Date: Sun 5 Jun 1988 23:13-EDT From: AIList Moderator Nick Papadakis Reply-To: AIList@AI.AI.MIT.EDU Us-Mail: MIT Mail Stop 38-390, Cambridge MA 02139 Phone: (617) 253-2737 Subject: AIList Digest V7 #19 To: AIList@AI.AI.MIT.EDU Status: R AIList Digest Monday, 6 Jun 1988 Volume 7 : Issue 19 Today's Topics: The Future of the Free Will Discussion Philosophy: a good jazz ensemble Who else isn't a science? Bad AI: A Clarification randomness ---------------------------------------------------------------------- Date: Sat 4 Jun 88 21:44:01-PDT From: Raymond E. Levitt Subject: Free Will Raymond E. Levitt Associate Professor Center for Integrated Facility Engineering Departments of Civil Engineering and Computer Science Stanford University ============================================================== Several colleagues and I would like to request that the free will debate - which seems endless - be set up on a different list with one of the more active contributors as a coordinator. The value of the AILIST as a source of current AI research issues, conferences, software queries and evaluations, etc., is diminished for us by having to plough through the philosophical dialectic in issue after issue of the AILIST. Perhaps you could run this message and take a poll of LIST readers to help decide this in a democratic way. Thanks for taking on the task of coordinating the AILIST. It is a great service to the community. Ray Levitt ------- [Editor's Note: Thank you, Mr. Levitt, and many thanks to all those who have written expressing interest or comments regarding AIList. I regret that I have not had time to respond to many of you individually, as I have lately been more concerned with the simple mechanics of generating digests and dealing with the average of sixty bounce messages per day than with the more substantive issues of moderation. However, a new COMSAT mail-delivery program is now orbiting, and we may perhaps be able to move away from the days of lost messages, week-long delays, and 50K digests ... My heartfelt apologies to all. Being rather new at this job, I have hesitated to express my opinion with respect to the free-will debate, preferring to retain the status quo and hoping that the problem would fix itself. But since Mr. Levitt is only the latest of several people who have complained about this particular issue, I feel I must take some action. Clearly this discussion is interesting and valuable to many of the participants, but equally clearly it is less so for many others. I have tried as far as possible to group the free-will discussions in digests apart from other matters, so people uninterested in the topic could simply 'delete' the offending digests unread. (There are many readers who only have access to the undigested stream and cannot do this.) Several people have suggested moving the discussion to a USENET list called 'tallk.philosophy'. The difficulty here is that AIList crosses USENET, INTERNET and BITNET, and not all readers would be able to contribute. In V7#6, John McCarthy said: > I am not sure that the discussion should progress further, but if > it does, I have a suggestion. Some neutral referee, e.g. the moderator, > should nominate principal discussants. Each principal discussant should > nominate issues and references. The referee should prune the list > of issues and references to a size that the discussants are willing > to deal with. They can accuse each other of ignorance if they > don't take into account the references, however perfunctorily. > Each discussant writes a general statement and a point-by-point > discussion of the issues at a length limited by the referee in > advance. Maybe the total length should be 20,000 words, > although 60,000 would make a book. After that's done we have another > free-for-all. I suggest four as the number of principal discussants > and volunteer to be one, but I believe that up to eight could > be accomodated without making the whole thing too unwieldy. > The principal discussants might like help from their allies. > > The proposed topic is "AI and free will". I would be more than willing to coordinate this effort, but I have, as yet, received no responses expressing an opinion one way or the other. I invite the readers of AIList who have found the free-will discussion interesting (as opposed to those who have not) to send me net mail at AILIST-REQUEST@AI.AI.MIT.EDU concerning the future of this discussion. Please send me a separate message, and do not intersperse your comments with other contributions, whether to the free-will debate or other matters. In the meantime, I will continue to send out digests covering the free-will topic, although separate from other material. - nick ] ------------------------------ Date: Sat, 4 Jun 88 14:21:06 EDT From: George McKee Subject: Artificial Free Will -- what's it good for? Obviously many people think that the question of whether or not humans have free will is important to a lot of people, and thinking about how it could be implemented in a computer program is an effective way to clarify exactly what