Date: Tue 12 Apr 1988 22:41-PDT From: AIList Moderator Kenneth Laws Reply-To: AIList@KL.SRI.COM Us-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList V6 #66 - Probability, Intelligence, Ethics, Future of AI To: AIList@KL.SRI.COM Status: R AIList Digest Wednesday, 13 Apr 1988 Volume 6 : Issue 66 Today's Topics: Theory - Probability, Opinion - Simulated Intelligence & Ethics of AI & The Future of AI ---------------------------------------------------------------------- Date: Thu, 7 Apr 88 00:13:57 HAE From: Spencer Star Reply-to: Subject: Probability: is it appropriate, necessary, or practical > a uniquely probabilistic approach to uncertainty may be > inappropriate, unnecessary and impractical. D. J. Spiegelhalter For the record, it was this quote from Spiegelhalter by Paul Creelman that prompted my questioning the accuracy of the quote. I have been able to track down the quote. A few words were dropped from what Spiegelhalter actually wrote. "Deviations from the methodological rigor and coherence of probability theory have been criticized, but in return it has been argued that a uniquely probabilistic approach to uncertainty may be inappropriate, unnecessary, and impractical." Spiegelhalter goes on to discuss these points and then reply to them point by point. Roses to Paul for sending me the source and brickbats to him for the liberties he took with it. ------------------------------ Date: 5 Apr 88 01:26:17 GMT From: mind!eliot@princeton.edu (Eliot Handelman) Subject: Simulated Intelligence Intelligence draws upon the resources of what Dostoevsky, in the "Notes from Underground", called the "advantageous advantage" of the individual who found his life circumscribed by "logarithms", or some form of computational determinism: the ability to veto reason. My present inclination is to believe that AI, in the long run, may only be a test for underlying mechanical constraints of of theories of intelligence, and therefore inapplicable to to the simulation of human intelligence. I'd like to hear this argued for or against. Best wishes to all, Eliot Handelman ------------------------------ Date: Mon, 4 Apr 88 12:18 EST From: Jeffy Subject: submission: Ethics of AI From: (I have deleted much text from in between the following lines) >But the payoff can be tremendous. >In the development stage, AI is expensive, >but in the long term it is cost effective. >the demand for AI is so great that we have no choice but to >push on. I would question the doctrine of "what is most _cost-effective_ (in the long term of course) is best." I think that, as Caroline Knight said, "Whatever the far future uses of AI are we can try to make the current uses as humane and as ethical as possible." I mean, what are we developing it for anyway? It often seems that AI is being developed for a specific purpose, but nobody seems to want to be explicit about what it is. Technology is not neutral. If you develop AI mainly as a war technology, then you will have a science that is most easily suited for war (as far as I know, DARPA is _the_ main funder for AI projects). Here is a quote from a book by Marcus Raskin and Herbert Bernstein: (they are talking about the Einstein-Bohr debate here, and how the results of Quantum Mechanics show us an observer created universe): "Bohr's position puts man, or at least his machines, at the center of scientific inquiry. If he is correct, science's style and purpose has to change. The problem has been that the physicists have not wanted to make any critical evaluation of their scientific work, an evaluation which their research cried out for just because of their belief that human beings remain at the center of inquiry, and man cannot know fundamental laws of nature. They rejected Einstins's conception of a Kantian reality and without saying it, his view of scientific purpose. Even though no fundamental laws can be found independent of man's beliefs and machines he constructs, scientists abjure making moral judgements as part of their work, even though they know - and knew - that the very character of science had changed. Standards for rational inquiry demand that moral judgements should be added as an integral part of any paricular experiment. Unless shown otherwise, I do not see how transformations of social systems, or the generation of a new consciousness can occur if we hold on to narrow conceptions of rational inquiry. Inquiry must now focus on relationsips. The reason is that rational inquiry is not, cannot, and should not be sealed from everyday life, institutional setting or the struggles which are carried on throughout the world. How rational inquiry is carried on, who we do it for, what we think about and _what we choose to see_ is never insulated or antiseptic. Once we communticate it through the medium of language, the symbols of mathematics, the metaphors and clishes of everyday like, we call forth in the mids of readers of fellow analysts other issues and considerations that may be outside of the four cornedrs of the experiment or inquiry. What they bring to what they see, read, or replicate is related to their purpose or agenda or their unconscious interpretations" - from: "NEW WAYS OF KNOWING", page 115 In any case, >But the payoff can be tremendous. >In the development stage, AI is expensive, >but in the long term it is cost effective. >the demand for AI is so great that we have no choice but to >push on. is a wrong way to view what one is doing, when one is doing AI. - Jeff Coggshall (JCOGGSHALL@HAMPVMS.BITNET) (This article is also in response to: From: ARMAND%BCVMS.BITNET@MITVMA.MIT.EDU Subject: STATUS OF EXPERT SYSTEMS?) ------------------------------ Date: 31 Mar 88 01:26:50 GMT From: jsnyder@june.cs.washington.edu (John Snyder) Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] In article <4640@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: >... But the demand for AI is so great that we have no choice