated in the design of two simple programming languages, one for sequential programming and the other for communicating sequential processes. Host: Professor John V. Guttag ------------------------------ Date: 12/02/83 09:17:19 From: ROSIE at MIT-ML Subject: Expert Systems Seminar [Forwarded by SASW@MIT-MC.] DATE: Thursday, December 8, 1983 TIME: 2.15 p.m. Refreshments 2.30 p.m. Lecture PLACE: NE43-AI Playroom Explainable Expert Systems Bill Swartout USC/Information Sciences Institute Traditional methods for explaining programs provide explanations by converting the code of the program to English. While such methods can sometimes adequately explain program behavior, they cannot justify it. That is, such systems cannot tell why what the system is doing is reasonable. The problem is that the knowledge required to provide these justifications was used to produce the program but is itself not recorded as part of the code and hence is unavailable. This talk will first describe the XPLAIN system, a previous research effort aimed at improving the explanatory capabilities of expert systems. We will then outline the goals and research directions for the Explainable Expert Systems project, a new research effort just starting up at ISI. The XPLAIN system uses an automatic programmer to generate a consulting program by refinement from abstract goals. The automatic programmer uses two sources of knowledge: a domain model, representing descriptive facts about the application domain, and a set of domain principles, representing problem-solving knowledge, to drive the refinement process forward. As XPLAIN creates an expert system, it records the decisions it makes in a refinement structure. This structure is then used to provide explanations and justifications of the expert system. Our current research focuses on three areas. First, we want to extend the XPLAIN framework to represent additional kinds of knowledge such as control knowledge for efficient execution. Second, we want to investigate the compilation process that moves from abstract to specific knowledge. While it does seem that human experts compile their knowledge, they do not always use the resulting specific methods. This may be because the specific methods often contain compiled-in assumptions which are usually (but not always) correct. Third, we intend to use the richer framework provided by XPLAIN for enhanced knowledge acquisition. HOST: Professor Peter Szolovits ------------------------------ End of AIList Digest ******************** 6-Dec-83 20:41:40-PST,20774;000000000001 Mail-From: LAWS created at 6-Dec-83 20:38:18 Date: Tue 6 Dec 1983 20:24-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #110 To: AIList@SRI-AI AIList Digest Wednesday, 7 Dec 1983 Volume 1 : Issue 110 Today's Topics: AI and Manufacturing - Request, Bindings - HPP, Programming Languages - Environments & Productivity, Vision - Cultural Influences on Perception, AI Jargon - Mental States of Machines, AI Challange & Expert Systems, Seminar - Universal Subgoaling ---------------------------------------------------------------------- Date: 5 Dec 83 15:14:26 EST (Mon) From: Dana S. Nau Subject: AI and Automated Manufacturing I and some colleagues at University of Maryland are doing a literature search on the use of AI techniques in Automated Manufacturing. The results of the literature search will comprise a report to be sent to the National Bureau of Standards as part of a research contract. We'd appreciate any relevant information any of you may have--especially copies of papers or technical reports. In return, I can send you (on request) copies of some papers I have published on that subject, as well as a copy of the literature search when it is completed. My mailing address is Dana S. Nau Computer Science Dept. University of Maryland College Park, MD 20742 ------------------------------ Date: Mon 5 Dec 83 08:27:28-PST From: HPP Secretary Subject: New Address for HPP [Reprinted from the SU-SCORE bboard.] The HPP has moved. Our new address is: Heuristic Programming Project Computer Science Department Stanford University 701 Welch Road, Bldg. C Palo Alto, CA 94304 ------------------------------ Date: Mon, 5 Dec 83 09:43:51 PST From: Seth Goldman Subject: Programming environments are fine, but... What are all of you doing with your nifty, adequate, and/or brain-damaged computing environments? Also, if we're going to discuss environments, it would be more productive I think to give concrete examples of the form: I was trying to do or solve X Here is how my environment helped me OR This is what I need and don't yet have It would also be nice to see some issues of AIList dedicated to presenting 1 or 2 paragraph abstracts of current work being pursued by readers and contributors to this list. How about it Ken? [Sounds good to me. It would be interesting to know whether progress in AI is currentlyheld back by conceptual problems or just by the programming effort of building large and user-friendly systems. -- KIL] Seth Goldman ------------------------------ Date: Monday, 5 December 1983 13:47:13 EST From: Robert.Frederking@CMU-CS-CAD Subject: Re: marcel on "lisp productivity question" I just thought I should mention that production system languages share all the desirable features of Prolog mentioned in the previous message, particularly being "rule-based computing with a clean formalism". The main differences with the OPS family of languages is that OPS uses primarily forward inference, instead of backwards inference, and a slightly different matching mechanism. Preferring one over the other depends, I suspect, on whether you think in terms of proofs or derivations. ------------------------------ Date: Mon, 5 Dec 83 10:23:17 pst From: evans@Nosc (Evan C. Evans) Subject: Vision & Such Ken Laws in AIList Digest 1:99 states: an adequate answer [to the question of why computers can't see yet] requires a guess at how it is that the human vision system can work in all cases. I cannot answer Ken's question, but perhaps I can provide some useful input. language shapes culture (Sapir-Whorf hypothesis) culture shapes vision (see following) vision shapes language (a priori) The influence of culture on perception (vision) takes many forms. A statistical examination (unpublished) of the British newspaper game "Where's the ball?" is worth consideration. This game has been appearing for some time in British, Australian, New Zealand, & Fijian papers. So far as I know, it has not yet made its ap- pearance in U.S. papers. The game is played thus: A photograph of some common sport involving a ball is published with the ball erased from the picture & the question, where's the ball? Various members of the readership send in their guesses & that closest to the ball's actual position in the unmodified photo wins. Some time back the responses to several rounds of this game were subjected to statistical analysis. This analysis showed that there were statistically valid differences associated with the cultural background of the participants. This finding was particularly striking in Fiji with a resident population comprising several very different cultural groups. Ball placement by the different groups tended to cluster at sig- nificantly different locations in the picture, even for a game like soccer that was well known & played by all. It is unfor- tunate that this work (not mine) has not been published. It does suggest two things: a.) a cultural influence on vision & percep- tion, & b.) a powerful means of conducting experiments to learn more about this influence. For instance, this same research was elaborated into various TV displays designed to discover where children of various age groups placed an unseen object to which an arrow pointed. The children responded enthusiastically to this new TV game, giving their answers by means of a light pen. Yet statistically significant amounts of data were collected ef- ficiently & painlessly. I've constructed the loop above to suggest that none of the three: vision, language, & culture should be studied out of context. E. C. Evans III ------------------------------ Date: Sat 3 Dec 83 00:42:50-PST From: PEREIRA@SRI-AI.ARPA Subject: Mental states of machines Steven Gutfreund's criticism of John McCarthy is unjustified. I haven't read the article in "Psychology Today", but I am familiar with the notion put forward by JMC and condemned by SG. The question can be put in simple terms: is it useful to attribute mental states and attitudes to machines? The answer is that our terms for mental states and attitudes ("believe", "desire", "expect", etc...) represent a classification of possible relationships between world states and the internal (inacessible) states of designated individuals. Now, for simple individuals and worlds, for example small finite automata, it is possible to classify the world-individual relationships with simple and tractable predicates. For more complicated systems, however, the language of mental states is likely to become essential, because the classifications it provides may well be computationally tractable in ways that other classifications are not. Remember that individuals of any "intelligence" must have states that encode classifications of their own states and those of other individuals. Computational representations of the language of mental states seem to be the only means we have to construct machines with such rich sets of states that can operate in "rational" ways with respect to the world and other individuals. SG's comment is analogous to the following criticism of our use of the terms like "execution", "wait" or "active" when talking about the states of computers: "it is wrong to use such terms when we all know that what is down there is just a finite state machine, which we understand so well mathematically." Fernando Pereira ------------------------------ Date: Mon 5 Dec 83 11:21:56-PST From: Wilkins Subject: complexity of formal systems From: Steven Gutfreund They then resort to arcane languages and to attributing 'mental' characteristics to what are basically fuzzy algorithms that have been applied to poorly formalized or poorly characterized problems. Once the problems are better understood and are given a more precise formal characterization, one no longer needs "AI" techniques. I think Professor McCarthy is thinking of systems (possibly not built yet) whose complexity comes from size and not from imprecise formalization. A huge AI program has lots of knowledge, all of it may be precisely formalized in first-order logic or some other well understood formalism, this knowledge may be combined and used by well understood and precise inference algorithms, and yet because of the (for practical purposes) infinite number of inputs and possible combinations of the individual knowledge formulas, the easiest (best? only?) way to desribe the behavior of the system is by attributing mental characteristics. Some AI systems approaching this complex already exist. This has nothing to do with "fuzzy algorithms" or "poorly formalized problems", it is just the inherent complexity of the system. If you think you can usefully explain the practical behavior of any well-formalized system without using mental characteristics, I submit that you haven't tried it on a large enough system (e.g. some systems today need a larger address space than that available on a DEC 2060 -- combining that much knowledge can produce quite complex behavior). ------------------------------ Date: 28 Nov 83 3:10:20-PST (Mon) From: harpo!floyd!clyde!akgua!sb1!sb6!bpa!burdvax!sjuvax!rbanerji@Ucb- Vax Subject: Re: Clarifying my "AI Challange" Article-I.D.: sjuvax.157 [...] I am reacting to Johnson, Helly and Dietterich. I really liked [Ken Laws'] technical evaluation of Knowledge-based programming. Basically similar to what Tom also said in defense of Knowledge-based programming but KIL said it much clearer. On one aspect, I have to agree with Johnson about expert systems and hackery, though. The only place there is any attempt on the part of an author to explain the structure of the knowledge base(s) is in the handbook. But I bet that as the structures are changed by later authors for various justified and unjustified reasons, they will not be clearly explained except in vague terms. I do not accept Dietterich's explanation that AI papers are hard to read because of terminology; or because what they are trying to do are so hard. On the latter point, we do not expect that what they are DOING be easy, just that HOW they are doing it be clearly explained: and that the definition of clarity follow the lines set out in classical scientific disciplines. I hope that the days are gone when AI was considered some sort of superscience answerable to none. On the matter of terminology, papers (for example) on algebraic topology have more terminology than AI: terminology developed over a longer period of time. But if one wants to and has the time, he can go back, back, back along lines of reference and to textbooks and be assured he will have an answer. In AI, about the only hope is to talk to the author and unravel his answers carefully and patiently and hope that somewhere along the line one does not get "well, there is a hack there..it is kind of long and hard to explain: let me show you the overall effect" In other sciences, hard things are explained on the basis of previously explained things. These explanantion trees are much deeper than in AI; they are so strong and precise that climbing them may be hard, but never hopeless. I agree with Helly in that this lack is due to the fact that no attempt has been made in AI to have workers start with a common basis in science, or even in scientific methodology. It has suffered in the past because of this. When existing methods of data representation and processing in theorem proving was found inefficient, the AI culture developed this self image that its needs were ahead of logic: notwithstanding the fact that the techniques they were using were representable in logic and that the reason for their seeming success was in the fact that they were designed to achieve efficiency at the cost (often high) of flexibility. Since then, those words have been "eaten": but at considerable cost. The reason may well be that the critics of logic did not know enough logic to see this. In some cases, their professors did--but never cared to explain what the real difficulty in logic was. Or maybe they believed their own propaganda. This lack of uniformity of background came out clear when Tom said that because of AI work people now clearly understood the difference between the subset of a set and the element of a set. This difference has been well known at least since early this century if not earlier. If workers in AI did not know it before, it is because of their reluctance to know the meaning of a term before they use