anies risen to meet it, they have more to gain than to lose by mutual support. Secondly, excluded companies can always set up a rival association. This association needs a code of practice, which members would agree to adhere to and which would serve as a basis for refusing membership. What form should such a code take, i.e. what counts as malpractice in AI? I suspect malpractice may be a lot harder to define in AI than in insurance, or medicine, or travel agency. Due to the state of the art, AI products cannot be perfect. No-one expects 100% accurate diagnosis of all known diseases. On the other hand a program which only works for slight variations of the standard demo is clearly a con. Where is the threshold to be drawn and how can it be defined? What consitutes an extravagent claim? Any product which claims to: understand any natural language input, or to make programming redundant, or to allow the user to volunteer any information, sounds decidedly smelly to me. Where do we draw the line? I would welcome suggestions and comments. Alan Bundy ------------------------------ Date: 22 Jun 84 6:44:56-EDT (Fri) From: hplabs!tektronix!uw-beaver!cornell!vax135!ukc!west44!greenw @ Ucb-Vax.arpa Subject: Human models Article-I.D.: west44.243 [The time has come, the Walrus said, to talk of many things...] Consider... With present computer technology, it is possible to build (simple) molecular models, and get the machine to emulate exactly what the atoms in the `real` molecule will do in any situation. Consider also... Software and hardware are getting more powerful; larger models can be built all the time. [...Of shoes and Ships...] One day someone may be able to build a model that will be an exact duplicate of a human brain. Since it will be perfect down to the last atom, it will also be able to act just like a human brain. i.e. It will be capable of thought. [...And Sealing Wax...] Would such an entity be considered `human`, for, though it would not be `alive` in the biological sense, someone talking on the telephone to its very sophisticated speech synthesiser, or reading a letter typed from it would consider it to be a perfectly normal, if not rather intelligent person. Hmmmmmm. One last thought... Even if all the correct education could be given it, might it still suffer from the HAL9000 syndrome [2001]; fear of being turned off if it did something wrong? [...of Cabbages and Kings.] Jules Greenwall, Westfield College, London, England. from... vax135 greenw (UNIX) \ / mcvax- !ukc!west44! / \ hou3b westf!greenw (PR1ME) The MCP is watching you... End of Line. ------------------------------ Date: 18 Jun 84 13:27:47-PDT (Mon) From: hplabs!hpda!fortune!crane @ Ucb-Vax.arpa Subject: Re: A Quick Question - Mind and Brain Article-I.D.: fortune.3615 Up to this point the ongoing discussion has neglected to take two things into account: (1) Subconscious memory - a person can be enabled (through hypnosis or by asking him the right way) to remember infinite details of any experience of this or prior life times. Does the mind selectively block out trivia in order focus on what's important currently? (2) Intuition - by this I mean huge leaps into discovery that have nothing to do with the application of logical association or sensual observation. This kind of stuff happens to all of us and cannot easily be explained by the physical/mechanical model of the human mind. I agree that if you could build a computer big enough and fast enough and taught it all the "right stuff", you could duplicate the human brain, but not the human mind. I don't intend to start a metaphysical discussion, but the above needs to be pointed out once in a while. John Crane ------------------------------ Date: Wed 20 Jun 84 10:01:39-PDT From: WYLAND@SRI-KL.ARPA Subject: The Turing Test - machines vs people Tony Robison (AIList V2 #74) and his comments about machine "soul" brings up the unsettling point - what happens when we make a machine that passes the Turing test? For: o One of the goals of AI (or at least some workers in the field - hedge, hedge) is to make a machine that will pass the Turing test. o Passing the Turing test means that you cannot distinguish between man and machine by their written responses to written questions (i.e., over a teletype). Today, we could extend the definition to include oral questions (i.e., over the telephone) by adding speech synthesis and recognition. o If you cannot tell the difference between person and machine by the formal social interaction of conversation, *how will the legal and social systems differentiate between them!!