mostow@fokker.rutgers.edu (Jack Mostow) (02/22/89)
NOBEL LAUREATE HERBERT SIMON TO VISIT RUTGERS Nobel laureate DR. HERBERT SIMON, Richard King Mellon University Professor of Computer Science and Psychology at Carnegie-Mellon University, will present two special Computer Science Department Colloquia at Rutgers University on Feb. 23. Dr. Simon, one of the founders of artificial intelligence, has been honored by such diverse bodies as the American Psychological Association, the Association for Computing Machinery, the American Political Science Association, the American Economic Association, and the Institute of Electrical and Electronic Engineers for his research on human decision-making and problem-solving and their implications for social institutions. He has published over 600 papers and 20 books. A member of the National Academy of Sciences since 1967, he received the Alfred Nobel Memorial Prize in Economic Sciences in 1978 and the National Medal of Science in 1986. Dr. Simon has been Chairman of the Board of Directors of the Social Science Research Council, and of the Behavioral Science Division of the National Research Council, and was a member of the President's Science Advisory Committee. "LOGIC PROGRAMMING: A WRONG ROAD FOR AI" Time: 10:30am Thursday, February 23 Place: Hill Center 114, Busch Campus, Rutgers University Audience: This talk assumes a graduate level computer science background. Abstract: Logic programming derives from the metaphor of formal logic, a technology for verifying statements rigorously rather than for discovering them, which is necessarily inefficient for the latter purpose. Logic programming, in languages like PROLOG, discourages the proliferation of inference rules (aka "operators"); but effective search and discovery algorithms use such rules freely, including rules that are not analytic but incorporate assumptions from the subject-matter domain. The discouragingly slow progress of automatic theorem proving can be attributed to the adherence to the logic metaphor rather than the more satisfactory heuristic search metaphor. STRIPS is an interesting example of the abandonment of logic programming for heuristic search in the face of realistic problems. "SCIENTIFIC DISCOVERY AS COMPUTATION" Time: 2:50pm Thursday, February 23 Place: Hill Center 114, Busch Campus, Rutgers University Audience: This talk is aimed at a general audience interested in artificial intelligence and the process of scientific discovery. Abstract: The past decade has seen the creation of a substantial number of AI programs that are capable of making discoveries at a non-trivial (professional) level. Such programs include Meta-Dendral, AM and EURISKO, BACON and its associates (DALTON, GLAUBER, STAHL), and KEDADA. From these programs, we have learned much about the heuristics required for discovery, including heuristics for searching spaces of functions, heuristics for exploiting surprise, and heuristics for inventing new concepts. The prospects will be discussed of extending these advances to all major types of scientific activities, including the invention and design of problem representations. Dr. Simon's host is Asst. Prof. Jack Mostow of the Computer Science Department. For more information, contact Ms. Carol Petty at 932-2928.
jack@cs.glasgow.ac.uk (Jack Campin) (03/01/89)
mostow@fokker.rutgers.edu (Jack Mostow) wrote: > [Herb Simon:] The past decade has seen the creation of a substantial number > of AI programs that are capable of making discoveries at a non-trivial > (professional) level. Such programs include Meta-Dendral, AM and EURISKO, > BACON and its associates (DALTON, GLAUBER, STAHL), and KEDADA. I am by no means up-to-date on AI, but I don't believe this. AM never did more than rediscover trivialities. Have any of the others made any discovery that could get into a refereed journal on its content rather than curiosity value? Has anyone ever used them for more than routine drudgework? Ever seen one of them cited in "Cell" or "Physics Review Letters" as making an essential contribution to an experimental design? Has any of them come up with a proof strategy subsequently used in a real mathematical paper? "Capable" is the sort of jam-tomorrow weasel-word we've had from Simon and his ilk for the last thirty years. What have these things *actually achieved*? -- Jack Campin * Computing Science Department, Glasgow University, 17 Lilybank Gardens, Glasgow G12 8QQ, SCOTLAND. 041 339 8855 x6045 wk 041 556 1878 ho INTERNET: jack%cs.glasgow.ac.uk@nss.cs.ucl.ac.uk USENET: jack@glasgow.uucp JANET: jack@uk.ac.glasgow.cs PLINGnet: ...mcvax!ukc!cs.glasgow.ac.uk!jack