it. This has also often come from their belief that precise definitions will rob their terms of their richness (not realising that once they have interpreted their terms by a program, they have a precise definition, only written in a much less comprehensible way: set theorists never had any difficulty understanding the diffeence between subsets and elements). If they were trained, they would know the techniques that are used in Science for defining terms. I disagree with Helly that Computer Science in general is unscientific. There has always been a precise mathematical basis of Theorem proving (AI, actually) and in computation and complexity theory. It is true, however, that the traditional techniques of experimental research have not been used in AI at all: people have tried hard to use it in software, but seem to be having difficulties. Would Helly disagree with me if I say that Newell and Simon's work in computer modelling of psychological processes have been carried out with at least the amount of scientific discipline that psychologists use? I have always seen that work as one of the success stories in AI. And at least some psychologists seem to agree. I agree with Tom that AI will have to keep going even if someone proves that P=NP. The reason is that many AI problems are amenable to N^2 methods already: except that N is too big. In this connection I have a question, in case someone can tell me. I think Rabin has a theorem that given any system of logic and any computable function, there is a true statement which takes longer to prove than that function predicts. What does this say about the relation between P and NP, if anything? Too long already! ..allegra!astrovax!sjuvax!rbanerji ------------------------------ Date: 1 Dec 83 13:51:36-PST (Thu) From: decvax!duke!mcnc!ncsu!fostel @ Ucb-Vax Subject: RE: Expert Systems Article-I.D.: ncsu.2420 Are expert systems new? Different? Well, how about an example. Time was, to run a computer system, one needed at least one operator to care and feed for the system. This is increasingly handled by sophisticated operating systems. As such is an operating system an "expert system"? An OS is usually developed using a style of programming which is quite different from those of wimpy, unskilled, un-enlightenned applications programmers. It would be very hard to build an operating system in the applications style. (I claim). The people who developed the style and practice it to build systems are not usually AI people although I would wager the presonality profiles would be quite similar. Now, that is I think a major point. Are there different type of people in Physics as compared to Biology? I would say so, having seen some of each. Further, biologists do research in ways that seem different (again, this is purely idiosynchratic evidence) differently than physists. Is it that one group know how to do science better, or are the fields just so differnt, or are the people attracted to each just different? Now, suppose a team of people got together and built an expert system which was fully capable of taking over the control of a very sophisticated (previously manual, by highly trained people) inventory, billing and ordering system. I claim that this is at least as complex as diagnosis of and dosing of particular drugs (e.g. mycin). My expert system was likely written in Cobol by people doing things in quite different ways from AI or systems hackers. One might want to argue that the productivity was much lower, that the result was harder to change and so on. I would prefer to see this in Figures, on proper comparisons. I suspect that the complexity of the commercial software I mentioned is MUCH greater than the usual problem attacked by AI people, so that the "productivity" might be comparable, with the extra time reflecting the complexity. For example, designing the reports and generating them for a large complex system (and doing a good job) may take a large fraction of the total time, yet such reporting is not usually done in the AI world. Traces of decisions and other discourse are not the same. The latter is easier I think, or at least it takes less work. What I'm getting at is that expert systems have been around for a long time, its only that recently AI people have gotten in to the arena. There are other techniques which have been applied to developing these, and I am waiting to be convinced that the AI people have a priori superior strategies. I would like to be so convinced and I expect someday to be convinced, but then again, I probably also fit the AI personality profile so I am rather biased. ----GaryFostel---- ------------------------------ Date: 5 Dec 1983 11:11:52-EST From: John.Laird at CMU-CS-ZOG Subject: Thesis Defense [Reprinted from the CMU-AI bboard.] Come see my thesis defense: Wednesday, December 7 at 3:30pm in 5409 Wean Hall UNIVERSAL SUBGOALING ABSTRACT A major aim of Artificial Intelligence (AI) is to create systems that display general problem solving ability. When problem solving, knowledge is used to avoid uncertainty over what to do next, or to handle the difficulties that arises when uncertainity can not be avoided. Uncertainty is handled in AI problem solvers through the use of methods and subgoals; where a method specifies the behavior for avoiding uncertainity in pursuit of a goal, and a subgoal allows the system to recover from a difficulty once it arises. A general problem solver should be able to respond to every task with appropriate methods to avoid uncertainty, and when difficulties do arise, the problem solver should be able to recover by using an appropriate subgoal. However, current AI problem solver are limited in their generality because they depend on sets of fixed methods and subgoals. In previous work, we investigated the weak methods and proposed that a problem solver does not explicitly select a method for goal, with the inherent risk of selecting an inappropriate method. Instead, the problem solver is organized so that the appropriate weak method emerges during problem solving from its knowledge of the task. We called this organization a universal weak method and we demonstrated it within an architecture, called SOAR. However, we were limited to subgoal-free weak methods. The purpose of this thesis is to a develop a problem solver where subgoals arise whenever the problem solver encounters a difficulty in performing the functions of problem solving. We call this capability universal subgoaling. In this talk, I will describe and demonstrate an implementation of universal subgoaling within SOAR2, a production system based on search in a problem space. Since SOAR2 includes both universal subgoaling and a universal weak method, it is not limited by a fixed set of subgoals or methods. We provide two demonstrations of this: (1) SOAR2 creates subgoals whenever difficulties arise during problem solving, (2) SOAR2 extends the set of weak methods that emerge from the structure of a task without explicit selection. ------------------------------ End of AIList Digest ******************** 10-Dec-83 15:21:29-PST,16132;000000000001 Mail-From: LAWS created at 10-Dec-83 15:15:30 Date: Sat 10 Dec 1983 14:46-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #111 To: AIList@SRI-AI AIList Digest Saturday, 10 Dec 1983 Volume 1 : Issue 111 Today's Topics: Call for Papers - Special Issue of AJCL, Linguistics - Phrasal Analysis Paper, Intelligence - Purpose of Definition, Expert Systems - Complexity, Environments - Need for Sharable Software, Jargon - Mental States, Administrivia - Spinoff Suggestion, Knowledge Representation - Request for Discussion ---------------------------------------------------------------------- Date: Thu 8 Dec 83 08:55:34-PST From: Ray Perrault Subject: Special Issue of AJCL American Journal of Computational Linguistics The American Journal of Computational Linguistics is planning a special issue devoted to the Mathematical Properties of Linguistic Theories. Papers are hereby requested on the generative capacity of various syntactic formalisms as well as the computational complexity of their related recognition and parsing algorithms. Articles on the significance (and the conditions for the significance) of such results are also welcome. All papers will be subjected to the normal refereeing process and must be accepted by the Editor-in-Chief, James Allen. In order to allow for publication in Fall 1984, five copies of each paper should be sent by March 31, 1984 to the special issue editor, C. Raymond Perrault Arpanet: Rperrault@sri-ai SRI International Telephone: (415) 859-6470 EK268 Menlo Park, CA 94025. Indication of intention to submit would also be appreciated. ------------------------------ Date: 8 Dec 1983 1347-PST From: MEYERS.UCI-20A@Rand-Relay Subject: phrasal analysis paper Over a month ago, I announced that I'd be submitting a paper on phrasal analysis to COLING. I apologize to all those who asked for a copy for not getting it to them yet. COLING acceptance date is April 2, so this may be the earliest date at which I'll be releasing papers. Please do not lose heart! Some preview of the material might interest AILIST readers: The paper is entitled "Conceptual Grammar", and discusses a grammar that uses syntactic and 'semantic' nonterminals. Very specific and very general information about language can be represented in the grammar rules. The grammar is organized into explicit levels of abstraction. The emphasis of the work is pragmatic, but I believe it represents a new and useful approach to Linguistics as well. Conceptual Grammar can be viewed as a systematization of the knowledge base of systems such as PHRAN (Wilensky and Arens, at UC Berkeley). Another motivation for a conceptual grammar is the lack of progress in language understanding using syntax-based approaches. A third motivation is the lack of intuitive appeal of existing grammars -- existing grammars offer no help in manipulating concepts the way humans might. Conceptual Grammar is an 'open' grammar at all levels of abstraction. It is meant to handle special cases, exceptions to general rules, idioms, etc. Papers on the implemented system, called VOX, will follow in the near future. VOX analyzes messages in the Navy domain. (However, the approach to English is completely general). If anyone is interested, I can elaborate, though it is hard to discuss such work in this forum. Requests for papers (and for abstracts of UCI AI Project papers) can be sent by computer mail, or 'snail-mail' to: Amnon Meyers AI Project Department of Computer Science University of California Irvine, CA 92717 PS: A paper has already been sent to CSCSI. The papers emphasize different aspects of Conceptual Grammar. A paper on VOX as an implementation of Conceptual Grammar is planned for AAAI. ------------------------------ Date: 2 Dec 83 7:57:46-PST (Fri) From: ihnp4!houxm!hou2g!stekas @ Ucb-Vax Subject: Re: Rational Psych (and science) Article-I.D.: hou2g.121 It is true that psychology is not a "science" in the way a physicist defines "science". Of course, a physicist would be likely to bend his definition of "science" to exclude psychology. The situation is very much the same as defining "intelligence". Social "scientists" keep tightening their definition of intelligence as required to exclude anything which isn't a human being. While AI people now argue over what intelligence is, when an artificial system is built with the mental ability of a mouse (the biological variety!) in no time all definitions of intelligence will be bent to include it. The real significance of a definition is that it clarifies the *direction* in which things are headed. Defining "intelligence" in terms of adaptability and self-consciousness are evidence of a healthy direction to AI. Jim ------------------------------ Date: Fri 9 Dec 83 16:08:53-PST From: Peter Karp Subject: Biologists, physicists, and report generating programs I'd like to ask Mr. Fostel how biologists "do research in ways that seem different than physicists". It would be pretty exciting to find that one or both of these two groups do science in a way that is not part of standard scientific method. He also makes the following claim: ... the complexity of the commercial software I mentionned is MUCH greater than the usual problem attacked by AI people... With the example that: ... designing the reports and generating them for a large complex system (and doing a good job) may take a large fraction of the total time, yet such reporting is not usually done in the AI world. This claim is rather absurd. While I will not claim that deciding on the best way to present a large amount of data is a trivial task, the point is that report generating programs have no knowledge about data presentation strategies. People who do have such knowledge spend hours and hours deciding on a good scheme and then HARD CODING such a scheme into a program. Surely one would not claim that a program consisting soley of a set of WRITELN (or insert your favorite output keyword) statements has any complexity at all, much less intelligence or knowledge? Just because a program takes a long time to write doesn't mean it has any complexity, in terms of control structures or data structures. And in fact this example is a perfect proof of this conjecture. ------------------------------ Date: 2 Dec 83 15:27:43-PST (Fri) From: sri-unix!hplabs!hpda!fortune!amd70!decwrl!decvax!duke!mcnc!shebs @utah-cs.UUCP (Stanley Shebs) Subject: Re: RE: Expert Systems Article-I.D.: utah-cs.2279 A large data-processing application is not an expert system because it cannot explain its action, nor is the knowledge represented in an adequate fashion. A "true" expert system would *not* consist of algorithms as such. It would consist of facts and heuristics organized in a fashion to permit some (relatively uninteresting) algorithmic interpreter to generate interesting and useful behavior. Production systems are a good example. The interpreter is fixed - it just selects rules and fires them. The expert system itself is a collection of rules, each of which represents a small piece of knowledge about the domain. This is of course an idealization - many "expert systems" have a large procedural component. Sometimes the existence of that component can even be justified... stan shebs utah-cs!shebs ------------------------------ Date: Wed, 7 Dec 1983 05:39 EST From: LEVITT%MIT-OZ@MIT-MC.ARPA Subject: What makes AI crawl From: Seth Goldman Subject: Programming environments are fine, but... What are all of you doing with your nifty, adequate, and/or brain-damaged computing environments? Also, if we're going to discuss environments, it would be more productive I think to give concrete examples... [Sounds good to me. It would be interesting to know whether progress in AI is currently held back by conceptual problems or just by the programming effort of building