Date: Sun 3 Jul 1988 00:54-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 V8 #2 To: AIList@AI.AI.MIT.EDU Status: R AIList Digest Sunday, 3 Jul 1988 Volume 8 : Issue 2 Today's Topics: No digests for one week Philosophy: On applying AI replicating the brain with a Turing machine metaepistemology ASL vs dance Auto_Suggestion? Announcements: Directions and Implications of Advanced Computing - DIAC-88 Intermediate Mechanisms For Activation Spreading ---------------------------------------------------------------------- Date: Sun, 3 Jul 88 00:49 EDT Subject: No digests for one week I have been unexpectedly called away for a period of about one week. Unless I am lucky and manage to obtain net access during that time, there will be no digests sent out. Apologies to all. - nick ------------------------------ Date: Fri, 1 Jul 88 14:18:14 EDT From: "Bruce E. Nevin" Subject: ON applying AI The following is excerpted without permission from The Boston Globe Magazine for June 26, 1988, pp. 39-42 of the cover article by D. C. Denison entitled "The ON team, software whiz Mitch Kapor's new venture": In the conference room, where the ON team has assembled for a group interview, a question is posed: What developments in artificial intelligence have made their project possible? The first response comes immediately: "The lack of progress." Then William Woods, ON's principal technologist, takes a turn. "What came out of artificial intelligence that's useful to us is kind of like what came out of the space program that's useful for everybody on Earth --" "Velcro," someone interrupts. "Tang." From the other side of the table. "Are we the Tang of artificial intelligence?" Woods continues undaunted. "Artificial intelligence has given us a tool kit of engineering techniques. AI has been driven by people who've been tilting at windmills, but their techniques are pretty good for what we want to do." Although the early promise of artificial-intelligence research has been tempered--we still don't have computers that can understand English or reason like a human expert--the possibilities are so seductive, so intriguing, and so potentially profitable that the field continues to attract some of the best minds in the computer field. Which is why it wans't surprising that when Mitchell Kapor left Lotus two years ago, he became a visiting scholar at MIT's Center for Cognitive Science, a leading center of artifial-intelligence-related research. And it's not at all surprising that when Kapor and [Peter] Miller put together the ON team, at least one AI veteran of Woods' stature was part of the group. Yet "artificial intelligence" has become such a buzzword, such an umbrella term, that when the topic is brought up, experts such as Woods and interested explorers such as Kapor take deep breaths and try to redefine the terms of discussion. "When you talk about AI," Kapor says, "you're talking about many, many things at once: a body of research, certain kinds of goals and aspirations that are characteristic of the people who are in it, certain fields of inquiry; you've got soft stuff, you've got hard stuff, you have mythology--the term 'AI' casts a broad shadow. "I also think there's a reason why so much attention is paid to the AI question in the nontechnical press," he continues, "and that is that there are some very bombastic people in the AI community who have spoken incredibly irresponsibly, who've made careers out of that. But if you make the assumption that because AI gets a lot of attention in the press there's a lot going on in the field, you might be making a big mistake." When Kapor and Woods first met soon after Kapor left Lotus, they discovered that they shared a similar view of the value of current AI research. First of all, they both felt that the goal most often attributed to artificial-intelligence research--the creation of a computer that "thinks" just like a human--was so remote as to be essentially impossible. Kapor's experience at MIT had convinced him that scientists still have no idea how people really think. Therefore, any attempt to design a computer that works the way people think is doomed. A more realistic approach, according to Kapor, would be to design computer programs that are compatible with the way people think, that help amplify a person's intelligence rather than try to duplicate it. Kapor felt that some artificial-intelligence techniques, when applied to that goal, could be very powerful. Last year, Woods, who was working at Applied Expert Systems, discovered that Kaapor had leased a floor in the same building, and he began stopping in for informal conversations. Eventually, after discussing the ON Technology project with Kapor and Miller, and studying their business plan, he accepted the position of principal technologist and moved his things down two floors into a large corner office. One of the possibilities [that opens up], which Woods will be actively working on during the next two years, is a more sympathetic fit between people