we're talking about. I think the McDermott's contributions show this -- they're getting pretty close to pseudocode that you could think about translating into executable programs. (But just to put in my historical two cents, I first saw this kind of analysis in a proceedings of the Pontifical Academy of Sciences article by D.M.MacKay in about 1968.) If free will is programmable, it's appropriate to then ask "why bother?", and "how will we recognize success?", i.e. to make explicit the scientific motivation for such a project, and the methodology used to evaluate it. I can see two potential reasons to work on building free will into a computer system: (1) formalizing free will into a program will finally show us the structure of an aspect of the human mind that's been confusing to philosophers and psychologists for thousands of years. (2) free-will-competent computer systems will have some valuable abilities missing from systems without free will. Reason 1 is unquestionably important to the cognitive sciences, and insofar as AI programs are an essential tool to cognitive scientists, *writing* a program that includes free will as part of its structure might be a worthwhile project. But *executing* a program embodying free will won't necessarly show us anything that we didn't know already. Free will in its sense as a consequence of the incompleteness of an individual's self-model has an essentially personal character, that doesn't get out into behavior except as verbal behavior in arguments about whether it exists at all. For instance, I haven't noticed in this discussion any mention of how you recognize free will in anyone other than yourself. If you can't tell whether I have free will or not, how will you recognize if my program has it without looking at the code? And if you always need to look at the code, what's the point in actually running the program, except for other, irrelevant reasons? (This same argument applies to consciousness, and explains why I, and maybe others out there as well, after sketching out some pseudocode that would have some conscious notion of its own structure, decided to leave implementation to the people who work on the formal semantics of reflective languages like "3lisp" or "brown". (See the proceedings of the Lisp and FP conferences, but be careful to avoid thinking about multiprocessing while reading them.)) Which brings us to Reason 2, and free will from the perspective of pure, pragmatic AI. As far as I can tell, the only way free will can affect behavior is by making it unpredictable. But since there are many other, easier ways to get unpredictability without having to invoke the demoniacal (or is it oracular?) Free Will, I'm back to "why bother?" again. Unpredictability in behavior is certainly valuable to an autonomous organism in a dangerous environment, both as an individual (e.g. a rabbit trying to outrun a hungry fox) and as a group (e.g. a plant species trying to find a less-crowded ecological niche), but in spite of my use of the word "trying" this doesn't need to involve any will, free or otherwise. In highly sophisticated systems like human societies, where statements of ability (like diplomas :-) are often effectively equivalent to demonstrations of ability, claiming "I have Free Will, you'll fail if you try to predict/control my behavior!" might well be quite effective in fending off a coercive challenge. But computer systems aren't in this kind of social situation (at least the ones I work with aren't). In fact they are designed to be as predictable as possible, and when they aren't, it indicates a failure either of understanding or in design. So again, I don't see the need for Artificial Free Will, fake or real. My background is largely psychology, so I think that it's valuable to understand how it is that people feel that their behavior is fundamentally unconstrained by external forces, especially social ones. But I also don't think that this illusion has any primary adaptive value, and I don't think there's anything to be gained by giving it to a computer. If this is true, then the proper place for this discussion is some cognitive-science list, which I'd be happy to read if I knew where to send my subscription request. - George McKee NU Computer Science ------------------------------ Date: Fri 03 Jun 1988 15:04 CDT From: Subject: >> RE>...a steeplejack and his mate, a good jazz ensemble, ...) >No. Fire and ambulance personnel have regulations, basketball has rules >and teams discuss strategy and tactics during practice, and even jazz >musicians use sheet music sometimes. I don't mean to say that implicit >communication doesn't exist, just that it's not as useful. I don't know >how to build steeples, but I'll bet it can be written down. This person obviously doesnt know much about music performance. Of course jazzers use sheet music but have you ever seen a page out of the Realbook, if you have, you never follow it literally. Even the more classically oriented stuff, the bandwidth of the information on the sheet no where nearly approaches the gestural