large and user-friendly systems. -- KIL] It's clear to me that, despite a relative paucity of new "conceptual" AI ideas, AI is being held back entirely by the latter "programming effort" problem, AND by the failure of senior AI researchers to recognize this and address it directly. The problem is regressive since programming problems are SO hard, the senior faculty typically give up programming altogether and lose touch with the problems. Nobody seems to realize how close we would be to practical AI, if just a handful of the important systems of the past were maintained and extended, and if the most powerful techniques were routinely applied to new applications - if an engineered system with an ongoing, expanding knowledge base were developed. Students looking for theses and "turf" are reluctant to engineer anything familiar-looking. But there's every indication that the proven techniques of the 60's/early 70's could become the core of a very smart system with lots of overlapping knowledge in very different subjects, opening up much more interesting research areas - IF the whole thing didn't have to be (re)programmed from scratch. AI is easy now, showing clear signs of diminishing returns, CS/software engineering are hard. I have been developing systems for the kinds of analogy problems music improvisors and listeners solve when they use "common sense" descriptions of what they do/hear, and of learning by ear. I have needed basic automatic constraint satisfaction systems (Sutherland'63), extensions for dependency-directed backtracking (Sussman'77), and example comparison/extension algorithms (Winston'71), to name a few. I had to implement everything myself. When I arrived at MIT AI there were at least 3 OTHER AI STUDENTS working on similar constraint propagator/backtrackers, each sweating out his version for a thesis critical path, resulting in a draft system too poorly engineered and documented for any of the other students to use. It was idiotic. In a sense we wasted most of our programming time, and would have been better off ruminating about unfamiliar theories like some of the faculty. Theories are easy (for me, anyway). Software engineering is hard. If each of the 3 ancient discoveries above was an available module, AI researchers could have theories AND working programs, a fine show. ------------------------------ Date: Thu, 8 Dec 83 11:56 EST From: Steven Gutfreund Subject: re: mental states of machines I have no problem with using anthropomorphic (or "mental") descriptions of systems as a heuristic for dealing with difficult problems. One such trick I especially approve of is Seymour Papert's "body syntonicity" technique. The basic idea is to get young children to understand the interaction of mathematical concepts by getting them to enter into a turtle world and become an active participant in it, and to use this perspective for understanding the construction of geometric structures. What I am objecting to is that I sense that John McCarthy is implying something more in his article: that human mental states are no different than the very complex systems that we sometimes use mental descriptions as a shorthand to describe. I would refer to Ilya Prigogine's 1976 Nobel Prize winning work in chemistry on "Dissapative Structures" to illustrate the foolishness of McCarthy's claim. Dissapative structures can be explained to some extent to non-chemists by means of the termite analogy. Termites construct large rich and complex domiciles. These structures sometimes are six feet tall and are filled with complex arches and domed structures (it took human architects many thousands of years to come up with these concepts). Yet if one watches termites at the lowest "mechanistic" level (one termite at a time), all one sees is a termite randomly placing drops of sticky wood pulp in random spots. What Prigogine noted was that there are parallels in chemistry. Where random underlying processes spontaneously give rise to complex and rich ordered structures at higher levels. If I accept McCarthy's argument that complex systems based on finite state automata exhibit mental characteristics, then I must also hold that termite colonies have mental characteristics, Douglas Hofstadter's Aunt Hillary also has mental characteristics, and that certain colloidal suspensions and amorphous crystals have mental characteristics. - Steven Gutfreund Gutfreund.umass@csnet-relay [I, for one, have no difficulty with assigning mental "characteristics" to inanimate systems. If a computer can be "intelligent", and thus presumably have mental characteristics, why not other artificial systems? I admit that this is Humpty-Dumpty semantics, but the important point to me is the overall I/O behavior of the system. If that behavior depends on a set of (discrete or continuous) internal states, I am just as happy calling them "mental" states as calling them anything else. To reserve the term mental for beings having volition, or souls, or intelligence, or neurons, or any other intuitive characteristic seems just as arbitrary to me. I presume that "mental" is intended to contrast with "physical", but I side with those seeing a physical basis to all mental phenomena. Philosophers worry over the distinction, but all that matters to me is the behavior of the system when I interface with it. -- KIL] ------------------------------ Date: 5 Dec 83 12:08:31-PST (Mon) From: harpo!eagle!mhuxl!mhuxm!pyuxi!pyuxnn!pyuxmm!cbdkc1!cbosgd!osu-db s!lum @ Ucb-Vax Subject: Re: defining AI, AI research methodology, jargon in AI Article-I.D.: osu-dbs.426 Perhaps Dyer is right. Perhaps it would be a good thing to split net.ai/AIList into two groups, net.ai and net.ai.d, ala net.jokes and net.jokes.d. In one the AI researchers could discuss actual AI problems, and in the other, philo- sophers could discuss the social ramifications of AI, etc. Take your pick. Lum Johnson (cbosgd!osu-dbs!lum) ------------------------------ Date: 7 Dec 83 8:27:08-PST (Wed) From: decvax!tektronix!tekcad!franka @ Ucb-Vax Subject: New Topic (technical) - (nf) Article-I.D.: tekcad.155 OK, some of you have expressed a dislike for "non-technical, philo- sophical, etc." discussions on this newsgroup. So for those of you who are tired of this, I pose a technical question for you to talk about: What is your favorite method of representing knowlege in a KBS? Do you depend on frames, atoms of data jumbled together randomly, or something in between? Do you have any packages (for public consumption which run on machines that most of us have access to) that aid people in setting up knowlege bases? I think that this should keep this newsgroup talking at least partially technically for a while. No need to thank me. I just view it as a public ser- vice. From the truly menacing, /- -\ but usually underestimated, <-> Frank Adrian (tektronix!tekcad!franka) ------------------------------ End of AIList Digest ******************** 14-Dec-83 10:18:34-PST,18110;000000000001 Mail-From: LAWS created at 14-Dec-83 10:17:18 Date: Wed 14 Dec 1983 10:03-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #112 To: AIList@SRI-AI AIList Digest Wednesday, 14 Dec 1983 Volume 1 : Issue 112 Today's Topics: Memorial Fund - Carl Engelman, Programming Languages - Lisp Productivity, Expert Systems - System Size, Scientific Method - Information Sciences, Jargon - Mental States, Perception - Culture and Vision, Natural Language - Flame ---------------------------------------------------------------------- Date: Fri 9 Dec 83 12:58:53-PST From: Don Walker Subject: Carl Engelman Memorial Fund CARL ENGELMAN MEMORIAL FUND Carl Engelman, one of the pioneers in artificial intelligence research, died of a heart attack at his home in Cambridge, Massachusetts, on November 26, 1983. He was the creator of MATHLAB, a program developed in the 1960s for the symbolic manipulation of mathematical expressions. His objective there was to supply the scientist with an interactive computational aid of a "more intimate and liberating nature" than anything available before. Many of the ideas generated in the development of MATHLAB have influenced the architecture of other systems for symbolic and algebraic manipulation. Carl graduated from the City College of New York and then earned an MS Degree in Mathematics at the Massachusetts Institute of Technology. During most of his professional career, he worked at The MITRE Corporation in Bedford, Massachusetts. In 1973 he was on leave as a visiting professor at the Institute of Information Science of the University of Turin, under a grant from the Italian National Research Council. At the time of his death Carl was an Associate Department Head at MITRE, responsible for a number of research projects in artificial intelligence. His best known recent work was KNOBS, a knowledge-based system for interactive planning that was one of the first expert systems applied productively to military problems. Originally developed for the Air Force, KNOBS was then adapted for a Navy system and is currently being used in two NASA applications. Other activities under his direction included research on natural language understanding and automatic programming. Carl published a number of papers in journals and books and gave presentations at many conferences. But he also illuminated every meeting he attended with his incisive analysis and his keen wit. While he will be remembered for his contributions to artificial intelligence, those who knew him personally will deeply miss his warmth and humor, which he generously shared with so many of us. Carl was particularly helpful to people who had professional problems or faced career choices; his paternal support, personal sponsorship, and private intervention made significant differences for many of his colleagues. Carl was a member of the American Association for Artificial Intelligence, the American Institute of Aeronautics and Astronautics, the American Mathematical Society, the Association for Computational Linguistics, and the Association for Computing Machinery and its Special Interest Group on Artificial Intelligence. Contributions to the "Carl Engelman Memorial Fund" should be sent to Judy Clapp at The MITRE Corporation, Bedford, Massachusetts 01730. A decision will be made later on how those funds will be used. ------------------------------ Date: Tue, 13 Dec 83 09:49 PST From: Kandt.pasa@PARC-MAXC.ARPA Subject: re: lisp productivity question Jonathan Slocum (University of Texas at Austin) has a large natural language translation program (thousands of lines of Interlisp) that was originally in Fortran. The compression that he got was 16.7:1. Also, I once wrote a primitive production rule system in both Pascal and Maclisp. The Pascal version was over 2000 lines of code and the Lisp version was about 200 or so. The Pascal version also was not as powerful as the Lisp version because of Pascal's strong data typing and dynamic allocation scheme. -- Kirk ------------------------------ Date: 9 Dec 83 19:30:46-PST (Fri) From: decvax!cca!ima!inmet!bhyde @ Ucb-Vax Subject: Re: RE: Expert Systems - (nf) Article-I.D.: inmet.578 I would like to add to Gary's comments. There are also issues of scale to be considered. Many of the systems built outside of AI are orders of magnitude larger. I was amazed to read that at one point the largest OPS production system, a computer game called Haunt, had so very few rules in it. A compiler written using a rule based approach would have 100 times as many rules. How big are the AI systems that folks actually build? The engineering component of large systems obscures the architectural issues involved in their construction. I have heard it said that AI isn't a field, it is a stage of the problem solving process. It seems telling that the ARPA 5-year speech recognition project was successful not with Hearsay ( I gather that after it was too late it did manage to met the performance requirements ), but by Harpy. Now, Harpy as very much like a signal processing program. The "beam search" mechanisms it used are very different than the popular approachs of the AI comunity. In the end it seems that it was an act of engineering, little insight into the nature of knowledge gained. The issues that caused AI and the rest of computing to split a few decades ago seem almost quaint now. Allan Newell has a pleasing paper about these. Only the importance of an interpreter based program development enviroment seem to continue. Can you buy a work station capable of sharing files with your 360 yet? [...] ben hyde ------------------------------ Date: 10 Dec 83 16:33:59-PST (Sat) From: decvax!ittvax!dcdwest!sdcsvax!davidson @ Ucb-Vax Subject: Information sciences vs. physical sciences Article-I.D.: sdcsvax.84 I am responding to an article claiming that psychology and computer science aren't sciences. I think that the author is seriously confused by his prefered usage of the term ``science''. The sciences based on mathematics, information processing, etc., which I will here call information sciences, e.g., linguistics, computer science, information science, cognitive science, psychology, operations research, etc., have very different methods of operation from sciences based upon, for example, physics. Since people often view physics as the prototypical science, they become confused when they look at information sciences. This is analogous to the confusion of the early grammarians who tried to understand English from a background in Latin: They decided that English was primitive and in need of fixing, and proceeded to create Grammar schools in which we were all supposed to learn how to speak our native language properly (i.e., with intrusions of latin grammar). If someone wants to have a private definition of the word science to include only some methods of operation, that's their privilege, as long as they don't want to try to use words to communicate with other human beings. But we shouldn't waste too much time definining terms, when we could be exploring the nature and utility of the methodologies used in the various disciplines. In that light, let me say something about the methodologies of two of the disciplines as I understand and practice them, respectively. Physics: There is here the assumption of a simple underlying reality, which we want to discover through elegant theorizing and experimenting. Compared to other disciplines, e.g., experimental psychology, many of the experimental tools are crude, e.g., the statistics used. A theoretical psychologist would probably find the distance that often separates physical theory from experiment to be enormous. This is perfectly alright, given the (assumed) simple nature of underlying reality. Computer Science: Although in any mathematically based science one might say that one is discovering knowledge; in many ways, it makes better sense in computer science to say that one is creating