* Our culture(s) is set up to judge people using conversation, written or oral: the legal arguments of courts, all of the testing through schools, psychological examination, etc. We have chosen the capability for rational conversation (including the potential capability for it in infants, etc.) as the test for membership in human society, rejecting membership based on physical characteristics such as body shape (men/women, "foreigners") and skin color, and the content of the conversations such as provided by cultural/ religious/political beliefs, etc. If we really do make machines that are *conversationally indistinguishable* from humans, we are going to have some interesting social problems, whether or not machines have "souls". Will we have to reject rational conversation as the test of membership in society? If so, what do we fall back on? (The term "meathead" may become socially significant!) And what sort of interesting things are going to happen to the social/legal/religious systems in the meantime? Dave Wyland WYLAND@SRI P.S. Asimov addressed these problems nicely in his renowned "I, Robot" series of stories. ------------------------------ Date: 18 Jun 1984 14:21 EDT (Mon) From: Peter Andreae Subject: Seminar - Precondition Analysis [Forwarded from the MIT bboard by SASW@MIT-MC.] PRECONDITION ANALYSIS - LEARNING CONTROL INFORMATION Bernard Silver Dept. of AI, University of Edinburgh 2pm Wednesday, June 20. 8th Floor Playroom I will describe LP, a program that learns equation solving strategies from worked examples. LP uses a new learning technique called Precondition Analysis. Precondition Analysis learns the control information that is needed for efficient progblem solving in domains with large search spaces. Precondition Analysis is similar in spirit to the recent work of Winston, Mitchell and DeJong. It is an analytic learning technique, and is capable of learning from a single example. LP has successfully learned many new equation solving strategies. ------------------------------ End of AIList Digest ******************** 21-Jun-84 22:18:40-PDT,13115;000000000001 Mail-From: LAWS created at 21-Jun-84 22:12:56 Date: Thu 21 Jun 1984 22:03-PDT 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 V2 #76 To: AIList@SRI-AI AIList Digest Friday, 22 Jun 1984 Volume 2 : Issue 76 Today's Topics: VLSI - Panel on Chips for AI & Trilogy CPU Failure, Databases - Oxford English Dictionary goes On-Line, Logic - Common Sense Summer, Mind & Brain - Artificial People & Neural Connections & Recall, Seminar - Natural Language Parsing ---------------------------------------------------------------------- Date: 20 June 1984 0512-EDT From: Dave Touretzky at CMU-CS-A Subject: panel on chips for AI [Forwarded from the CMU bboard by Laws@SRI-AI.] Dana Seccombe is looking for people to participate in a panel discussion at ISSCC (International Solid State Circuits Conference) to be held in February '85 in New York City. The topic of the panel is issues in the realization of AI systems using VLSI technology, e.g. AI inference engines, 5th generation architectures, or Lisp processors that are or could be implemented using VLSI. If you would be interested in participating in this panel, please contact Mr. Seccombe at (408) 257-7000 x4854. DON'T contact me, because I don't know any more about it than what you've just read. ------------------------------ Date: 19 Jun 1984 11:07:46-EDT From: Doug.Jensen at CMU-CS-G Subject: Trilogy CPU design fails [Forwarded from the CMU bboard by Laws@SRI-AI.] After 4 years and $220 million, Gene Amdahl's Trilogy Corp. has declared their attempt to build a computer from 2.5" diameter whole wafer VLSI a failure. They never got even one wafer functioning correctly much less ever powered up a machine. Trilogy thus follows in the path of TI and many other whole wafer failures before them over the past decade; the others were less known because they were military projects. Trilogy was one of, and probably THE, most publicized and heavily funded new startup in the history of the computer business. They were spending $7 million/month and estimated that they would need at least another $100 million just to get them to mid-85, while their first machine was still two years beyond that (more than two years later than they estimated when they started in 1980). Each 2.5" wafer was to contain about 60K ECL gates, with four layers of metalization, and dissipate about 1000 watts. The CPU was to have nine wafers and excute 32 MIPS. Trilogy was even further behind on the other computer subsystems. They now say they may try a smaller machine, or just subsystems (e.g., memories), or just wafers and related technology. DEC, Sperry, and CII-HB were among the investors in Trilogy. ------------------------------ Date: 13-Jun-84 02:30 PDT From: William Daul Augmentation Systems Division / MDC Subject: Oxford English Dictionary goes On-Line [Forwarded from the Human-Nets Digest by Laws@SRI-AI.] LONDON -- ...the Oxford University Press has announced plans to publish a computerized version of the venerable Oxford English Dictionary. With the help of a $1.4 million donation from IBM United Kingdom Ltd., the British publisher will produce the first fully integrated edition of the 13-volume dictionary since the original work was begun in 1884. That first edition took 44 years to complete; the publisher said it will be able to complete the second edition in a fraction of that time. ... The New Oxford English Dictionary, as the new version has been named, will constitute the largest electronic dictionary data base in the world. The present multi-volume version consists of more than 20,000 printed pages. Computerization of the dictionary is a massive undertaking that will involve the data entry of about 60 million words used to record, describe and illustrate 500,000 words and phrases. The Oxford University Press has hired International Computaprint Corp. of Fort Washington, Pa., to do the data entry. A staff of 120 people has been assigned the task of completing the data entry by this September. ... Additionally, the company (IBM) is providing two data processing specialists who will work on the dictionary project for two years. Once the electronic dictionary is finished, it could be made available on-line, on magnetic tape, on laser/video disk or possible, on a single integrated circuit... The publisher estimated the project will cost $10 million. The British government awarded the company a 3 year grant of roughly $420,000 -- or 25% of the development cost -- for the dictionary. The University of Waterloo in Ontario will conduct a survey for the publisher of the potential users of an electronic dictionary. The university will also help develop software that would be needed to take advantage of an electronic dictionary. ------------------------------ Date: Wed 20 Jun 84 22:06:12-PDT From: Dikran Karagueuzian Subject: Newsletter, June 21, No. 37 [Forwarded from the CSLI Newsletter by Laws@SRI-AI.] COMMON SENSE SUMMER CSLI is sponsoring a summer-long workshop called "Common Sense Summer." It has long been agreed that language use and intelligent behavior in general require a great deal of knowledge about the commonsense world. But heretofore no one has embarked on a large-scale effort to encode this knowledge. The aim of Common Sense Summer is to make the first three months of such an effort. We are attempting to axiomatize in formal logic significant amounts of commonsense knowledge about the physical, psychological and social worlds. We are concentrating on eight domains: shape and texture, spatial relationships, lexical semantics of cause and possession, properties of materials, certain mental phenomena, communication, relations between textual entities and entities in the world, and responsibility. We are attempting to make these axiomatizations mutually consistent and mutually supportive. We realize, of course, that all that can be accomplished during the summer is tracing the broad outlines of each of the domains and, perhaps, discovering several hard problems. Nine graduate students from several universities are participating in the workshop full-time. In addition, a number of other active researchers in the fields of knowledge representation, natural language, and vision are participating in meetings of various sizes and purposes. There will be two or three presentations during the summer, giving progress reports for the general public. The workshop is being coordinated by the writer. --Jerry Hobbs [...] ------------------------------ Date: Wed, 20 Jun 84 17:25:03 PDT From: Charlie Crummer Subject: Human Models The foundation of the reasoning constructed by Jules Greenwall in his note depends on being able to specify exactly the behavior of atoms in molecules. The precise description required depends on the molecular physics. Unfor- tunately study is still going on. The study of the molecule is a many-body problem for which there is no closed-form solution. Another fly in the ointment is the fact that the behavior of atoms in molecules depends, albeit in second order, on the nature of the nucleus. This is another branch of physics that is very active, i.e. much is not known. What one would get for a model built on such a fuzzy foundation is of dubious value. --Charlie ------------------------------ Date: 18 Jun 84 10:07:07-PDT (Mon) From: hplabs!hao!seismo!rlgvax!cvl!umcp-cs!aplvax!lwt1 @ Ucb-Vax.arpa Subject: Re: A Quick Question - Mind and Brain Article-I.D.: aplvax.663 The other thing to note is that while each 'memory cell' in a computer has ~2 