Date: Sun 6 Nov 1988 17:45-EST From: AIList Moderator Nick Papadakis Reply-To: AIList@AI.AI.MIT.EDU Us-Mail: MIT LCS, 545 Tech Square, Rm# NE43-504, Cambridge MA 02139 Phone: (617) 253-6524 Subject: AIList Digest V8 #121 To: AIList@AI.AI.MIT.EDU Status: R AIList Digest Monday, 7 Nov 1988 Volume 8 : Issue 121 Philosophy: AI as CS and the scientific epistemology of the common sense world Lightbulbs and Related Thoughts Re: Bringing AI back home Computer science as a subset of artificial intelligence Re: Limits of AI -or- The Kaleidoscope of Ideas ---------------------------------------------------------------------- Date: 31 Oct 88 2154 PST From: John McCarthy Subject: AI as CS and the scientific epistemology of the common sense world [In reply to message sent Mon 31 Oct 1988 20:39-EST.] Intelligence can be studied (1) through the physiology of the brain, (2) through psychology, (3) through studying the tasks presented in the achievement of goals in the common sense world. No one of the approaches can be excluded by a priori arguments, and I believe that all three will eventually succeed, but one will succeed more quickly than the other two and will help mop up the other two. I have left out sociology, because I think its contribution will be peripheral. AI is the third approach. It proceeds mainly in computer science departments, and many of its methods are akin to other computer science topics. It involves experimenting with computer programs and sometimes hardware and rarely includes either psychological or physiological experiments with humans or animals. It isn't further from other computer science topics than they are from each other and there are more and more hybrids of AI with other CS topics all the time. Perhaps Simon doesn't like the term AI, because his and Newell's work involves a hybrid with psychology and has involved psychological experiments as well as experimental computer programming. Surely some people should pursue that mixture, which has sometimes been fruitful, but most AI researchers stick to experimental programming and also AI theory. In my opinion the core of AI is the study of the common sense world and how a system can find out how to achieve its goals. Achieving goals in the common sense world involves a different kind of information situation than science has had to deal with previously. This fact causes most scientists to make mistakes in thinking about it. Some pick an aspect of the world that permits a conventional mathematical treatment and retreat into it. The result is that their results often have only a metaphorical relation to intelligence. Others demand differential equations and spend their time rejecting approaches that don't have them. Why does the common sense world demand a different approach? Here are some reasons. (1) Only partial information is available. It is partial not merely quantitatively but also qualitatively. We don't know all the relevant phenomena. Nevertheless, humans can often achieve goals using this information, and there is no reason humans can't understand the processes required to do it well enough to program them in computers. (2) The concepts used in common sense reasoning have a qualitatively approximate character. This is treated in my paper ``Ascribing Mental Qualities to Machines.'' (3) The theories that can be obtained will not be fully predictive of behavior. They will predict only when certain conditions are met. Curiously, while many scientists demand fully predictive theories, when they build digital hardware, they accept engineering specifications that aren't fully predictive. For example, consider a flip-flop with a J input, a K input and a clock input. The manufacturer specifies what will happen if the clock is turned on for long enough and then turned off provided exactly one of the J and K inputs remains high during this period and the other remains low. The specifications do not say what will happen if both are high or both are low or if they change while the clock is turned on. The manufacturer doesn't guarantee that all the flip-flops he sells will behave in the same way under these conditions or that he won't change without notice how they behave. All he guarantees is what will happen when the flip-flop is used in accordance with the ``design rules''. Computer scientists are also quite properly uninterested in non-standard usage. This contrasts with linear circuit theory which in principle tells how a linear circuit will respond to any input function of time. Newtonian and non-relativistic quantum mechanics tell how particles respond to arbitrary forces. Quantum field theory seems to be more picky. Many programs have specified behavior only for inputs meeting certain conditions, and some programming languages refrain from specifying what will happen if certain conditions aren't met. The implementor make the compiler do whatever is convenient or even not figure out what will happen. What we can learn about the common sense world is like what is specified about the flip-flop, only even more limited. Therefore, some people regard the common sense world as unfair and refuse to do science about it. ------------------------------ Date: 2 Nov 88 13:55:48 GMT From: pitstop!sundc!rlgvax!tony@sun.com (Tony Stuart) Subject: Lightbulbs and Related Thoughts On the way into work this morning I was stopped at a light near an office building. They were washing the windows using a large crane. This lead me to think about the time that a light was out in CCI's sign and they used a crane to replace it. I began to wonder whether they replace all the lights while they have the crane up or just the one that is out. Maybe it depends on how