as much as discovering. Someone will invent a new language, a new architecture, or a new algorithm, and people will abandon older languages, architectures and algorithms. A physicist would find this strange, because these objects are no less valid for having been surpassed (the way an outdated physical theory would be), but are simply no longer interesting. Let me stop here, and solicit some input from people involved in other disciplines. What are your methods of investigation? Are you interested in creating theories about reality, or creating artificial or abstract realities? What is your basis for calling your discipline a science, or do you? Please do not waste any time saying that some other discipline is not a science because it doesn't do things the way yours does! -Greg ------------------------------ Date: Sun, 11 Dec 83 20:43 EST From: Steven Gutfreund Subject: re: mental states Ken Laws in his little editorializing comment on my last note seems to have completely missed the point. Whether FSA's can display mental states is an argument I leave to others on this list. However, John McCarthy's definition allows ant hills and colloidal suspensions to have mental states. ------------------------------ Date: Sun, 11 Dec 1983 15:04:10 EST From: AXLER.Upenn-1100@Rand-Relay (David M. Axler - MSCF Applications Mgr.) Subject: Culture and Vision Several people have recently been bringing up the question of the effects of culture on visual perception. This problem has been around in anthropology, folkloristics, and (to some extent) in sociolinguistics for a number of years. I've personally taken a number of graduate courses that focussed on this very topic. Individuals interested in this problem (or, more precisely, group of problems) should look into the Society for the Anthropology of Visual Communication (SAVICOM) and its journal. You'll find that the terminology is often unfamiliar, but the concerns are similar. The society is based at the University of Pennsylvania's Annenberg School of Communications, and is formally linked with such relevant groups as the American Anthro- pological Assn. Folks who want more info, citations, etc. on this can also contact me personally by netmail, as I'm not sure that this is sufficiently relevant to take up too much of AI's space. Dave Axler (Axler.Upenn-1100@Rand-Relay) [Extract from further correspondence with Dave:] There is a thing called "Visual Anthropology", on the other hand, which deals with the ways that visual tools such as film, video, still photography, etc., can be used by the anthropologist. The SAVICOM journal occasionally has articles dealing with the "meta" aspects of visual anthropology, causing it, at such times, to be dealing with the anthropology of visual anthropology (or, at least, the epistemology thereof...) --Dave Axler ------------------------------ Date: Mon 12 Dec 83 21:16:43-PST From: Martin Giles Subject: A humanities view of computers and natural language The following is a copy of an article on the Stanford Campus report, 7th December, 1983, in response to an article describing research at Stanford. The University has just received a $21 million grant for research in the fields of natural and computer languages. Martin [I have extracted a few relevant paragraphs from the following 13K-char flame. Anyone wanting the full text can contact AIList-Request or FTP it from COHN.TXT on SRI-AI. I will deleted it after a few weeks. -- KIL] Mail-From: J.JACKSON1 created at 10-Dec-83 10:29:54 Date: Sat 10 Dec 83 10:29:54-PST From: Charlie Jackson Subject: F; (Gunning Fog Index 20.18); Cohn on Computer Language Study To: bboard@LOTS-A Following is a letter found in this week's Campus Report that proves Humanities profs make as good flames as any CS hacker. Charlie THE NATURE OF LANGUAGE IS ALREADY KNOWN WITHOUT COMPUTERS Following is a response from Robert Greer Cohn, professor of French, to the Nov. 30 Campus Report article on the study of computer and natural language. The ambitious program to investigate the nature of language in connection with computers raises some far-reaching questions. If it is to be a sort of Manhattan project, to outdo the Japanese in developing machines that "think" and "communicate" in a sophisticated way, that is one thing, and one may question how far a university should turn itself towards such practical, essentially engineering, matters. If on the other hand, they are serious about delving into the nature of languages for the sake of disinterested truth, that is another pair of shoes. Concerning the latter direction: no committee ever instituted has made the kind of breakthrough individual genius alone can accomplish. [...] Do they want to know the nature of language? It is already known. The great breakthrough cam with Stephane Mallarme, who as Edmund Wilson (and later Hugh Kenner) observed, was comparable only to Einstein for revolutionary impact. He is responsible more than anyone, even Nietzsche, for the 20th-century /episteme/, as most French first-rank intellectuals agree (for example, Foucault, in "Les mots et les choses"; Sartre, in his preface to the "Poesies"' Roland Barthes who said in his "Interview with Stephen Hearth," "All we do is repeat Mallarme"; Jakobson; Derrida; countless others). In his "Notes" Mallarme saw the essence of language as "fiction," which is to say it is based on paradox. In the terms of Darwin, who describes it as "half art, half instinct," this means that language, as related to all other reality (hypothetically nonlinguistic, experimental) is "metaphorical" -- as we now say after Jakobson -- i.e. above and below the horizontal line of on-going, spontaneous, comparatively undammmed, life-flow or experience; later, as the medium of whatever level of creativity, it bears this relation to the conventional and rational real, sanity, sobriety, and so on. In this sense Chomsky's view of language as innate and determined is a half-truth and not very inspired. He would have been better off if he had read and pondered, for example, Pascal, who three centuries ago knew that "nature is itself only a first 'custom'"; or Shakespeare: "The art itself is nature" (The Winter's Tale). [...] But we can't go into all the aspects of language here. In terms of the project: since, on balance, it is unlikely the effects will go the way of elite French thought on the subject, there remains the probability that they will try to recast language, which is at its best creatively free (as well as determined at its best by organic totality, which gives it its ultimate meaning, coherence, harmony), into the narrow mold of the computer, even at /its/ best. [...] COMPUTERS AND NEWSPEAK In other words, there is no way to make a machine speak anything other than newspeak, the language of /1984/. They may overcome that flat dead robotic tone that our children enjoy -- by contrast, it gives them the feeling that they are in command of life -- but the thought and the style will be sprirtually inert. In that sense, the machines, or the new language theories, will reflect their makers, who, in harnessing themselves to a prefabricated goal, a program backed by a mental arms race, will have been coopted and dehumanized. That flat (inner or outer) tone is a direct result of cleaving to one-dimensionality, to the dimension of the linear and "metonymic," the dimension of objectivity, of technology and science, uninformed and uninspired by the creatively free and whole-reflecting ("naive") vertical, or vibrant life itself. That unidimensionality is visible in the immature personalities of the zealots who push these programs: they are not much beyond children in their Frankenstein eagerness to command the frightening forces of the psyche, including sexuality, but more profoundly, life itself, in its "existential" plenitude involving death. People like that have their uses and can, with exemplary "tunnel vision," get certain jobs done (like boring tunnels through miles of rock). A group of them can come up with /engineering/ breakthroughs in that sense, as in the case of the Manhattan project. But even that follows the /creative/ breakthroughs of the Oppenheimers and Tellers and Robert D. (the shepherd in France) and is rather pedestrian endeavor under the management of some colonel. When I tried to engage a leader of the project in discussion about the nature of language, he refused, saying, "The humanities and sciences are father apart than ever," clearly welcoming this development. This is not only deplorable in itself; far worse, according to the most accomplished mind on /their/ side of the fence in this area; this man's widely-hailed thinking is doomed to a dead end, because of its "unidimensionality!" This is not the place to go into the whole saddening bent of our times and the connection with totalitarianism, which is "integrated systems" with a vengeance. But I doubt that this is what our founders had in mind. ------------------------------ End of AIList Digest ******************** 16-Dec-83 10:09:38-PST,15021;000000000001 Mail-From: LAWS created at 16-Dec-83 10:07:31 Date: Fri 16 Dec 1983 10:02-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #113 To: AIList@SRI-AI AIList Digest Friday, 16 Dec 1983 Volume 1 : Issue 113 Today's Topics: Alert - Temporal Representation & Fuzzy Reasoning Programming Languages - Phrasal Analysis Paper, Fifth Generation - Japanese and U.S. Views, Seminars - Design Verification & Fault Diagnosis ---------------------------------------------------------------------- Date: Wed 14 Dec 83 11:21:47-PST From: Ken Laws Subject: CACM Alert - Temporal Representation & Fuzzy Reasoning Two articles in the Nov. issue of CACM (just arrived) may be of special interest to AI researchers: "Maintaining Knowledge about Temporal Intervals," by James F. Allen of the U. of Rochester, is about representation of temporal information using only intervals -- no points. While this work does not lead to a fully general temporal calculus, it goes well beyond state space and date line systems and is more powerful and efficient than event chaining representations. I can imagine that the approach could be generalized to higher dimensions, e.g., for reasoning about the relationships of image regions or objects in the 3-D world. "Extended Boolean Information Retrieval," by Gerald Salton, Edward A. Fox, and Harry Wu, presents a fuzzy logic or hierarchical inference method for dealing with uncertainties when evaluating logical formulas. In a formula such as ((A and B) or (B and C)), they present evidential combining formulas that allow for: * Uncertainty in the truth, reliability, or applicability of the the primitive terms A and B; * Differing importance of establishing the primitive term instances (where the two B terms above could be weighted differently); * Differing semantics of the logical connectives (where the two "and" connectives above could be threshold units with different thresholds). The output of their formula evaluator is a numerical score. They use this for ranking the pertinence of literature citations to a database query, but it could also be used for evidential reasoning or for evaluating possible worlds in a planning system. For the database query system, they indicate a method for determining term weights automatically from an inverted index of the database. The weighting of the Boolean connectives is based on the infinite set of Lp vector norms. The connectives and[INF] and or[INF] are the ones of standard logic; and[1] and or[1] are equivalent and reduce formula evaluation to a simple weighted summation; intermediate connective norms correspond to "mostly" gates or weighted neural logic models. The authors present both graphical illustrations and logical theorems about these connectives. -- Ken Laws ------------------------------ Date: 14 Dec 83 20:05:25-PST (Wed) From: hplabs!hpda!fortune!phipps @ Ucb-Vax Subject: Re: Phrasal Analysis Paper/Programming Languages Applications ? Article-I.D.: fortune.1981 Am I way off base, or does this look as if the VOX project would be of interest to programming languages (PL) researchers ? It might be interesting to submit to the next "Principles of Programming Languages" (POPL) conference, too. As people turn from traditional programming languages (is Ada really pointing the way of the future ? ) to other ways (query languages and outright natural language processing) to obtain and manipulate information and codified knowledge, I believe that AI and PL people will find more overlap in their ends, although probably not their respective interests, approaches, and style. This institutionalized mutual ignorance doesn't benefit either field. One of these days, AI people and programming languages people ought to heal their schism. I'd certainly like to hear more of VOX, and would cheerfully accept delivery of a copy of your paper (US Mail (mine): PO Box 2284, Santa Clara CA 95055). My apologies for using the net for a reply, but he's unreachable thru USENET, and I wanted to make a general point anyhow. -- Clay Phipps -- {allegra,amd70,cbosgd,dsd,floyd,harpo,hollywood,hpda,ihnp4, magic,megatest,nsc,oliveb,sri-unix,twg,varian,VisiA,wdl1} !fortune!phipps ------------------------------ Date: 12 Dec 83 15:29:10 PST (Monday) From: Ron Newman Subject: New Generation computing: Japanese and U.S. views [The following is a direct submission to AIList, not a reprint. It has also appeared on the Stanford bboards, and has generated considerable discussion there. I am distributing this and the following two reprints because they raise legitimate questions about the research funding channels available to AI workers. My distribution of these particular messages should not be taken as evidence of support for or against military research. -- KIL] from Japan: "It is necessary for each researcher in the New Generation Computer technology field to work for world prosperity and the progress of mankind. "I think it is the responsibility of each researcher, engineer and scientist in this field to ensure that KIPS [Knowledge Information Processing System] is used for good, not harmful, purposes. It is also necessary to investigate KIPS's influence on society concurrent with KIPS's development." --Tohru Moto-Oka, University of Tokyo, editor of the new journal "New Generation Computing", in the journal's founding statement (Vol. 1, No. 1, 1983, p. 2) and from the U.S.: "If the new generation technology evolves as we now expect, there will be unique new opportunities for military applications of computing. For example, instead of fielding simple guided missiles or remotely piloted vehicles, we might launch completely autonomous land, sea, and air vehicles capable of complex, far-ranging reconnaissance and attack misssions. The possibilities are quite startling, and suggest that new generation computing could fundamentally change the nature of future