but to >push on. We always have the choice not to develop a technology; what may be lacking are reasons or will. jsnyder@june.cs.washington.edu John R. Snyder {ihnp4,decvax,ucbvax}!uw-beaver!jsnyder Dept. of Computer Science, FR-35 University of Washington 206/543-7798 Seattle, WA 98195 ------------------------------ Date: 1 Apr 88 22:26:06 GMT From: arti@vax1.acs.udel.edu (Arti Nigam) Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] In article <4565@june.cs.washington.edu> you write: > >We always have the choice not to develop a technology; what may be lacking >are reasons or will. I heard this from one of the greats in computer-hardware-evolution, only I don't remember his name. What he said, and I say, is essentially this; if you are part of an effort towards progress, in whatever field or domain, you should have some understanding of WHERE you are going and WHY you want to get there. Arti Nigam ------------------------------ Date: 31 Mar 88 15:10:30 GMT From: ulysses!sfmag!sfsup!saal@ucbvax.Berkeley.EDU (S.Saal) Subject: Re: The future of AI - my opinion I think the pessimism about AI is a bit more subtle. Whenever something is still only vaguely understood, it is considered a part of AI. Once we start understanding the `what,' `how,' and (sometimes) `why' we no longer consider it a part of AI. For example, all robotics used to be part of AI. Now robotics is a field unto itself and only the more difficult aspects (certain manipoulations, object recognition, etc) are within AI anymore. Similarly so for expert systems. It used to be that ES were entirely within the purview of AI. That was when the AI folks had no real idea how to do ESs and were trying all sorts of methods. Now they understand them and two things have happened: expert systems are an independant branch of computer science and people have found that they no longer need to rely on the (advanced) AI type languages (lisp, etc) to get the job done. Ironically, this makes AI a field that must make itself obsolete. As more areas become understood, they will break off and become their own field. If not for finding new areas, AI would run out of things for it to address. Does this mean it isn't worth while to study AI? Certainly not. If for no other reason than AI is the think tank, problem _finder_ of computer science. So what if no problem in AI itself is ever solved? Many problems that used to be in AI have been, or are well on their way to being, solved. Yes, the costs are high, and it may not look as though much is actually coming out of AI research except for more questions, but asking the questions and lookling for the answers in the way that AI does, is a valid and useful approach. -- Sam Saal ..!attunix!saal Vayiphtach HaShem et Peah HaAtone ------------------------------ Date: 3 Apr 88 11:03:30 GMT From: bloom-beacon.mit.edu!boris@bloom-beacon.mit.edu (Boris N Goldowsky) Subject: Re: The future of AI - my opinion In article <2979@sfsup.UUCP> saal@sfsup.UUCP (S.Saal) writes: Ironically, this makes AI a field that must make itself obsolete. As more areas become understood, they will break off and become their own field. If not for finding new areas, AI would run out of things for it to address. Isn't that true of all sciences, though? If something is understood, then you don't need to study it anymore. I realize this is oversimplifying your point, so let me be more precise. If you are doing some research and come up with results that are useful, people will start using those results for their own purposes. If the results are central to your field, you will also keep expanding on them and so forth. But if they are not really of central interest, the only people who will keep them alive are these others... and if, as in the case of robotics, they are really useful results they will be very visibly and profitably kept alive. But I think this can really happen in any field, and in no way makes AI "obsolete." Isn't finding new areas what science is all about? Bng -- Boris Goldowsky boris@athena.mit.edu or @adam.pika.mit.edu %athena@eddie.UUCP @69 Chestnut St.Cambridge.MA.02139 @6983.492.(617) ------------------------------ Date: 4 Apr 88 16:25:49 GMT From: hubcap!mrspock@gatech.edu (Steve Benz) Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] >From article <1134@its63b.ed.ac.uk>, by gvw@its63b.ed.ac.uk (G Wilson): > In article <4640@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: >> Moreover, your opinion that conventional techniques can >>replace AI is ludicrous. Consider the area of natural language. What >>conventional techniques that you know of can extract information from >>natural language text or translate a passage from English to French? > > Errmmm...show me *any* program which can do these things? To date, > AI has been successful in these areas only when used in toy domains. In a real world (real world at least as far as real money will carry you...) project here, we developed a nearly-natural-language system that deals with the "toy domain" of reading mail, querying databases, and some other stuff. It may be a toy, but some folks were willing to lay out some signifigant number of dollars to get it. These applications are based off of a lazy-evaluation, functional language (I wouldn't call that a "conventional technique.") But the best part about the whole thing (as far as our contract monitor is concerned) is that it really wasn't all that expensive to do--less than 20 man-months went into the development of the language and fitting out the old menu-driven software with the new technique. Overall, it was a highly successful venture, allowing us to create high-quality user-interfaces very quickly, and develop them semi-independently of the application itself. None of the "conventional techniques" we had used before allowed us this. So you see, AI has application, I think the problem is that AI techniques like expert systems, and functional/logic