close the lights are to the end of their rated life. This got me thinking about the lights in the vanity at home. Two of the four have blown in the last couple of weeks. I remarked to Anne how it was interesting that lightbulbs do start to blow out at around the same time. This lead me to suggest that we remember to replace the blown out lightbulbs. The point is that an external stimulus, seeing the men wash the windows of the building, lead to a problem to solve, replacing the lights in the vanity. I have no doubt that if I had replaced those lights already then the train of thought would have continued until I encountered a problem that needed attention. The mind seems optimized for problem solving and perhaps one reason for miscellaneous ramblings is that they uncover problems. On a similar track, I have often thought that once we find a solution to a problem it is much more difficult to search for another solution. Over evolutionary history it is likely that life was sufficiently primitive that a single good solution was sufficient. The brain might be optimized such that the first good solution satisifies the problem seeking mode and to go beyond that solution requires concious effort. This is an argument for not resorting to a textbook as the first line of problem solving. The textbook is sure to give a good solution but perhaps not the best. With the textbook solution in mind it may be much more difficult to come up with an original solution that is better than the textbook one. For this reason it is best to try to solve the problem internally before going to some external device. There may also be some insite into how to make computers think. Lets say I designed my computer to follow trains of thought and at each thought it looked for unresolved questions. If there were no unresolved questions it would continue onto the next linked thought. Otherwise it would look for the solution to the problem. If the search did not turn up the information in memory it would result in the formation of a question. Anne suggests that these trains of thought are often triggered by external stimulae that a computer would not have. She says that we live in a sea of stimulae. I've often wondered about the differences between short term and long term memory. Here's a computer model for it. Assume that short term memory is information stored as sentences and long term memory is information stored in data structures with organized field name/field value/relationship links. Information is initially stored in the sentence based short term memory. In a background process, or when our minds are otherwise idle, a task searches through the short term memory for data that might resolve questions (holes) in the long term memory. (Which is searched I don't really know.) This information in the short term memory is then appropriately cataloged in the long term memory. Another task is responsible for purging sentences from the short term memory. It could use a first in-first out or more likely a least frequently used algorithm. A side effect of this model is that information in short term memory cannot be used unless there is a hole in the long term memory. This leads to problems in bootstrapping the process, but assuming there is a solution to that problem, it also models behavior that is present in humans. This is the case of feeling that one hears a word or phrase a lot after he knows what it means. Another part of the side effect is that one cannot use information that he has unless it fits. This means that it must be discarded until the long term memory is sufficiently developed to accept it. -- Anthony F. Stuart, {uunet|sundc}!rlgvax!tony CCI, 11490 Commerce Park Drive, Reston, VA 22091 ------------------------------ Date: 2 Nov 88 15:54:37 GMT From: umix!umich!itivax!dhw@uunet.UU.NET (David H. West) Subject: Re: Bringing AI back home In article <1776@crete.cs.glasgow.ac.uk> Gilbert Cockton writes: > >In a previous article, Ray Allis writes: >>If AI is to make progress toward machines with common sense, we >>should first rectify the preposterous inverted notion that AI is >>somehow a subset of computer science, >Nothing preposterous at all about this. AI is about applications of >computers, and you can't sensibly apply computers without using computer >science. All that this shows is that AI has a non-null intersection with CS, not that it is a subset of it. > Many people would be happy if AI boy scouts came down >from their technological utopian fantasies and addressed the sensible >problem of optimising human-computer task allocation in a humble, >disciplined and well-focussed manner. Many people would have been happier (in the short term) if James Watt had stopped his useless playing with kettles and gone out and got a real job. >There are tasks in the world. Computers can assist some of these >tasks, but not others. Understanding why this is the case lies at the >heart of proper human-machine system design. The problem with hard AI is >that it doesn't want to know that a real division between automatable >and unautomatable tasks does exist in practice. You seem to believe that this boundary is fixed. Well, it will be unless people work on moving it. > Why are so >many AI workers so damned ignorant of the problems with >operationalising definitions of intelligence, as borne out by nearly a >century of psychometrics here? There was a time when philosophers concerned themselves with questions such as whether magnets move towards each other out of love or against their will. Why were those who wanted instead to measure forces so damned ignorant of the problems with the philosophical approach? Maybe they weren't, and that's *why* they preferred to measure forces. >Common sense is a labelling activity >for beliefs which are assumed to be common within a (sub)culture. Partially. >Such social constructs cannot have a machine embodiment, nor can any Cannot? Why not? "Do not at present" I would accept. >The minute words like common sense and intelligence are used, the >relevant discipline becomes the sociology of knowledge. *A* relevent discipline. AI is at present more concerned with the structure and machine implementation of common sense than with its detailed content. -David West dhw%iti@umix.cc.umich.edu {uunet,rutgers,ames}!umix!itivax!dhw CDSL, Industrial Technology Institute, PO Box 1485, Ann Arbor, MI 48106 ------------------------------ Date: Wed, 2 Nov 88 14:55:01 pst From: Ray Allis Subject: Computer science as a subset of artificial intelligence In <1776@crete.cs.glasgow.ac.uk> Gilbert Cockton writes: >In a previous article, Ray Allis writes: >>If AI is to make progress toward machines with common sense, we >>should first rectify the preposterous inverted notion that AI is >>somehow a subset of computer science, >Nothing preposterous at all about this. AI is about applications of >computers, I was disagreeing with that too-limited definition of AI. *Computer science* is about applications of computers, *AI* is about the creation of intelligent artifacts. I don't believe digital computers, or rather physical symbol systems, can be intelligent. It's more than difficult, it's not possible. >> or call the research something other than "artificial intelligence". >Is this the real thrust of your argument? Most people would agree, >even Herb Simon doesn't like the term and says so in "Sciences of the >Artificial". No, "I said what I meant, and I meant what I said". The insistence that "artificial intelligence research" is subsumed under computer science is preposterous. That position precludes the development of intelligent machines. >> Computer science has nothing whatever to say about much of what we call >> intelligent behavior, particularly common sense. >Only sociology has anything to do with either of these, so to >place AI within CS is to lose nothing. Only the goal. >Intelligence is a value judgement, not a definable entity. "Intelligence" is not a physical thing you can touch or put in a bottle, like water or carbon dioxide. "Intelligent" is an adjective, usually modifying the noun "behavior", and it does describe something measurable; a quality of behavior an organism displays in coping with its environment. I think intelligent behavior is defined more objectively than, say, the quality of an actor's performance in A Midsummer Night's Dream, which IS a value judgement. > Common sense is a labelling activity >for beliefs which are assumed to be common within a (sub)culture. > >Such social constructs cannot have a machine embodiment, nor can any >academic discipline except sociology sensibly address such woolly >epiphenomena. I do include cognitive psychology within this exclusion, >as no sensible cognitive psychologist would use terms like common sense >or intelligence. The mental phenomena which are explored >computationally by cognitive psychologists tend to be more basic and >better defined aspects of individual behaviour. The minute words like >common sense and intelligence are used, the relevant discipline becomes >the sociology of knowledge. Common sense does not depend on social consensus. I mean by common sense those behaviors which nearly everyone acquires in the course of existence, such as reluctance to put your hand into the fire. I contend that symbol systems in general, digital computers in particular, and therefore computer science, are inadequate for artifacts which "have common sense". Formal logic is only a part of human intellectual being, computer science is about the mechanization of that part, AI is (or should be) about the entire intellect. Hence AI is something more ambitious than CS, and not a subcategory. That's why I used the word "inverted". >Gilbert Cockton, Department of Computing Science, The University, Glasgow > gilbert@uk.ac.glasgow.cs !ukc!glasgow!gilbert Ray Allis Boeing Computer Services, Seattle, Wa. ray@boeing.com bcsaic!ray ------------------------------ Date: 3 Nov 88 15:12:29 GMT From: bwk@mitre-bedford.arpa (Barry W. Kort) Subject: Re: Limits of AI -or- The Kaleidoscope of Ideas In article <2211@datapg.MN.ORG> sewilco@datapg.MN.ORG writes: > Life, and thus evolution, is merely random exceptions to entropy. There is an emerging theory on the evolution of complex stable systems. (See for example Ilya Prigogine's book, _Order out of Chaos_.) The mathematical theory of fixed points, and the related system-theoretic idea of eigenfunctions and eigenvalues suggest that stable, recurring modes or patterns may emerge naturally from any system when "the outputs are shorted to the inputs". Consider for instance, the map whose name is "The Laws of Physics and Chemistry". Plug in some atoms and molecules into this map (or processor) and you get out atoms and molecules. By the Fixed Point Theorem, one would expect there to exist a family of atoms and molecules which remain untransformed by this map. And this family could have arbitrarily complex members. DNA comes to mind. (Crystals are another example of a self-replicating, self-healing structure). So the "random exceptions to entropy" may not be entirely random. They may be the eigenvalues and eigenfunctions of the system. The Mandelbrot Set has shown us how exquisitely beautiful and complex structures can arise out of simple recursion and feedback loops. --Barry Kort ------------------------------ End of AIList Digest ********************