jbn@glacier.STANFORD.EDU (John B. Nagle) (03/04/89)
See "Why AM and Eurisko Appear to Work", by Lenat and Brown, in Huberman's "Computational Ecology" work. This is an important paper, and explains why AM seemed to do so well at first but eventually hit a wall. They say it best: "Although we generally described it as 'exploring in the space of math concepts', what it really was doing from moment to moment was "syntactically mutating small LISP programs'. Rather than disdaining it for that reason, we saw that that was its salvation, its chief source of power, the reason that it had such a high hit rate; AM was exploting the natural tie between LISP and mathematics." John Nagle
robinson@pravda.gatech.edu (Stephen M. Robinson) (03/07/89)
In article <2493@crete.cs.glasgow.ac.uk> jack@cs.glasgow.ac.uk (Jack Campin) writes: >mostow@fokker.rutgers.edu (Jack Mostow) wrote: >> [Herb Simon:] The past decade has seen the creation of a substantial number >> of AI programs that are capable of making discoveries at a non-trivial >> (professional) level. Such programs include Meta-Dendral, AM and EURISKO, >> BACON and its associates (DALTON, GLAUBER, STAHL), and KEDADA. >I am by no means up-to-date on AI, but I don't believe this. .... > .... What have these things *actually achieved*? >Jack Campin * Computing Science Department, Glasgow University, 17 Lilybank BACON, GLAUBER, DALTON and STAHL are programs designed not to discover new things but rather to show that it is possible to account for scientific discovery computationally without having to make appeal to some mysterious "intuition," especially when run cooperatively. See the book _Scientific Discovery_ by Langley, Simon and Bradshaw, 1987. These programs are tools for studying the process of discovery more than programs designed to make discoveries which are completely new to their domains. Stephen M. Robinson AI Group School of Information and Computer Science Georgia Institute of Technology Atlanta, GA 30332 InterNet: robinson@pravda.gatech.edu UUCPNet: ...!{uiucdcs,akgua,allegra,amd,hplabs,ihnp4,seismo,ut-ngp} !gatech!pravda!robinson Phone: (404)894-8932
hundt@paul.rutgers.edu (Thomas M. Hundt) (03/08/89)
---------- >I am by no means up-to-date on AI, but I don't believe this. .... > .... What have these things *actually achieved*? BACON, GLAUBER, DALTON and STAHL are programs designed not to discover new things but rather to show that it is possible to account for scientific discovery computationally without having to make appeal to some mysterious "intuition," especially when run cooperatively. See the book _Scientific Discovery_ by Langley, Simon and Bradshaw, 1987. These programs are tools for studying the process of discovery more than programs designed to make discoveries which are completely new to their domains. ---------- Right, Simon's theme in his second talk was basically that it doesn't take genius to be a "genius". Or more precisely, any intelligent person following a certain logical series of steps may arrive at an important discovery, given the underlying information and a given stimulus. F'r instance, there was Fleming and his mold that killed bacteria. Now, here was a guy who knew a lot about bacteria, and knew what to expect. So, he found that on a Petri dish he didn't bother to clean ("sometimes you find interesting things that way") a mold was killing some bacteria. This got his attention, because it did not fit in with his expectations. He asked questions: is it a particular type of bacteria? What about the mold? etc. Logical questions. Generalize the occurrance, use experiments to find the answers. Draw new conclusions. Design new experiments. Etc. Simon's example programs showed that is possible to mechanize these steps, at least to some degree. . . . He also mentioned that to become a World Class Scholar (a big term these days at Rutgers, where they're always hiring one or another; we mis-pronounce the abbreviation "wixel"), one needs to know a *lot* about one's field, 50000 or so facts, where a "fact" is about equivalent to knowledge about an English word. Most people have vocabularies in the range of 50000 words, to give an idea. The other thing you need is 10 years intense use of that knowledge, in one's field of expertise. -- w ["] | Thomas M. Hundt :: hundt@occlusal.rutgers.edu | |__'_ | Gradual Student :: Electrical & Computer Eng. | H \/| Rutgers University :: 201/932-5843 | X | 272 Hamilton St. #96 :: 201/247-6723 | _/ \_ | New Brunswick, NJ 08901 "Limit guns not speed" |