dimensions required to musically or otherise correctly interpret the piece. For one, there is tradition, which is passed through schools, performance ensembles, and now recording media. If you are a trumpet player in an orchestra, and you see a dot over a note, that dot means different things depending on which composer, period, genre, tempo, etc, ect, ect. With jazz there are even more intangibles, like you can be on top of the beat or lay in the pocket. Nowhere is a written method which can gaurantee that you are going to get the right feel, man, you just gotto feel it baby *snap*. You still may want to call jazz a language, of course it is, it has meaning, but it is not something that can be put down in a machine readable format. Jeff Beer, UUCJEFF@ECNCDC.BITNET ================================== Language is a virus --- Laurie Anderson ------------------------------ Date: 3 Jun 88 20:22:32 GMT From: maui!bjpt@locus.ucla.edu (Benjamin Thompson) Subject: Re: Who else isn't a science? In article <10510@agate.BERKELEY.EDU> weemba@garnet.berkeley.edu writes: >Gerald Edelman, for example, has compared AI with Aristotelian >dentistry: lots of theorizing, but no attempt to actually compare >models with the real world. AI grabs onto the neural net paradigm, >say, and then never bothers to check if what is done with neural >nets has anything to do with actual brains. This is symptomatic of a common fallacy. Why should the way our brains work be the only way "brains" can work? Why shouldn't *A*I workers look at weird and wonderful models? We (basically) don't know anything about how the brain really works anyway, so who can really tell if what they're doing corresponds to (some part of) the brain? Ben ------------------------------ Date: Sat, 4 Jun 88 00:14 EST From: EBARNES%HAMPVMS.BITNET@MITVMA.MIT.EDU Subject: Re: Sorry, no philosophy allowed here Editors: >If you can't write it down, you can't program it. This comes down to two questions. Can we build machines with original thought capabilities, and what is meant by `program'. I think that it is possible to build machines which will think originally. The question now becomes: "Is what we do to set these "free thinking" machines up considered programing". It would not be a strict set of instructions, but we would surely instill the rules of deductive reasoning in the machine. Whether or not this is "programing" is an uniteresting question. Call it what you will, one way makes the original statement true and the other way makes it false. Eric Barnes ------------------------------ Date: 4 Jun 88 15:41:26 GMT From: trwrb!aero!venera.isi.edu!smoliar@bloom-beacon.mit.edu (Stephen Smoliar) Subject: Re: Bad AI: A Clarification In article <1299@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: > Research requires skill. Research into humanity requires special >skills. Computer scientists and mathematicians are not taught these skills. > There is no questioning the premise of the first sentence. I am even willing to grant, further, that artificial intelligence (or at least aspects which are of particularly interest to me) may be regarded as "research into humanity." However, after that, Cockton's argument begins to fall apart. Just what are those "special skills" which such research "requires?" Does anyone have them? Does Cockton regard familiarity with the humanistic literature as such a skill? I suspect there could be some debate as to whether or not extensive literary backgroud is a skill, particularly when the main virtue of such knowledge is that it provides one with a history of how one's predecessors have failed on similar tasks. There is no doubt that it is valuable to know that certain paths lead to dead ends; but when there are so many forks in the road, it is not always easy to determine WHICH fork was the one which ultimately embodied the incorrect decision. Perhaps I am misrepresenting Cockton by throwing too much weight on "being well read." In that case, he can set the record straight by doing a better job of characterizing those skills which he feels computer scientists and mathematicians lack. Then he can tell us how many humanists have those skills and have exercised them in the investigation of intelligence with a discipline which he seems to think the AI community lacks. Let he who is without guilt cast the first stone, Mr. Cockton! (While we're at it, is your house made of glass, by any chance?) One final note on bad AI. I don't think there is anyone reading this newsgroup who would doubt that there is bad AI. However, in another article, Cockton seems quite willing to admit (as most of us knew already) that there is bad sociology, too. One of the more perceptive writers on social behavior, Theodore Sturgeon (who had the good sense to articulate his views in the palatable form of science fiction), once observed that 90% of X is crud, for any value of X . . . that can be AI, sociology, or classical music. Bad AI is easy enough to find and even easier to pick on. Rather than biting the finger of the bad stuff, why not take the time to look where