conflicts." --Defense Advanced Research Projects Agency, "Strategic Computing: New Generation Computing Technology: A Strategic Plan for its Development and Application to Critical Problems in Defense," 28 October 1983, p. 1 ------------------------------ Date: 13 Dec 83 18:18:23 PST (Tuesday) From: Ron Newman Subject: Re: New Generation computing: Japanese and U.S. views [Reprinted from the SU-SCORE bboard.] My juxtaposition of quotations is intended to demonstrate the difference in priorities between the Japanese and U.S. "next generation" computer research programs. Moto-Oka is a prime mover behind the Japanese program, and DARPA's Robert Kahn is a prime mover behind the American one. Thus I consider the quotations comparable. To put it bluntly: the Japanese say they are developing this technology to help solve human and social problems. The Americans say they are developing this technology to find more efficient ways of killing people. The difference in intent is quite striking, and will undoubtedly produce a "next-generation" repetition of an all too familiar syndrome. While the U.S. pours yet more money and scientific talent into the military sinkhole, the Japanese invest their monetary and human capital in projects that will produce profitable industrial products. Here are a couple more comparable quotes, both from IEEE Spectrum, Vol. 20, No. 11, November 1983: "DARPA intends to apply the computers developed in this program to a number of broad military applications... "An example might be a pilot's assistant that can respond to spoken commands by a pilot and carry them out without error, drawing upon specific aircraft, sensor, and tactical knowledge stored in memory and upon prodigious computer power. Such capability could free a pilot to concentrate on tactics while the computer automatically activated surveillance sensors, interpreted radar, optical, and electronic intelligence, and prepared appropriate weapons systems to counter hostile aircraft or missiles.... "Such systems may also help in military assessments on a battlefield, simulating and predicting the consequences of various courses of military action and interpreting signals acquired on the battlefield. This information could be compiled and presented as sophisticated graphics that would allow a commander and his staff to concentrate on the larger strategic issues, rather than having to manage the enormous data flow that will[!] characterize future battles." --Robert S. Cooper and Robert E. Kahn, DARPA, page 53. "Fifth generation computers systems are exptected to fulfill four major roles: (1) enhancement of productivity in low-productivity areas, such as nonstandardized operations in smaller industries; (2) conservation of national resources and energy through optimal energy conversion; (3) establishment of medical, educational, and other kinds of support systems for solving complex social problems, such as the transition to a society made up largely of the elderly; and (4) fostering of international cooperation through the machine translation of languages." --Tohru Moto-Oka, University of Tokyo, page 46 Which end result would *you* rather see? /Ron ------------------------------ Date: Tue 13 Dec 83 21:29:22-PST From: John B. Nagle Subject: Comparable quotes [Reprinted from the SU-SCORE bboard.] The goals of an effort funded by the military will be different than those of an effort aimed at trade dominance. Intel stayed out of the DoD VHSIC program because the founder of Intel felt that concentrating on fast, expensive circuits would be bad for business. He was right. The VHSIC program is aimed at making a few hundred copies of an IC for a few thousand each. Concentration on that kind of product will bankrupt a semiconductor company. We see the same thing in AI. There is getting to be a mini-industry built around big expensive AI systems on big expensive computers. Nobody is thinking of volume. This is a direct consequence of the funding source. People think in terms of keeping the grants coming in, not selling a million copies. If money came from something like MITI, there would be pressure to push forward to a volume product just to find out if there is real potential for the technology in the real world. Then there would be thousands of people thinking about the problems in the field, not just a few hundred. This is divirging from the main thrust of the previous flame, but think about this and reply. There is more here than another stab at the big bad military. ------------------------------ Date: Tue 13 Dec 83 10:40:04-PST From: Sumit Ghosh Subject: Ph.D. Oral Examination: Special Seminar [Reprinted from the SU-SCORE bboard.] ADA Techniques for Implementing a Rule-Based Generalised Design Verifier Speaker: Sumit Ghosh Ph.D. Oral Examination Monday, 19th Dec '83. 3:30pm. AEL 109 This thesis describes a top-down, rule-based design verifier implemented in the language ADA. During verification of a system design, a designer needs several different kinds of simulation tools such as functional simulation, timing verification, fault simulation etc. Often these tools are implemented in different languages, different machines thereby making it difficult to correlate results from different kinds of simulations. Also the system design must be described in each of the different kinds of simulation, implying a substantial overhead. The rule-based approach enables one to create different kinds of simulations, within the same simulation environment, by linking appropriate type of models with the system nucleus. This system also features zooming whereby certain subsections of the system design (described at a high level) can be expanded at a lower level, at run time, for a more detailed simulation. The expansion process is recursive and should be extended down to the circuit level. At the present implementation stage, zooming is extended to gate level simulation. Since only those modules that show discrepancy (or require detailed analysis) during simulation are simulated in details, the zoom technique implies a substantial reduction in complexity and CPU time. This thesis further contributes towards a functional deductive fault simulator and a generalised timing verifier. ------------------------------ Date: Mon 12 Dec 83 12:46-EST From: Philip E. Agre Subject: Walter Hamscher at the AI Revolving Seminar [Reprinted from the MIT-AI bboard.] AI Revolving Seminar Walter Hamscher Diagnostic reasoning for digital devices with static storage elements Wendesday 14 December 83 4PM 545 Tech Sq 8th floor playroom We view diagnosis as a process of reasoning from anomalous observations to a set of components whose failure could explain the observed misbehaviors. We call these components "candidates." Diagnosing a misbehaving piece of hardware can be viewed as a process of generating, discriminating among, and refining these candidates. We wish to perform this diagnosis by using an explicit representation of the hardware's structure and function. Our candidate generation methodology is based on the notions of dependency directed backtracking and local propagation of constraints. This methodology works well for devices without storage elements such as flipflops. This talk presents a representation for the temporal behavior of digital devices which allows devices with storage elements to be treated much the same as combinatorial devices for the purpose of candidate generation. However, the straightforward adaptation requires solutions to subproblems that are severely underconstrained. This in turn leads to an overly conservative and not terribly useful candidate generator. There exist mechanism-oriented solutions such as value enumeration, propagation of variables, and slices; we review these and then demonstrate what domain knowledge can be used to motivate appropriate uses of those techniques. Beyond this use of domain knowledge within the current representation, there are alternative perspectives on the problem which offer some promise of alleviating the lack of constraint. ------------------------------ End of AIList Digest ******************** 18-Dec-83 12:04:33-PST,17282;000000000001 Mail-From: LAWS created at 18-Dec-83 12:01:20 Date: Sun 18 Dec 1983 11:48-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #114 To: AIList@SRI-AI AIList Digest Sunday, 18 Dec 1983 Volume 1 : Issue 114 Today's Topics: Intelligence - Confounding with Culture, Jargon - Mental States, Scientific Method - Research Methodology ---------------------------------------------------------------------- Date: 13 Dec 83 10:34:03-PST (Tue) From: hplabs!hpda!fortune!amd70!dual!onyx!bob @ Ucb-Vax Subject: Re: Intelligence = culture Article-I.D.: onyx.112 I'm surprised that there have been no references to culture in all of these "what is intelligence?" debates... The simple fact of the matter is, that "intelligence" means very little outside of any specific cultural reference point. I am not referring just to culturally-biased vs. non-culturally-biased IQ tests, although that's a starting point. Consider someone raised from infancy in the jungle (by monkeys, for the sake of the argument). What signs of intelligence will this person show? Don't expect them to invent fire or stone axes; look how long it took us the first time around. The most intelligent thing that person could do would be on par with what we see chimpanzees doing in the wild today (e.g. using sticks to get ants, etc). What I'm driving at is that there are two kinds of "intelli- gence"; there is "common sense and ingenuity" (monkeys, dolphins, and a few people), and there is "cultural methodology" (people only). Cultural methodologies include all of those things that are passed on to us as a "world-view", for instance the notion of wearing clothes, making fire, using arithmetic to figure out how many people X bags of grain will feed, what spices to use when cooking, how to talk (!), all of these things were at one time a brilliant conception in someones' mind. And it didn't catch on the first time around. Probably not the second or third time either. But eventually someone convinced other people to try his idea, and it became part of that culture. And using that as a context gives other people an opportunity to bootstrap even further. One small step for a man, a giant leap for his culture. When we think about intelligence and get impressed by how wonder- ful it is, we are looking at its application in a world stuffed to the gills with prior context that is indispensible to every- thing we think about. What this leaves us with is people trying to define and measure a hybrid of common sense and culture without noticing that what they are interested in is actually two different things, plus the inter-relations between those things, so no wonder the issue seems so murky. For those who may be interested, general systems theory, general semantics, and epistemology are some fascinating related sub- jects. Now let's see some letters about what "common sense" is in this context, and about applying that common sense to (cultural) con- texts. (How recursive!) ------------------------------ Date: Tue, 13 Dec 83 11:24 EST From: Steven Gutfreund Subject: re: mental states I am very intriguied by Ferenando Pereira's last comment: Sorry, you missed the point that JMC and then I were making. Prygogine's work (which I know relatively well) has nothing to say about systems which have to model in their internal states equivalence classes of states of OTHER systems. It seems to me impossible to describe such systems unless certain sets of states are labeled with things like "believe(John,have(I,book))". That is, we start associating classes of internal states to terms that include mentalistic predicates. I may be missing the point, since I am not sure what "model in their internal states equivelence classes of states of OTHER systems" means. But I think you are saying is that `reasoning systems' that encode in their state information about the states of other systems (or their own) are not coverered by Ilya Prygogine's work. I think think you are engaging in a leap of faith here. What is the basis for believing that any sort of encoding of the state of other systems is going on here. I don't think even the philosophical guard phrase `equivalence class' protects you in this case. To continue in my role of sceptic: if you make claims that you are constructing systems that model their internal state (or other systems' internal states) [or even an equivalence class of those states]. I will make a claim that my Linear Programming Model of an computer parts inventory is also exhibiting `mental reasoning' since it is modeling the internal states of that computer parts inventory. This means that Prygogine's work is operative in the case of FSA based `reasoning systems' since they can do no more modeling of the internal state of another system than a colloidal suspension, or an inventory control system built by an operations research person. - Steven Gutfreund Gutfreund.umass@csnet-relay ------------------------------ Date: Wed 14 Dec 83 17:46:06-PST From: PEREIRA@SRI-AI.ARPA Subject: Mental states of machines The only reason I have to believe that a system encodes in its states classifications of the states of other systems is that the systems we are talking about are ARTIFICIAL, and therefore this is part of our design. Of course, you are free to say that down at the bottom our system is just a finite-state machine, but that's about as helpful as making the same statement about the computer on which I am typing this message when discussing how to change its time-sharing resource allocation algorithm. Besides this issue of convenience, it may well be the case that certain predicates on the states of other or the same system are simply not representable within the system. One does not even need to go as far as incompleteness results in logic: in a system which has means to represent a single transitive relation (say, the immediate accessibility relation for a maze), no logical combination can represent the transitive closure (accessibility relation) [example due to Bob Moore]. Yet the transitive closure is causally connected to the initial relation in the sense that any change in the latter will lead to a change in the former. It may well be the case (SPECULATION WARNING!) that some of the "mental state" predicates have this character, that is, they cannot be represented as predicates over lower-level notions such as states. -- Fernando Pereira ------------------------------ Date: 12 Dec 83 7:20:10-PST (Mon) From: hplabs!hao!seismo!philabs!linus!utzoo!dciem!mmt @ Ucb-Vax Subject: Re: Mental states of machines Article-I.D.: dciem.548 Any discussion of the nature and value of mental states in either humans of machines should include consideration of the ideas of J.G.Taylor (no relation). In his "Behavioral