programming simply haven't filtered out of the University in sufficient quantity to make an impact on the marketplace. The average BS-in-CS-graduate probably has had a very limited exposure to these techniques, hence he/she will be afraid of the unknown and will prefer to stick with "conventional techniques." To say that AI will never catch on is like saying that high-level languages should never have cought on. At one point it looked unlikely that HLL would gain wide acceptance, better equipment and better understanding by the programming community made them practical. - Steve mrspock@hubcap.clemson.edu ...!gatech!hubcap!mrspock ------------------------------ Date: 3 Apr 88 08:51:59 GMT From: mcvax!ukc!its63b!gvw@uunet.uu.net (G Wilson) Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] In article <4640@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: > Moreover, your opinion that conventional techniques can >replace AI is ludicrous. Consider the area of natural language. What >conventional techniques that you know of can extract information from >natural language text or translate a passage from English to French? Errmmm...show me *any* program which can do these things? To date, AI has been successful in these areas only when used in toy domains. >The future of AI is going to be full of unrealistic hype and disappointing >failures. Just like its past, and present. Does anyone think AI would be as prominent as it is today without (a) the unrealistic expectations of Star Wars, and (b) America's initial nervousness about the Japanese Fifth Generation project? > But the demand for AI is so great that we have no choice but to >push on. Manifest destiny?? A century ago, one could have justified continued research in phrenology by its popularity. Judge science by its results, not its fashionability. I think AI can be summed up by Terry Winograd's defection. His SHRDLU program is still quoted in *every* AI textbook (at least all the ones I've seen), but he is no longer a believer in the AI research programme (see "Understanding Computers and Cognition", by Winograd and Flores). Greg Wilson ------------------------------ Date: 31 Mar 88 20:02:22 GMT From: necntc!linus!philabs!ttidca!hollombe@AMES.ARC.NASA.GOV (The Polymath) Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] In article <962@daisy.UUCP> klee@daisy.UUCP (Ken Lee) writes: >What do people think of the PRACTICAL future of artificial intelligence? My empoloyers just sponsored a week-long in-house series of seminars, films, vendor presentations and demonstrations of expert systems technology. I attended all of it, so I think I can reasonably respond to this. Apparently, the expert systems/knowledge engineering branch of so called AI (of which, more later) has made great strides in the last few years. There are many (some vendors claim thousands) of expert system based commercial applications running in large and small corporations all over the country. In the past week we saw presentations by Gold Hill Computers (GOLDWORKS), Aion Corp. (ADS), Texas Instruments (Personal Consultant Plus) and Neuron Data (Nexpert Object). The presentations were impressive, even taking into account their sales nature. None of the vendors is in any financial trouble, to say the least. All claimed many delivered, working systems. A speaker from DEC explained that their Vax configurator systems couldn't have been developed without an expert system (they tried and failed) and is now one of the oldest and most famous expert systems running. It was pointed out by some of the speakers that companies using expert systems tend to keep a low profile about it. They consider their systems as company secrets, proprietary information that gives them an edge in their market. Personal Impressions: The single greatest advantage of expert systems seems to be their rapid prototyping capability. They can produce a working system in days or weeks that would require months or years, if it could be done at all, with conventional languages. That system can subsequently be modified very easily and rapidly to meet changing conditions or include new rules as they're discovered. Once a given algorithm has stabilized over time, it can be re-written in a more conventional language, but still accessed by the expert system. The point being that the algorithm may never have been determined at all but for the adaptable rapid prototyping environment. (The DEC Vax configurator, mentioned above, is an example of this. Much of it, but not all, has been converted to conventional languages). As for expense, prices of systems vary widely, but are coming down. TI offers a board with a LISP mainframe-on-a-chip (their term) that will turn a MAC-II into a LISP machine for as little as $7500. Other systems went as high as an order of magnitude over that. I personally think these won't really take off 'til the price drops another order of magnitude to put them in the hands of the average home hacker. Over all, I'd have to say that expert systems, at least, are alive and well with a bright future ahead of them. About Artificial Intelligence: I maintain this is a contradiction in terms, and likely to be so for the forseeable future. If we take "intelligence" to mean more than expert knowledge of a very narrow domain there's nothing in existence that can equal the performance of any mammal, let alone a human being. We're just begining to explore the types of machine architectures whose great^n- grandchildren might, someday, be able to support something approaching true AI. I'll be quite amazed to see it in my lifetime (but the world has amazed me before (-: ). -- The Polymath (aka: Jerry Hollombe, hollombe@TTI.COM) Illegitimati Nil Citicorp(+)TTI Carborundum 3100 Ocean Park Blvd. (213) 452-9191, x2483 Santa Monica, CA 90405 {csun|philabs|psivax|trwrb}!ttidca!hollombe ------------------------------ End of AIList Digest ********************