the finger of the good stuff is really pointing? ------------------------------ Date: 4 Jun 88 16:09:56 GMT From: trwrb!aero!venera.isi.edu!smoliar@bloom-beacon.mit.edu (Stephen Smoliar) Subject: Re: AI and Sociology In article <1301@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: > AI can be >mulitidisciplinary, but it is, for me, unique in its insistence on a single >paradigm which MUST distort the researcher's view of humanity, as well as the >research consumer's view on a bad day. Indefensible. > . . . and patently untrue! Perhaps Mr. Cockton has suffered from an attempt to study AI in such a dogmatic environment. His little anecdote about the advior who put him off AI is quite telling. I probably would have been put off by such an attitude, too. Fortunately, I could affort the luxury of changing advisors without changing my personal interest in questions I wanted to pursue. First of all, it is most unclear that there is any single paradigm for the pursuit of artificial intelligence. Secondly, it is at least somewhat unclear that any paradigm which certainly will INFLUENCE one's view of humanity also necessarily DISTORTS it. To assume that the two thousands years of philosophy which have preceded us have provided an undistorted view of humanity is arrogance in its most ignorant form. Finally, having settled that there is more than one paradigm, we can hardly accuse the AI community of INSISTING on any paradigm. > >Again, I challenge AI's rejection of social criticisms of its paradigm. We >become what we are through socialisation, not programming (although some >teaching IS close to programming, especially in mathematics). Thus a machine >can never become what we are, because it cannot experience socialisation in >the >same way as a human being. Thus a machine can never reason like us, as it can >never absorb its model of reality in a proper social context. Again, there >are >well documented examples of the effect of social neglect on children. >Machines >will not suffer in the same way, as they only benefit from programming, and >not all forms of human company. Actually, if there is any agreement at all in the AI community it is in the conviction to be sceptical of all authoritative usage of the word "never." I, personally, do not feel that any social criticisms are being rejected wholesale. However, AI is a very difficult area to pursue (at least if you are really interested in a research pursuit, as opposed to marketing a new shell for building expert systems). One of the most important keys to getting any sort of viable result at all is understanding how to break off a piece of the whole, big, intimidating problem whose investigation is likely to provide some insight. This generally leads to the construction of a model, usually in the form of a software artifact. The next key is to investigate that model to see what it has REALLY told us. A good example of such an investigation is the one by Brown and Lenat on what AM and EURISKO APPEAR (their words) to work. There are valid questions about socialization which can probably be formulated in terms of communities of automata. However, we need to form a better vision of what we can expect by way of the behavior of individual automata before we can express those questions in any useful way. There is no doubt that this will take some time. However, there is at least a glimmer of hope that when we get around to expressing them, we will have a better idea of what we are talking about than those who have chosen to reject the abstraction of automata out of hand. ------------------------------ Date: 4 Jun 88 16:21:47 GMT From: trwrb!aero!venera.isi.edu!smoliar@bloom-beacon.mit.edu (Stephen Smoliar) Subject: Re: Aah, but not in the fire brigade, jazz ensembles, rowing eights,... In article <239@proxftl.UUCP> tomh@proxftl.UUCP (Tom Holroyd) writes: >In article <1171@crete.cs.glasgow.ac.uk>, gilbert@cs.glasgow.ac.uk (Gilbert >Cockton) writes: >> In article <5499@venera.isi.edu> smoliar@vaxa.isi.edu.UUCP (Stephen Smoliar) >>writes: >> >The problem comes in deciding >> >WHAT needs to be explicitly articulated and what can be left in the >> >"implicit >> >background." >> ... >> For people who haven't spent all their life in academia or >> intellectual work, there will be countless examples of carrying out >> work in near 100% implicit background (watch fire and ambulance >> personelle who've worked together as a team for ages, watch a basketball >> team, a steeplejack and his mate, a good jazz ensemble, ...) > >No. Fire and ambulance personnel have regulations, basketball has rules >and teams discuss strategy and tactics during practice, and even jazz >musicians use sheet music sometimes. I don't mean to say that implicit >communication doesn't exist, just that it's not as useful. I don't know >how to build steeples, but I'll bet it can be written down. > Take a look at Herb Simon's article in ARTIFICIAL INTELLIGENCE about "ill-structured problems" and then decide whether or not you want to make that bet. ------------------------------ Date: 5 Jun 88 17:29:29 