and their computers. "I want to take an abstract perspective of what people's mental machinery does very well and what a machine can do well," Woods says, "and design ways that you can couple the two together to complement each other. For example, machines can do long sequences of complicated steps without leaving out one. People will forget something with frequency. On the other hand, people can walk down the street without falling in holes. To get a mechanical artifact to do that is a challenge that hasn't even been aapproximately approached after several decades of research." Woods pauses to frame his thoughts. "But if you could get the right interface technology and conceptual framework, on the machine side, to match up with what people really want to do on our side--that would be a very nice arrangement. Bruce Nevin bn@cch.bbn.com ------------------------------ Date: Fri, 1 Jul 88 09:55:16 EDT From: Duke Briscoe Subject: Re: replicating the brain with a Turing machine >Date: Wed, 29 Jun 88 9:26:50 PDT >From: jlevy.pa@Xerox.COM >Subject: Re: AIList Digest V7 #46 replicating the brain with a > Turing machine > >Andy Ylikoski asks why you can't replicate the brain's exact functions >with a Turing machine. First off, the brain is not a single machine but >a whole bunch of them. Therefore "replacing it with a Turing machine" >wouldn't get you there. I think this is not a valid point because a single Turing machine (TM) can simulate the actions of a group of parallel TMs. >Turing machines have an inherent limitation in that they are not >reactive i.e. they are unable to react to the environment directly. On >the other hand, the brain is in direct communication with a number of >input devices (eyes, ears, nose, touch-sense, etc.), all of which are >sending data at the same time. TMs are usually only used as a theoretical tool. If you were actually going to implement one, you could have a multi-track input tape with one tape having an alphabet representing sensory input sampled at an appropriate rate. Issues of real-time response discussed below. >An interesting question is whether the brain's software suffers from the >Church-Rosser problem which is present in functional languages - >basically, you cannot, in a functional language, see that a certain >source of input is empty and later detect input on it. It seems that >this is not so, since we are able to close our eyes and later open them, >seeing again. In a functional program to simulate a brain, you are assuming that closing your eyes equates to closing an input stream, while in fact real optic nerves continue sending information even when the eyes are closed. Even though I have just shown that I think the points above are invalid, I'm still not sure that brain functions can be theoretically modelled by a TM. TMs operate in discrete steps, while material objects act in continuous dimensions of time and space (as far as we know, otherwise perhaps the universe is a giant, parallel Turing-equivalent computer). Assuming reality is continuous, a TM model might closely approximate something material for some period of time, but would eventually diverge. Plus there is the whole problem that any physical TM implementation would have problems such as unavoidable bit errors which would invalidate its exact correspondence to the abstract TM. However, physical implementations, even using non-organic materials, of computers should still theoretically be capable of the same computing powers as organic brains. There just seem to be limitations in using a restricted TM model to prove things about brain computable functions. Maybe an expanded TM model is needed which takes into account physical properties of space-time. Or perhaps the space-time is discrete at some level we have not yet detected, in which case the current plain TM would be adequate. After all, electric charges seem to be discrete. ------------------------------ Date: 2 Jul 88 19:11:40 GMT From: proxftl!bill@bikini.cis.ufl.edu (T. William Wells) Subject: Re: metaepistemology In a previous article, YLIKOSKI@FINFUN.BITNET writes: > In AIList Digest V7 #41, John McCarthy > writes: > > >I want to defend the extreme point of view that it is both > >meaningful and possible that the basic structure of the > >world is unknowable. It is also possible that it is > >knowable. I did not see the origins of this debate but it appears to be nothing more than an attempt to defend the Kantian noumenal vs. phenomenal distinction. Instead of wasting time debating this issue, why don't those of you who are interested go and study some philosophy? And, for those of you who are going to say "but I have", carefully compare this view with Kant and you will see that they are in essence identical. ------------------------------ Date: Fri, 1 Jul 88 08:16:09 EDT From: "Bruce E. Nevin" Subject: ASL vs dance I have not studied ASL, but it seems prima facie likely that the gesture system of a sign language used by the deaf would have both a formal and an