Basis of Perception" Yale University Press, 1962, he sets out mathematically a basis for changes in perception/behaviour dependent on transitions into different members of "sets" of states. These "sets" look very like the mental states referenced in the earlier discussion, and may be tractable in studies of machine behaviour. They also tie in quite closely with the recent loose talk about "catastrophes" in psychology, although they are much better specified than the analogists' models. The book is not easy reading, but it is very worthwhile, and I think the ideas still have a lot to offer, even after 20 years. Incidentally, in view of the mathematical nature of the book, it is interesting that Taylor was a clinical psychologist interested initially in behaviour modification. Martin Taylor {allegra,linus,ihnp4,uw-beaver,floyd,ubc-vision}!utzoo!dciem!mmt ------------------------------ Date: 14 Dec 1983 1042-PST From: HALL.UCI-20B@Rand-Relay Subject: AI Methods After listening in on the communications concerning definitions of intelligence, AI methods, AI results, AI jargon, etc., I'd like to suggest an alternate perspective on these issues. Rather than quibbling over how AI "should be done," why not take a close look at how things have been and are being done? This is more of a social-historical viewpoint, admitting the possibility that adherents of differing methodological orientations might well "talk past each other" - hence the energetic argumentation over issues of definition. In this spirit, I'd like to submit the following for interested AILIST readers: Toward a Taxonomy of Methodological Perspectives in Artificial Intelligence Research Rogers P. Hall Dennis F. Kibler TR 108 September 1983 Department of Information and Computer Science University of California, Irvine Irvine, CA 92717 Abstract This paper is an attempt to explain the apparent confusion of efforts in the field of artificial intelligence (AI) research in terms of differences between underlying methodological perspectives held by practicing researchers. A review of such perspectives discussed in the existing literature will be presented, followed by consideration of what a relatively specific and usable taxonomy of differing research perspectives in AI might include. An argument will be developed that researchers should make their methodological orientations explicit when communicating research results, both as an aid to comprehensibility for other practicing researchers and as a step toward providing a coherent intellectual structure which can be more easily assimilated by newcomers to the field. The full report is available from UCI for a postage fee of $1.30. Electronic communications are welcome: HALL@UCI-20B KIBLER@UCI-20B ------------------------------ Date: 15 Dec 1983 9:02-PST From: fc%usc-cse%USC-ECL@MARYLAND Subject: Re: AIList Digest V1 #112 - science In my mind, science has always been the practice of using the 'scientific method' to learn. In any discipline, this is used to some extent, but in a pure science it is used in its purest form. This method seems to be founded in the following principles: 1 The observation of the world through experiments. 2 Attempted explanations in terms of testable hypotheses - they must explain all known data, predict as yet unobserved results, and be falsifiable. 3 The design and use of experiments to test predictions made by these hypotheses in an attempt to falsify them. 4 The abandonment of falsified hypotheses and their replacement with more accurate ones - GOTO 2. Experimental psychology is indeed a science if viewed from this perspective. So long as hypotheses are made and predictions tested with some sort of experiment, the crudity of the statistics is similar to the statistical models of physics used before it was advanced to its current state. Computer science (or whatever you call it) is also a science in the sense that our understanding of computers is based on prediction and experimentation. Anyone that says you don't experiment with a computer hasn't tried it. The big question is whether mathematics is a science. I guess it is, but somehow any system in which you only falsify or verify based on the assumptions you made leaves me a bit concerned. Of course we are context bound in any other science, and can't often see the forest for the trees, but on the other hand, accidental discovery based on experiments with results which are unpredictable under the current theory is not really possible in a purely mathematical system. History is probably not a science in the above sense because, although there are hypotheses with possible falsification, there is little chance of performing an experiment in the past. Archeological findings may be thought of as an experiment of the past, but I think this sort of experiment is of quite a different nature than those that are performed in other areas I call science. Archeology by the way is probably a science in the sense of my definition not because of the ability to test hypotheses about the past through experimental diggings, but because of its constant development and experimental testing of theory in regards to the way nature changes things over time. The ability to determine the type of wood burned in an ancient fire and the year in which it was burned is based on the scientific process that archeologists use. Fred ------------------------------ Date: 13 Dec 83 15:13:26-PST (Tue) From: hplabs!hao!seismo!philabs!linus!utzoo!dciem!mmt @ Ucb-Vax Subject: Re: Information sciences vs. physical sciences Article-I.D.: dciem.553 *** This response is routed to net.philosophy as well as the net.ai where it came from. Responders might prefer to edit net.ai out of the Newsgroups: line before posting. I am responding to an article claiming that psychology and computer science arn't sciences. I think that the author is seriously confused by his prefered usage of the term ``science''. I'm not sure, but I think the article referenced was mine. In any case, it seems reasonable to clarify what I mean by "science", since I think it is a reasonably common meaning. By the way, I do agree with most of the article that started with this comment, that it is futile to define words like "science" in a hard and fast fashion. All I want here is to show where my original comment comes from. "Science" has obviously a wide variety of meanings if you get too careful about it, just as does almost any word in a natural language. But most meanings of science carry some flavour of a method for discovering something that was not known by a method that others can repeat. It doesn't really matter whether that method is empirical, theoretical, experimental, hypothetico-deductive, or whatever, provided that the result was previously uncertain or not obvious, and that at least some other people can reproduce it. I argued that psychology wasn't a science mainly on the grounds that it is very difficult, if not impossible, to reproduce the conditions of an experiment on most topics that qualify as the central core of what most people think of as psychology. Only the grossest aspects can be reproduced, and only the grossest characterization of the results can be stated in a way that others can verify. Neither do theoretical approaches to psychology provide good prediction of observable behaviour, except on a gross scale. For this reason, I claimed that psychology was not a science. Please note that in saying this, I intend in no way to downgrade the work of practicing psychologists who are scientists. Peripheral aspects, and gross descriptions are susceptible to attack by our present methods, and I have been using those methods for 25 years professionally. In a way it is science, but in another way it isn't psychology. The professional use of the word "psychology" is not that of general English. If you like to think what you do is science, that's fine, but remember that the definition IS fuzzy. What matters more is that you contribute to the world's well-being, rather than what you call the way you do it. -- Martin Taylor {allegra,linus,ihnp4,uw-beaver,floyd,ubc-vision}!utzoo!dciem!mmt ------------------------------ Date: 14 Dec 83 20:01:52-PST (Wed) From: hplabs!hpda!fortune!rpw3 @ Ucb-Vax Subject: Re: Information sciences vs. physical sc - (nf) Article-I.D.: fortune.1978 I have to throw my two bits in: The essence of science is "prediction". The missing steps in the classic paradigm of hypothesis-experiment-analysis- presented above is that "hypothesis" should be read "theory-prediction-" That is, no matter how well the hypothesis explains the current data, it can only be tested on data that has NOT YET BEEN TAKEN. Any sufficiently overdetermined model can account for any given set of data by tweaking the parameters. The trick is, once calculated, do those parameters then predict as yet unmeasured data, WITHOUT CHANGING the parameters? ("Predict" means "within an reasonable/acceptable confidence interval when tested with the appropriate statistical methods".) Why am I throwing this back into "ai"? Because (for me) the true test of whether "ai" has/will become a "science" is when it's theories/hypotheses can successfully predict (c.f. above) the behaviour of existing "natural" intelligences (whatever you mean by that, man/horse/porpoise/ant/...). ------------------------------ End of AIList Digest ******************** 20-Dec-83 21:56:56-PST,12199;000000000001 Mail-From: LAWS created at 20-Dec-83 21:56:22 Date: Tue 20 Dec 1983 21:48-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #115 To: AIList@SRI-AI AIList Digest Wednesday, 21 Dec 1983 Volume 1 : Issue 115 Today's Topics: Neurophysics - Left/Right-Brain Citation Request, Knowledge Representation, Science & Computer Science & Expert Systems, Science - Definition, AI Funding - New Generation Computing ---------------------------------------------------------------------- Date: 16 Dec 83 13:10:45-PST (Fri) From: decvax!microsoft!uw-beaver!ubc-visi!majka @ Ucb-Vax Subject: Left / Right Brain Article-I.D.: ubc-visi.571 From: Marc Majka I have heard endless talk, and read endless numbers of magazine-grade articles about left / right brain theories. However, I have not seen a single reference to any scientific evidence for these theories. In fact, the only reasonably scientific discussion I heard stated quite the opposite conclusion about the brain: That although it is clear that different parts of the brain are associated with specific functions, there is no logical (analytic, mathematical, deductive, sequential) / emotional (synthetic, intuitive, inductive, parallel) pattern in the hemispheres of the brain. Does anyone on the net have any references to any studies that have been done concerning this issue? I would appreciate any directions you could provide. Perhaps, to save the load on this newsgroup (since this is not an AI question), it would be best to mail directly to me. I would be happy to post a summary to this group. Marc Majka - UBC Laboratory for Computational Vision ------------------------------ Date: 15 Dec 83 20:12:46-PST (Thu) From: decvax!wivax!linus!utzoo!watmath!watdaisy!rggoebel @ Ucb-Vax Subject: Re: New Topic (technical) - (nf) Article-I.D.: watdaisy.362 Bob Kowalski has said that the only way to represent knowledge is using first order logic. ACM SIGART Newsletter No. 70, February 1980 surveys many of the people in the world actually doing representation research, and few of them agree with Kowalski. Is there anyone out there than can substantiate a claim for actually ``representing'' (what ever that means) ``knowledge?'' Most of the knowledge representation schemes I've seen are really deductive information description languages with quasi-formal extensions. I don't have a good definition of what knowledge is...but ask any mathematical logician (or mathematical philosopher) what they think about calling something like KRL a knowledge representation language. Randy Goebel Logic Programming Group University of Waterloo Waterloo, Ontario, CANADA N2L 3G1 ------------------------------ Date: 13 Dec 83 8:14:51-PST (Tue) From: hplabs!hao!seismo!philabs!linus!security!genrad!wjh12!foxvax1!br unix!jah @ Ucb-Vax Subject: Re: RE: Expert Systems Article-I.D.: brunix.5992 I don't understand what the "size" of a program has to do with anything. The notion that size is important seems to support the idea that the word "science" in "computer science" belongs in quote marks. That is, that CS is just a bunch of hacks anyhow. The theory folks, whom I think most of us would call computer scientists, write almost no programs. Yet, I'd say their contribution to CS is quite important (who analyzed the sorting algorithm you used this morning?) At least some parts of AI are still Science (with a capital "S"). We are exploring issues involving cognition and memory, as well as building the various programs that we call "expert systems" and the like. Pople's group, for example, are examining how it is that expert doctors come to make diagnoses. He is interested in the computer application, but also in the understanding of the underlying process. Now, while we're flaming, let me also mention that some AI programs have been awfully large. If you are into the "bigger is better" mentality, I suggest a visit to Yale and a view of some of the language programs there. How about FRUMP, which in its 1978 version took up three processes each using over 100K of memory, the source code was several hundred pages, and it contained word definitions for over 10,000 words. A little bigger than Haunt?? Pardon all this verbiage, but I think AI has shown itself both on the scientific level, by contributions to the field of psychology, (and linguistics for that matter) and by contributions to the state of the art in computer technology, and also in the engineering level, by designing and building some very large programs and some new programming techniques and tools. -Jim Hendler ------------------------------ Date: 19 Dec 1983 15:00-EST From: Robert.Frederking@CMU-CS-CAD.ARPA Subject: Re: Math as science Actually, my library's encyclopedia says that mathematics isn't a science, since it doesn't study phenomena, but rather is "the language of science". Perhaps part of the fuzziness about AI-as-science is that we are creating most of the phenomena we are studying, and the more theoretical components of what we are doing look a lot like mathematical logic, which isn't a science. ------------------------------ Date: Mon, 19 Dec 1983 10:21:47 EST From: AXLER.Upenn-1100@Rand-Relay (David M. Axler - MSCF Applications Mgr.) Subject: Defining "Science" For better or worse, there really isn't such a thing as a prototypical science. The meaning of the word 'science' has always been different in different realms of discourse: what the "average American" means by the term differs from what a physicist means, and neither of them