GMT From: glacier!jbn@labrea.stanford.edu (John B. Nagle) Subject: Re: Bad AI: A Clarification On this subject, one should read Drew McDermott's "Artificial Intelligence meets Natural Stupidity" (ACM SIGART newsletter, #57, April 1976.) His comments are all too apt today. John Nagle ------------------------------ Date: 5 Jun 88 18:07:42 GMT From: glacier!jbn@labrea.stanford.edu (John B. Nagle) Subject: Re: Ill-structured problems In article <5644@venera.isi.edu> Stephen Smoliar writes: >Take a look at Herb Simon's article in ARTIFICIAL INTELLIGENCE about >"ill-structured problems" and then decide whether or not you want to make that bet. A reference to the above would be helpful. Little progress has been made on ill-structured problems in AI. This reflects a decision in the AI community made in the early 1970s to defer work on those hard problems and go for what appeared to be an easier path, the path of logic/language/formal representation sometimes referred to as "mainstream AI". In the early 1970s, both Minsky and McCarthy were working on robots; McCarthy proposed to build a robot capable of assembling a Heathkit color TV kit. This was a discrete component TV, requiring extensive soldering and hand-wiring to build, not just some board insertion. The TV kit was actually purchased, but the robot assembly project went nowhere. Eventually, somebody at the SAIL lab assembled the TV kit, which lived in the SAIL lounge for many years, providing diversion for a whole generation of hackers. Embarassments like this tended to discourage AI workers from attempting projects where failure was so painfully obvious. With more abstract problems, one can usually claim (one might uncharitably say "fake") success by presenting one's completed system only with carefully chosen problems that it can deal with. But in dealing with the physical world, one regularly runs into ill-structured problems that can't be bypassed. This can be hazardous to your career. If you fail, your thesis committee will know. Your sponsor will know. Your peers will know. Worst of all, you will know. So most AI researchers abandoned the problems of vision, navigation, decision-making in ill-structured physical environments, and a number of other problems which must be solved before there is any hope of dealing effectively with the physical world. Efforts were focused on logic, language, abstraction, and "understanding". Much progress was made; we now have a whole industry devoted to the production of systems with a superficial but useful knowledge of a wide assortment of problems. Still, in the last few years, the state of the art in that area seems to have reached a plateau. That set of ideas may have been mined out. Certainly the public claims made a few years ago have not been furfilled. (I will refrain from naming names; that's not my point today.) The phrase "AI winter" is heard in some quarters. ------------------------------ Date: Sat, 4 Jun 88 17:48:57 EDT From: aboulang@WILMA.BBN.COM Reply-to: aboulanger@bbn.com Subject: randomness In AIList Digest V7 #4, Barry Kort writes: >If I wanted to give my von Neumann machine a *true* random number >generator, I would connect it to an A/D converter driven by thermal >noise (i.e. a toasty resister). I recall that a Zener diode is a good source of noise (but cannot remember the spectrum it gives). It could be a good idea to utilize a Zener / A-D converter random number generator in Monte Carlo simulations. Andy Ylikoski Ahem, all this stuff about analog sources being better random sources is a bit of a "scientific" urban myth. It is instructive to go back to the papers of the early 60's and see what it took to utilize analog random sources. The basic problem in analog sources is correlation. To wit: "A Hybrid Analog-Digital Pseudo-Random Noise Generator", R.L.T. Hampton, AFIPS Conference Proceedings, Vol 25, 1964 Spring Joint Computer Conference. 287-301. To quote a little: "By precision clamping, the RMS level of binary noise can be closely controlled, but the non-stationarity of the circuits used to obtain electrical noise, even form stationary mechanism such an a radio-active source, still create problems and expense. For example, the 80 Kc random-telegraph wave generator .... required a fairly sophisticated and not completely satisfactory count-rate control loop. In the design of University of Arizona's new ASTRAC II iterative differential analyzer ... it was decided to abandon analog noise generation completely. Instead, the machine will employ a digital shift-register sequence generator ..." If you would like to investigate recent high-quality theoretical work on this matter, see the paper: "Generating Quasi-random Sequences from Semi-random Sources", Miklos Santha & Umesh V. Vazirani, Journal of Computer and System Sciences, Vol 33, No 1, August 1986, 75-87. They propose a clever method to eliminate the correlations in analog sources. Help stamp out scientific urban myths! Albert Boulanger aboulanger@bbn.com BBN Labs ------------------------------ End of AIList Digest ********************