expressive aspect, just as the gesture system of ordinary spoken phonology does. In the phonology of a given language, there is a limited inventory of usually <50 contrasts, differences that make a difference. The phonetic `content' of these contrasts (the actual sounds used to embody the contrasts in a given utterance by a given speaker at a given time) is subject to remarkably free `stretching', which languages exploit for expressive purposes as well as in dialect variation. Leigh Lisker long ago speculated that the function of semantically empty greeting rituals ("Hello, how are you?" "Fine, and you?") is to provide an opportunity for conversants to tune in on the fundamental frequency of each other's voice and calibrate for the relative location of vowel formants. Calibrating for the phonetic envelope each uses to embody the contrasts of their shared language is also a likely function. I would expect that deaf folks have to attune themselves to the gestural style and expressive range of conversants, but I can't think of anything analogous to the fundamental frequency and vowel formants in phonology. I would astonished if there were no analogs of phonemic contrasts in ASL utterances, no fundamental and stable `differences that make a difference' to other ASL users, and I would be be very interested to learn what they are like. In language, it is the formal aspect, the system of contrasts or `differences that make a difference', and the information structures that they support, that are the ostensive focus. This is surely the case with the sign languages of the Deaf also. In dance, by contrast, it is the expressive aspect that is typically the main point, and the formal structure is subsidiary, merely a channel for expressive communication, else the piece is seen as dry, technical, academic, uninspired. One may apply such adjectives to a conversation, but with scarcely the same devastating critical effect! Conversely, a critic who discussed what a choreographer was saying without comment on how she or he said it would generally be thought to be missing the point. An interesting thing here is that the expressive aspects of language use actually do influence people much more than the literal content (words 7%, tone 32%, kinesics 61%: Albert Murahbian, _Public Places & Private Spaces_; Ray Birdwhistell, _Kinesics & Context_). In this respect, we very much need an understanding and representation of the expressive `stretching' of a formal structure, since that is where most of human communication takes place (as distinct from simple transmission of literal information). This is a big part of the difference between linguistic competence (Chomsky) and communicative competence (Hymes). An AI that has the first (a hard enough problem!) but not the second will always be missing the point and misconstruing the literal meaning of what is said. I should think that the notations developed by Ray Birdwhistell and his colleagues at the Annenberg School of Communication would be more apt than Laban dance notation, because they concern the unconscious, culturally inherited expressive art form of ordinary human communication rather than a consciously cultivated art form. And of course Manfred Clynes makes claims about the underlying form of all communicative expression. Bruce Nevin bn@cch.bbn.com ------------------------------ Date: Sat, 02 Jul 88 16:05:09 +0100 From: "Gordon Joly, Statistics, UCL" Subject: Auto_Suggestion? > From AIList Vol 3 # 161 gcj> Date: Mon, 4 Nov 85 09:58:29 GMT gcj> From: gcj%qmc-ori.uucp@ucl-cs.arpa gcj> Subject: Vision Systems and American Sign Language gcj> gcj> One of goals of AI research is to produce speech recognition systems. gcj> Has there been a proposal to produce a vision system that can ``read'' gcj> ASL? gcj> gcj> Gordon Joly > From AIList Vol 4 # 49 ph> Date: 28 Jun 88 09:52 PDT ph> From: hayes.pa@Xerox.COM ph> Subject: Re: AIList Digest V7 #45 ph> ph> On dance notation: ph> A quick suggestion for a similar but perhaps even thornier problem: ph> a notation for the movements involved in deaf sign language. ph> I am not sure if Pat and I are really thinking of the same thing... Gordon Joly. Surface mail: Dr. G.C.Joly, Department of Statistical Sciences, University College London, Gower Street, LONDON WC1E 6BT, U.K. E-mail: | Tel: +44 1 387 7050 JANET (U.K. national network) gcj@uk.ac.ucl.stats | extension 3636 (Arpa/Internet form: gcj@stats.ucl.ac.uk) | Relays: ARPA,EAN: @nss.cs.ucl.ac.uk | CSNET: %nss.cs.ucl.ac.uk@relay.cs.net | BITNET: %ukacrl.bitnet@cunyvm.cuny.edu, @ac.uk EARN: @ukacrl.bitnet, @AC.UK By uucp/Usenet: ....!uunet!mcvax!ukc!stats.ucl.ac.uk!gcj ------------------------------ Date: 30 Jun 88 18:46:41 GMT From: bcsaic!douglas@june.cs.washington.edu (Douglas Schuler) Subject: Directions and Implications of Advanced Computing - DIAC-88 DIRECTIONS AND IMPLICATIONS OF ADVANCED COMPUTING DIAC-88 Twin Cities, Minnesota August 21, 1988 Earle Browne Continuing Education Center, University of Minnesota