would agree with an individual working in one of the 'softer' fields. This is not something we want to change, in my view. The belief that there must be one single standardized definition for a very general term is not a useful one, especially when the term is one that does not describe a explicit, material thing (e.g., blood, pencil, etc.). Abstract terms are always dependent on the social context of their use for their definition; it's just that academics often forget (or fail to note) that contexts other than their own fields exist. Even if we try and define science in terms of its usage of the "scientific method," we find that there's no clear definition. If you've yet to read it, I strongly urge you to take a look at Kuhn's "The Structure of Scientific Revolutions," which is one of the most important books written about science. He looks at what the term has meant, and does mean, in various disciplines at various periods, and examines very carefully how the definitions were, in reality, tied to other socially-defined notions. It's a seminal work in the study of the history and sociology of science. The social connotations of words like science affect us all every day. In my personal opinion, one of the major reasons why the term 'computer science' is gaining popularity within academia is that it dissociates the field from engineering. The latter field has, at least within most Western cultures, a social stigma of second-class status attached to it, precisely because it deals with mundane reality (the same split, of course, comes up twixt pure and applied mathematics). A good book on this, by the way, is Samuel Florman's "The Existential Pleasures of Engineering"; his more recent volume, "Blaming Technology", is also worth your time. --Dave Axler ------------------------------ Date: Fri 16 Dec 83 17:32:56-PST From: Al Davis Subject: Re: AIList Digest V1 #113 In response to the general feeling that Gee the Japanese are good guys and the Americans are schmucks and war mongers view, and as a member of one of the planning groups that wrote the DARPA SC plan, I offer the following questions for thought: 1. If you were Bob Kahn and were trying to get funding to permit continued growth of technology under the Reagan administration, would you ask for $750 million and say that you would do things in such a way as to prevent military use? 2. If it were not for DARPA how would we be reading and writing all this trivia on the ARPAnet? 3. If it were not for DARPA how many years (hopefully fun, productive, and challenging) would have been fundamentally different? 4. Is it possible that the Japanese mean "Japanese society" when they target programs for "the good of ?? society"? 5. Is it really possible to develop advanced computing technology that cannot be applied to military problems? Can lessons of destabilization of the US economy be learned from the automobile, steel, and TV industries? 6. It is obvious that the Japanese are quick to take, copy, etc. in terms of technology and profit. Have they given much back? Note: I like my Sony TV and Walkman as much as anybody does. 7. If DARPA is evil then why don't we all move to Austin and join MCC and promote good things like large corporate profit? 8. Where would AI be if DARPA had not funded it? Well the list could go on, but the direction of this diatribe is clear. I think that many of us (me too) are quick to criticize and slow to look past the end of our noses. One way to start to improve society is to climb down off the &%^$&^ ivory tower ourselves. I for one have no great desire to live in Japan. Al Davis ADAVIS @ SRI-KL ------------------------------ Date: Tue, 20 Dec 1983 09:13 EST From: HEWITT%MIT-OZ@MIT-MC.ARPA Subject: New Generation computing: Japanese and U.S. motivations Ron, I believe that you have painted a misleading picture of a complex situation. From talking to participants involved, I believe that MITI is funding the Japanese Fifth Generation Project primarily for commercial competitive advantage. In particular they hope to compete with IBM more effectively than as plug-compatible manufacturers. MITI also hopes to increase Japanese intellectual prestige. Congress is funding Strategic Computing to maintain and strengthen US military and commercial technology. A primary motivation for strengthening the commercial technology is to meet the Japanese challenge. ------------------------------ Date: 20 Dec 83 20:41:06 PST (Tuesday) From: Ron Newman Subject: Re: New Generation computing: Japanese and U.S. motivations Are we really in disagreement? It seems pretty clear from my quotes, and from numerous writings on the subject, that the Japanese intend to use the Fifth Generation Project to strengthen their position in commercial markets. We don't disagree there. It also seems clear that, as you say, "Congress is funding a project called Strategic Computing to maintain and strengthen US military and commercial technology." That should be parsed as "Military technology first, with hopes of commercial spinoff." If you think that's a misleading distortion, read the DARPA Strategic Computing Report. Pages 21 through 29 contain detailed specifications of the requirements of three specific military applications. There is no equivalent specification of non-military application requirements--only a vague statement on page 9 that commercial spinoffs will occur. Military requirements and terminology permeate the entire report. If the U.S. program is aimed at military applications, that's what it will produce. Any commercial or industrial spinoff will be incidental. If we are serious about strengthening commercial computer technology, then that's what we should be aiming for. As you say, that's certainly what the Japanese are aiming for. Isn't it about time that we put our economic interests first, and the military second? /Ron ------------------------------ End of AIList Digest ******************** 22-Dec-83 19:42:33-PST,10328;000000000001 Mail-From: LAWS created at 22-Dec-83 19:41:19 Date: Thu 22 Dec 1983 19:37-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #116 To: AIList@SRI-AI AIList Digest Friday, 23 Dec 1983 Volume 1 : Issue 116 Today's Topics: Optics - Request for Camera Design, Neurophysiology - Split Brain Research, Expert Systems - System Size, AI Funding - New Generation Computing, Science - Definition ---------------------------------------------------------------------- Date: Wed, 21 Dec 83 14:43:29 PST From: Philip Kahn Subject: REFERENCES FOR SPECIALIZED CAMERA DESIGN USING FIBER OPTICS In a conventional TV camera, the image falls upon a staring array of transducers. The problem is that it is very difficult to get very close to the focal point of the optical system using this technology. I am looking for a design of a camera imaging system that projects the light image onto a fiber optic bundle. The optical fibers are used to transport the light falling upon each pixel away from the camera focal point so that the light may be quantitized. I'm sure that such a system has already been designed, and I would greatly appreciate any references that would be appropriate to this type of application. I need to computer model such a system, so the pertinent optical physics and related information would be MOST useful. If there are any of you that might be interested in this type of camera system, please contact me. It promises to provide the degree of resolution which is a constraint in many vision computations. Visually yours, Philip Kahn ------------------------------ Date: Wed 21 Dec 83 11:38:36-PST From: Richard F. Lyon Subject: Re: AIList Digest V1 #115 In reply to on left/right brain research: Most of the work on split brains and hemispheric specialization has been done at Caltech by Dr. Roger Sperry and colleagues. The 1983 Caltech Biology annual report has 5 pages of summary results, and 11 recent references by Sperry's group. Previous year annual reports have similar amounts. I will mail copies if given an address. Dick Lyon DLYON@SRI-KL ------------------------------ Date: Wednesday, 21 December 1983 13:48:54 EST From: John.Laird@CMU-CS-H Subject: Haunt and other production systems. A few facts on productions systems. 1. Haunt consists of 1500 productions and requires 160K words of memory on a KL10. (So Frumps is a bit bigger than Haunt.) 2. Expert systems (R1, XSEL and PTRANS) written in a similar language consist of around 1500-2500 productions. 3. An expert system to perform VLSI design (DAA) consists of around 200 productions. ------------------------------ Date: 19 Dec 83 17:37:56-PST (Mon) From: decvax!dartvax!lorien @ Ucb-Vax Subject: Re: Humanistic Japanese vs. Military Americans Article-I.D.: dartvax.536 Does anyone know of any groups doing serious AI in the U.S. or Europe that emulate the Japanese attitude? --Lorien ------------------------------ Date: Wed 21 Dec 83 13:04:21-PST From: Andy Freeman Subject: Re: AIList Digest V1 #115 "If the U.S. program is aimed at military applications, that's what it will produce. Any commercial or industrial spinoff will be incidental." It doesn't matter what DoD and the Japanese project aim for. We're not talking about a spending a million on designing bullets but something much more like the space program. The meat of that specification was "American on Moon with TV camera" but look what else happened. Also, the goal was very low volume, but many of the products aren't. Hardware, which is probably the majority of the specification, could be where the crossover will be greatest. Even if they fail to get "a lisp machine in every tank", they'll succeed in making one for an emergency room. (Camping gear is a recent example of something similar.) Yes, they'll be able to target software applications, but at least the tools, skills, and people move. What distinguishes a US Army database system anyway? I can understand the objection that the DoD shouldn't have "all those cycles", but that isn't one of the choices. (How they are to be used is, but not through the research.) The new machines are going to be built - if nothing else the Dod can use Japanese ones. Even if all other things were equal, I don't think the economic ones are, why should they have all the fun? -andy ------------------------------ Date: Wednesday, 21 December 1983, 19:27-EST From: Hewitt at MIT-MC Subject: New Generation Computing: Japanese and U.S. motivations Ron, For better or worse, I do not believe that you can determine what will be the motivations or structure of either the MITI Fifth Generation effort or the DARPA Strategic Computing effort by citing chapter and verse from the two reports which you have quoted. /Carl ------------------------------ Date: Wed, 21 Dec 83 22:55:04 EST From: BRINT Subject: AI Funding - New Generation Computing It seems to me that intelligent folks like AIList readers should realize that the only reason Japan can fund peaceful and humanitarian research to the exclusion of military projects is that the United States provides the military protection and security guarantees (out of our own pockets) that make this sort of thing possible. (I believe Al Davis said it well in the last Digest.) ------------------------------ Date: 22 Dec 83 13:52:20 EST From: STEINBERG@RUTGERS.ARPA Subject: Strategic Computing: Defense vs Commerce Yes, it is a sad fact about American society that a project like Strategic Computing will only be funded if it is presented as a defense issue rather than a commercial/economic one. (How many people remember that the original name for the Interstate Highway system had the word "Defense" in it?) This is something we can and should work to change, but I do not believe that it is the kind of thing that can be changed in a year or two. So, we are faced with the choice of waiting until we change society, or getting the AI work done in a way that is not perfectly optimal for producing commercial/economic results. It should be noted that achieving the military goals will require very large advances in the underlying technology that will certainly have very large effects on non-military AI. It is not just a vague hope for a few spinoffs. So while doing it the DOD way may not be optimal it is not horrendously sub-optimal. There is, of course, a moral issue of whether we want the military to have the kinds of capabilities implied by the Strategic Computing plan. However, if the answer is no then you cannot do the work under any funding source. If the basic technology is achieved in any way, then the military will manage to use it for their purposes. ------------------------------ Date: 18 Dec 83 19:47:50-PST (Sun) From: pur-ee!uiucdcs!parsec!ctvax!uokvax!andree @ Ucb-Vax Subject: Re: Information sciences vs. physical sc - (nf) Article-I.D.: uiucdcs.4598 The definitions of Science that were offered, in defense of "computer Science" being a science, were just irrelevant. A field can lay claim to Science, if it uses the "scientific method" to make advances, that is: Hypotheses are proposed. Hypotheses are tested by objective experiments. The experiments are objectively evaluated to prove or disprove the hypotheses. The experiments are repeatable by other people in other places. - Keremath, care of: Robison decvax!ittvax!eosp1 or: allegra!eosp1 I have to disagree. Your definition of `science' excludes at least one thing that almost certainly IS a science: astronomy. The major problem here is that most astronomers (all extra-solar astronomers) just can not do experiments. Which is why they call it `obervational astronomy.' I would guess what is needed is three (at least) flavors of science: 1) experimental sciences: physics, chemistry, biology, psychology. Any field that uses the `scientific method.' 2) observational sciences: astronomy, sociology, etc. Any field that, for some reason or another, must be satisfied with observing phenomena, and cannot perform experiments. 3) ? sciences: mathematics, some cs, probably others. Any field that explores the universe of the possible, as opposed to the universe of the actual. What should the ? be? I don't know. I would tend to favor `logical,' but something tells me a lot of people will object. Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V1 #117 To: AIList@SRI-AI AIList Digest Friday, 30 Dec 1983 Volume 1 : Issue 117 Today's Topics: Reply - Fiber Optic Camera, Looping Problem - Loop Detection and Classical Psychology, Logic Programming - Horn Clauses, Disjunction, and Negation, Alert - Expert Systems & Molecular Design, AI Funding - New Generation Discussion, Science - Definition ---------------------------------------------------------------------- Date: 23 Dec 1983 11:59-EST From: David.Anderson@CMU-CS-G.ARPA Subject: fiber optic camera? The University of Pittsburgh Observatory is experimenting with just such an imaging system in one of their major projects, trying to (indirectly) observe planetary systems around nearby stars. They claim that the fiber optics provide so much more resolution than the photography they used before that they may well succeed. Another major advantage to them is that they have been able to automate the search; no more days spent staring at photographs. --david ------------------------------ Date: Fri 23 Dec 83 12:01:07-EST From: Michael Rubin Subject: Loop detection and classical psychology I wonder if we've been incorrectly thinking of the brain's loop detection mechanism as a sort of monitor process sitting above a train of thought, and deciding when the latter is stuck in a loop and how to get out of it. This approach leads to the problem of who monitors the monitor, ad infinitum. Perhaps the brain detects loops in *hardware*, by classical habituation. If each neuron is responsible for one production (more or less), then a neuron involved in a loop will receive the same inputs so often that it will get tired of seeing those inputs and fire less frequently (return a lower certainty value), breaking the loop. The detection of higher level loops such as "Why am I trying to get this PhD?" implies that there is a hierarchy of little production systems (or whatever), one for each chunk of knowledge. [Next question - how are chunks formed? Maybe there's a low-level explanation for that too, having to do with classical conditioning....] BTW, I thought of this when I read some word or other so often that it started looking funny; that phenomenon has gotta be a misfeature of loop detection. Some neuron in the dictionary decides it's been seeing that damn word too often, so it makes its usual definition less certain; the parse routine that called it gets an uncertain definition back and calls for help. --Mike Rubin ------------------------------ Date: 27 Dec 1983 16:30:08-PST From: marcel.uiuc@Rand-Relay Subject: Re: a trivial reasoning problem? This is an elaboration of why a problem I submitted to the AIList seems to be unsolvable using regular Horn clause logic, as in Prolog. First I'll present the problem (of my own devising), then my comments, for your critique. Suppose you are shown two lamps, 'a' and 'b', and you are told that, at any time, 1. at least one of 'a' or 'b' is on. 2. whenever 'a' is on, 'b' is off. 3. each lamp is either on or off. WITHOUT using an exhaustive generate-and-test strategy, enumerate the possible on-off configurations of the two lamps. If it were not for the exclusion of dumb-search-and-filter solutions, this problem would be trivial. The exclusion has left me baffled, even though the problem seems so logical. Check me on my thinking about why it's so difficult. 1. The first constraint (one or both lamps on) is not regular Horn clause logic. I would like to be able to state (as a fact) that on(a) OR on(b) but since regular Horn clauses are restricted to at most one positive literal I have to recode this. I cannot assert two independent facts 'on(a)', 'on(b)' since this suggests that 'a' and 'b' are always both on. I can however express it in regular Horn clause form: not on(b) IMPLIES on(a) not on(a) IMPLIES on(b) As it happens, both of these are logically equivalent to the original disjunction. So let's write them as Prolog: on(a) :- not on(b). on(b) :- not on(a). First, this is not what the disjunction meant. These rules say that 'a' is provably on only when 'b' is not provably on, and vice versa, when in fact 'a' could be on no matter what 'b' is. Second, a question ?- on(X). will result in an endless loop. Third, 'a' is not known to be on except when 'b' is not known to be on (which is not the same as when 'b' is known to be off). This sounds as if the closed-world assumption might let us get away with not being able to prove anything (if we can't prove something we can always assume its negation). Not so. We do not know ANYTHING about whether 'a' or 'b' are on OR off; we only know about constraints RELATING their states. Hence we cannot even describe their possible states, since that would require filling in (by speculative hypothesis) the states of the lamps. What is wanted is a non-regular Horn clause, but some of the nice properties of Logic Programming (eg completeness and consistency under the closed-world assumption, alias a reasonable negation operator) do not apply to non-regular Horn clauses. 2. The second constraint (whenever 'a' is on, 'b' is off) shares some of the above problems, and a new one. We want to say on(a) IMPLIES not on(b), or not on(b) :- on(a). but this is not possible in Prolog; we have to say it in what I feel to be a rather contrived manner, namely on(b) :- on(a), !, fail. Unfortunately this makes no sense at all to a theoretician. It is trying to introduce negative information, but under the closed-world assumption, saying that something is NOT true is just the same as not saying it at all, so the clause is meaningless. Alternative: define a new predicate off(X) which is complementary to on(X). That is the conceptualization suggested by the third problem constraint. 3. off(X) :- not on(X). on(X) :- not off(X). This idea has all the problems of the first constraint, including the creation of another endless loop. It seems this problem is beyond the capabilities of present-day logic programming. Please let me know if you can find a solution, or if you think my analysis of the difficulties is inaccurate. Marcel Schoppers U of Illinois at Urbana-Champaign {pur-ee|ihnp4}!uiucdcs!marcel ------------------------------ Date: Mon 26 Dec 83 22:15:06-PST From: Ken Laws Subject: High Technology Articles The January issue of High Technology has a fairly good introduction to expert systems for commercial applications. As usual for this magazine, there are corporate names and addresses and product prices. The article mentions that there are probably fewer than 200 "knowledge engineers" in the country, most at universities and think tanks; an AI postdoc willing to go into industry, but with no industry experience, can command $70K. The business outlook section is not the usual advice column for investors, just a list of some well-known AI companies. The article is also unusual in that it bases a few example of knowledge representation and inference on the fragment BIRD IS-A MAMMAL. Another interesting article is "Designing Molecules by Computer". Several approaches are given, but one seems particularly pertinent to the recent AIList discussion of military AI funding. Du Pont researchers are studying how a drug homes in on its receptor site. They use an Army program that generates line-of-sight maps for TV-controlled antitank missiles to "fly" a drug in and observe how its ability to track its receptor site on the enzyme surface is influenced by a variety of force fields and solvent interactions. A different simulation with a similar purpose uses robotic software for assembling irregular components to "pick up" the drug and "insert" it in the enzyme. -- Ken Laws ------------------------------ Date: 23 December 1983 21:41 est From: Dehn at MIT-MULTICS (Joseph W. Dehn III) Subject: "comparable" quotes Person at University of Tokyo, editor of a scientific/engineering journal, says computers will be used to solve human problems. Person at DARPA says computers will be used to make better weapons ("ways of killing people"). Therefore, Japanese are humane, Americans are warmongers. Huh? What is somebody at DARPA supposed to say is the purpose of his R&D program? As part of the Defense Department, that agency's goal SHOULD be to improve the defense of the United States. If they are doing something else, they are wasting the taxpayer's money. There are undoubtedly other considerations involved in DARPA's activities, bureaucratic, economic, scientific, etc., but, nobody should be astonished when an official statement of purpose states the official purpose! Assuming the nation should be defended, and assuming that advanced computing can contribute to defense, it makes sense for the national government to take an interest in advanced computing for defense. Thus, the question should not be, "why do Americans build computers to kill people", but rather why don't they, like the Japanese, ALSO, and independent of defense considerations (which are, as has been pointed out, different in Japan), build computers " to produce profitable industrial products"? Of course, before we try to solve this puzzle, we should first decide that there is something to be solved. Is somebody suggesting that because there are no government or quasi-government statements of purpose, that Americans are not working on producing advanced and profitable computer products? What ARE all those non-ARPA people doing out there in netland, anyway? Where are IBM's profits coming from? How can we meaningfully compare the "effort" being put into computer research in Japan and the U.S.? Money? People? How about results? Which country has produced more working AI systems (you pick the definition of "working" and "AI")? -jwd3 ------------------------------ Date: 29 Dec 1983 09:11:34-PST From: Mike Brzustowicz Subject: Japan again. Just one more note. Not only do we supply Japan's defense, but by treaty they cannot supply their own (except for a very small national guard-type force). ------------------------------ Date: 21 Dec 83 19:49:32-PST (Wed) From: harpo!eagle!mhuxl!ulysses!princeton!eosp1!robison @ Ucb-Vax Subject: Re: Information sciences vs. physical sc - (nf) Article-I.D.: eosp1.466 I disagree - astronomy IS an experimental science. Even before the age of space rockets, some celebrated astronomical experiments have been performed. In astronomy, as in all sciences, one observes, makes hypotheses, and then tries to verify the hypotheses by observation. In chemistry and physics, a lot of attention is paid to setting up an experiment, as well as observing the experiment; in astronomy (geology as well!), experiments consist mostly of observation, since there is hardly anything that people are capable of setting up. Here are some pertinent examples: (1) An experiment to test a theory about the composition of the sun has been going on for several years. It consists of an attempt to trap neutrinos from the sun in a pool of chlorine underground. The amount of neutrinos detected has been about 1/4 of what was predicted, leading to new suggestions about both the composition of the sun, and (in particle physics) the physical properties of neutrinos. (2) An experiment to verify Einstein's theory of relativity, particularly the hypothesis that the presence of large masses curves space (gravitational relativity) -- Measurements of Mercury's apparent position, during an eclipse of the sun, were in error to a degree consistent with Einstein's theory. Obviously, Astronomical experiments will seem to lie half in the realm of physics, since the theories of physics are the tools with which we try to understand the skies. Astronomers and physicists, please help me out here; I'm neither. In fact, I don't even believe in neutrinos. - Keremath, care of: Robison decvax!ittvax!eosp1 or: allegra!eosp1 ------------------------------ Date: Thu, 29 Dec 83 15:44 EST From: Hengst.WBST@PARC-MAXC.ARPA Subject: Re: AIList Digest V1 #116 The flaming on the science component of computer science intrigues me because it parallels some of the 1960's and 1970's discussion about the science component of social science. That particular discussion, to which Thomas Kuhn also contributed, also has not yet reached closure which leaves me with the feeling that science might best be described as a particular form of behavior by practitioners who possess certain qualifications and engage in certain rituals approved by members of the scientific tribe. Thus, one definition of science is that it is whatever it is that scientists do in the name of science ( a contextual and social definition). Making coffee would not be scientific activity but reading a professional book or entertaining colleagues with stimulating thoughts and writings would be. From this perspective, employing the scientific method is merely a particular form of engaging in scientific practice without judging the outcome of that scientific practice. Relying upon the scientific method by unlicensed practitioners would not result in science but in lay knowledge. This means that authoritative statements by members of scientific community are automatically given a certain truth value. "Professor X says this", "scientific study Y demonstrates that . . ." should all be considered as scientific statements because they are issued as authorative statements in the name of science. This interpretation of science discounts the role of Edward Teller as a credible spokesman in the area of nuclear weapons policy in foreign affairs. The "licensing" of the practitioners derives from the formalization of the training and education in the particular body of knowledge: eg. a university degree is a form of license. Scientific knowledge can differentiate itself from other forms of knowledge on the basis of attempts (but not necesssarily success) at formalization. Physical sciences study phenomena which lend themselves to better quantification (they do have better metrics!) and higher levels of formalization. The deterministic bodies of knowledge of the physical science allow for better prediction than the heavily probabilistic bodies of knowledge of the social science which facilitate explanation more so than prediction. I am not sure if a lack of predictive power or lack of availability of the scientific method (experimental design in its many flavors) makes anyone less a scientist. The social sciences are rich in description and insight which in my judgment compensates for a lack of hierarchical, deductive formal knowledge. From this point of view computer science is science if it involves building a body of knowledge with attempts at formulating rules in some consistent and verfiable manner by a body of trained practitioners. Medieval alchemy also qualifies due to its apprenticeship program (rules for admitting members) and its rules for building knowledge. Fortunately, we have better rules now. Acco ------------------------------ Date: Thu 29 Dec 83 23:38:18-PST From: Ken Laws Reply-to: AIList-Request@SRI-AI Subject: Philosophy of Science Discussion I hate to put a damper on the discussion of Scientific Method, but feel it is my duty as moderator. The discussion has been intelligent and entertaining, but has strayed from the central theme of this list. I welcome discussion of appropriate research techniques for AI, but discussion of the definition and philosophy of science should be directed to Phil-Sci@MIT-OZ. (Net.ai members are free to discuss whatever they wish, of course, but I will not pass further messages on this topic to the ARPANET readership.) -- Ken Laws ------------------------------ End of AIList Digest ********************