connections, each 'memory cell' in the brain has ~100. Since processing power is relative to (cells * connections), a measure of relative capacities is not sufficient for comparison between the brain and the CRAY. -Lloyd W. Taylor ... seismo!umcp-cs!aplvax!lwt1 ---I will have had been there before, soon--- ------------------------------ Date: Thu, 21 Jun 84 06:39 EDT From: dmrussell.pa@XEROX.ARPA Subject: Objection to Crane: A Quick Question - Mind and Brain -- V2 Sorry, but I must make a serious objection to your claim that "... a person can be enabled (through hypnosis or by asking him the right way) to remember infinite details of any experience of this or prior life times ... " I object to the use of the term "infinite" in describing memory. That simply isn't true. If you just mean "large number", then say so. The infinite memory capacity problem was addressed once (in either AIDigest or HumanNets, I've forgotten) and found indefensible. The phrase "prior life times" assumes reincarnation, a completely unsupported assumption. "of any experience" demands that all experiences can be recalled, not just *recognized*, or *restored* but recalled! Do you really want the references to show that this isn't true? Memory recall under hypnosis has been found to be just as reconstructive (perhaps more so) as normal memory. Hypnotic states buy you some recall, but not that much! We haven't taken these things into account because they simply aren't true, or at the very least, can't be supported by anything other than religious belief. -- D.M. Russell. -- ------------------------------ Date: 18 Jun 84 15:08:10-PDT (Mon) From: ihnp4!ihldt!stewart @ Ucb-Vax.arpa Subject: Re: A Quick Question - Mind and Brain Article-I.D.: ihldt.2382 > (1) Subconscious memory - a person can be enabled (through > hypnosis or by asking him the right way) to remember > infinite details of any experience of this or prior life > times. I don't know where the "prior life" part came from, but this claim is usually an incorrect extrapolation of studies that indicate no such thing. What has been established is that people can be induced to remember things that they considered forgotten. This isn't by a long shot the same thing as saying that we remember everything that's ever happened to us. If you have evidence to support this claim, by all means present it. If not, please spare us. Bob Stewart ihldt!stewart ------------------------------ Date: Thu, 21 Jun 84 08:23 EDT From: Dehn@MIT-MULTICS.ARPA (Joseph W. Dehn III) Subject: Turing test - legal implications ...computers someday might act like people... ...legal system is based on capability for rational conversation... ...what will we do????... ...will we have to reject rational conversation as the test of membership in society?... Sorry, I must have forgotten, but why exactly do we WANT to distinguish between humans and machines? -jwd3 ------------------------------ Date: Thu, 21 Jun 84 14:14 EST From: Huhns Subject: Seminar - Natural Language Parsing CONSTRAINT PROPAGATION SENTENCE PARSING Somnuek Anakwat Center for Machine Intelligence College of Engineering University of South Carolina 2pm Thursday, June 21, Room 230 An algorithm for parsing English sentences by the method of constraint propagation is presented. This method can be used to recognize English sentences and indicate whether those sentences are syntactically correct or incorrect according to grammar rules. The central idea of constraint propagation in sentence analysis is to form all possible combinations of the parts of speech from adjacent words in the input sentence, and then compare those combinations with English grammar rules for allowable combinations. The parts of speech for each word may be modified, left alone, or eliminated according to these rules. The analysis of these combinations of the parts of speech normally proceeds from left to right. The most significant feature of the algorithm presented is that grammar constraints propagate backward when it is possible. The algorithm is very useful when the given sentence contains words which have multiple properties. The algorithm also has an efficient parallel implementation. Results of applying the algorithm to several English sentences are included. An interpretation of the algorithm's performance and some topics for future research are discussed as well. ------------------------------ End of AIList Digest ******************** 22-Jun-84 05:39:05-PDT,15944;000000000001 Mail-From: LAWS created at 22-Jun-84 05:34:49 Date: Fri 22 Jun 1984 05:12-PDT 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 V2 #77 To: AIList@SRI-AI AIList Digest Friday, 22 Jun 1984 Volume 2 : Issue 77 Today's Topics: AI Tools - Q'NIAL, Cogniti