Advanced computing technologies are presented as instruments and images of both near and distant futures. Some of these futures radically challenge our conceptions of work, security, leisure, and common purpose. Will we be drawn into these futures as passive participants or will we actively select and shape alternative futures in our own interests? Few computing disciplines lie so directly at the intersection of these issues as does Artificial Intelligence. This summer thousands of computer professionals will descend on the Twin Cities for the annual conference of the American Association for Artificial Intelligence (AAAI). Sunday, August 21, the day before the AAAI conference, Computer Professionals for Social Responsibility (CPSR) will sponsor a one day symposium, "Directions and Implications of Advanced Computing." DIAC-88 aims to examine the social and political contexts of advanced computing, asking what futures are obtainable, for whom, and at what cost? Douglas Engelbart, the DIAC-88 plenary speaker, will share his perspective on using the computer to address global problems. Since the late 1950's, Engelbart has worked with systems that augment the human intellect including his NLS/Augment system, a hypertext system that pioneered "windows" and a "mouse." The driving force behind Engelbart's professional career has been his recognition of social impacts of computing technology. The plenary session will be followed by presentations of research papers and a panel discussion. The panel, John Ladd (Brown University), Deborah Johnson (Rens- salaer Polytechnic), Claire McInerney (College of St. Catherine) and Glenda Eoyang (Excel Instruction) will address the question, "How Should Ethical Values be Imparted and Sustained in the Computing Community?" Presented Papers Computer Literacy: A Study of Primary and Secondary Schools, Ronni Rosenberg Dependence Upon Expert Systems: The Dangers of the Computer as an Intellectual Crutch, Jo Ann Oravec Computerized Voting, Eric Nilsson Computerization and Women's Knowledge, Lucy Suchman and Brigitte Jordan Some Prospects for Computer Aided Negotiation, Douglas Schuler Computer Accessibility for Disabled Workers: It's the Law (invited paper) Richard E. Ladner Send symposium registration to: DIAC-88, CPSR/Los Angeles, P.O. Box 66038 Los Angeles, CA 90066-0038. Enclose check payable to CPSR/DIAC-88 with registration. For additional information, call David Pogoff, 612-933-6431. NAME ___________________________________________________ ADDRESS _________________________________________________ ________________________________________________________ ________________________________________________________ Phone (home) _____________________ (work) ______________________ Please check one: Symposium Registration Regular O $50 (Includes Proceedings and Lunch) CPSR Member O $35 Student/Low Income O $25 I cannot attend, but want the symposium proceedings O $15 There will a reception following the symposium. Proceedings will be distributed to registrants at the symposium. Non-attendees will receive proceedings by October 15, 1988. -- ** MY VIEWS MAY NOT BE IDENTICAL TO THOSE OF THE BOEING COMPANY ** Doug Schuler (206) 865-3226 [allegra,ihnp4,decvax]uw-beaver!uw-june!bcsaic!douglas douglas@boeing.com ------------------------------ Date: Fri, 1 Jul 88 09:10:39 EDT From: dlm@research.att.com Subject: talk announcement Title: Intermediate Mechanisms For Activation Spreading or Why can't neural networks talk to expert systems? Speaker:Jim Hendler University of Maryland Institute for Advanced Computer Studies University of Maryland, College Park Date: Tuesday, July 19 Time: 1:30 Place: AT&T Bell Laboratories - Murray Hill 3D-473 Abstract: Spreading activation, in the form of computer models and cognitive theories, has recently been under- going a resurgence of interest in the cognitive science and AI communities. Two competing schools of thought have been forming. One technique concentrates on the spreading of symbolic information through an associa- tive knowledge representation. The other technique has focused on the passage of numeric information through a network. In this talk we show that these two tech- niques can be merged. We show how an ``intermediate level'' mechanism, that of symbolic marker-passing, can be used to provide a limited form of interaction between traditional asso- ciative networks and subsymbolic networks. We describe the marker-passing technique, show how a notion of microfeatures can be used to allow similarity based reasoning, and demonstrate that a back-propogation learning algorithm can build the necessary set of microfeatures from a well-defined training set. We discuss several problems in natural language and plan- ning research and show how the hybrid system can take advantage of inferences that neither a purely symbolic nor a purely connectionist system can make at present. Sponsor: Diane Litman (allegra!diane) ------------------------------ End of AIList Digest ********************