Date: Fri 12 Aug 1988 00:07-EDT From: AIList Moderator Nick Papadakis Reply-To: AIList@mc.lcs.mit.edu Us-Mail: MIT LCS, 545 Tech Square, Rm# NE43-504, Cambridge MA 02139 Phone: (617) 253-6524 Subject: AIList Digest V8 #44 To: AIList@mc.lcs.mit.edu Status: R AIList Digest Friday, 12 Aug 1988 Volume 8 : Issue 44 Today's Topics: Spang Robinson Reports Will computers dominate chess? (EURISKO) ---------------------------------------------------------------------- Date: Wed, 3 Aug 88 21:23:47 CDT From: smu!leff@uunet.UU.NET (Laurence Leff) Subject: bm940 Summary of Spang Robinson Report on Artificial Intelligence Volume 4, No. 6, June 1988 The lead article is on research directions. Randy Davis at MIT is developing deep-knowledge based systems for dealing with relationships between devices. The work focusses on digital circuits to do circuit design, test generation and diagnosis. AI Squared is a new company using this technology for medical instrumentations. They are delivering a system for CAT Scanners called FELIX. The article discusses Xerox Parc, how they bring such disciplines as anthropology and psychology into AI efforts and study how people "actually do design work." Xerox Parc is also looking at office system to keep track of office documents. Price Waterhouse is doing research into auditing, tax planning and consulting with a Big Eight accounting systems. Richard Fikes, now at Price Waterhouse, is working on aprojectin international corporate tax planning. They are also working on integrating textual material that does notfit into a structureddomain model, but which is applicable, using a hypertext-like technique. Lockheed is adding spatial and temporal systems to their expert system tool, LES. They are also working on result explanation by means other than backtracking through rules and rule-base validation. ________________________________________ Neural Networks: Robert Hecht-Neilsen, et. al. will have a proof that a three layer Back Error Propagation neural network will always converge,under certain conditions. Conditions are the use of 32 bit floating point math, a square integrable mapping function and the mapping regions must be compact and bounded. Stephen Gallant has constructed a mechanism for neural-network explanation. It uses an input vector with only three values (false, unknown and true) and is faster than Back error Propagation. The system has been patented. ++++++++++++++++++++++++++++++++++++++++ DARPA has appropriated 60 million for AI research, which is 80 percent of total AI research. Jack Schwartz, the new DARPA Information Science and Technology Office director, is favoring AI research which lasts two to three years and has clearly definable results. Areas like logic and those needing "intensive computation" are "considered overly ambitious." Cuts of between ten and thirty three percent are expected for AI research. There will be emphasis on robotics and algorithms including AI. ________________________________________ Shorts: Gold Hill Computers let go 20 out of 105 employees, most in sales. 1988 sales flat after a tripling in 1987. No cuts in development staff. Financia is a new company in England that will develop packaged PC Expert systems to advise in Equities and Futures markets. Intellicorp has been selected to be an Autorized Marketing Aid for IBM RT's. Geosource and Knowledge Systems are joining forces to develop and market geophysical and geological applications for the energy industry. Coopers and Lybrand has created Insurance ExperTAX that helps Insurance companies identifying tax accrual issues and tax planning opportunities. Neuron Data announced that its product now runs on HP 9000 series 300 and series 800 technical work stations. Inference has made its ART available for the TI MicroExplorer and Sun 4. ------------------------------ Date: Mon, 8 Aug 88 11:12:13 CDT From: smu!leff@uunet.UU.NET (Laurence Leff) Subject: bm954 Summary of Spang Robinson Report on Artificial Intelligence, July 1988, Volume 4, No 7 Lead Article is on Knowledge-Based System Methodology and teaching of same. It describes training efforts at various firms such as IBM, TI, DEC and accounting firms. Some of these training programs provide "automated methodologies" and Arthur D. Little provides automated assistants for these items. Some include sample systems, e. g. Cullinet's database performance analyzer. (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( Learning Systems System Technique Price VP-Expert Kavanaugh Maps $124.95 Mac Smarts Kavanaugh Maps $250.00 Super Expert ID3 $195.00 Rule Master ID3 $495.00 generates C code KnowledgeMaker ID3 $ 95.00 generates Prolog roles, M. 1, Insight 2 1st Class ID3 $495.00 tree can be edited Fusion ID3 $1295.00 produces C or pascal code IXL ID3 $495.00 uses statistical methods to predict relationships, produces confiedence factors Beagle genetic $200.00 produces Fortran, Cor pPascal learning DuPont has used 1st Class and VP-Expert. An example-based protottype for a Mylar manufactuirng machine was up in an afternoon after conventional rule-based aporaches failed. An insurance company achieved expert level performance in two weeks, 400 examples and is now in "beta" ()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()() AI Software/International Marketing. Crystal is a system that supports an inductive approach and costs about a thousand dollars. Systems Designer Internation sells a SD-Prolog for $499. Gold Hill percentage of sales in: Japan 12 percent Europe 10 percent *(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*(*( Shorts: Hect-Nielsen (neural networks) received 3 million dollars in second financing round. Gensym and GigaMos settled their lawsuit confidentially. Survey of large financial servies show 43 percent doing something regarding expert systems. Banks have 60 percent. Texas Instruments will merge the Data Systems Group to the Comptuer systems Group. AI activities were in the former. Gensym will offer G2 on HP 9000 system. (real time xpert system) Lucid has joint marketing agreement to sell products in Japan. Aion Corporation and Cincom systems have cooperative marketing agreement. Intelligent Technology will be distributing ClienTrak relationship management system. It manages Key sales activities. A bridge between for V. I. Dataviews interactive graphics and Neuron NEXPERT object will be developed by the companies. Canadian Artificial Intelligence Products has received grant from Telecom Canada to develop hypertext system. ------------------------------ Date: Mon, 8 Aug 88 11:42:58 CDT From: smu!leff@uunet.UU.NET (Laurence Leff) Subject: bm953 Summary of Spang Robinson Report on Supercomputing and Parallel Prcessing July 1988, Volume 2, No. 7 Lead Issue is on "Linda" 14 VAXen losely coupled outperformed CRAY-1 on a Compute-Intensive task at Sandia. Linda is a Yale-developed package being enhanced by Scientific Computing Associates. They have versions for Encore, Sequent. Implementations have been developed for shared memory systems rimarily but can be run on INtel Scinetific systems. Implementations of Linda exist for Fortran, Lisp and Pascal. U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7U7 Applied Intelligent systm has developped a masive machine vision systems using a proprietary chip featuring eight single-bit processes. It is similar to the Thinking machine. The system ranges from a one board system with sixty-four processing elements to one with 1024 processor elements. A new system will have a 32-processor chips. A new system will have 10 MFLOP system at $500.00. They also have a product called LAYERS which provides an object-oriented C-based system. ()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()()() The next article is on the Kartashevs who hold various Supercumputing conferences. However, some people find the "kartashev style" abrasive. A competing conference, "Supercomputing World 1989" is a reaction against the system. %^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^%^% Shorts: Cray received orders for its CRAY 2S from: National Center for Supercomputer Applications, University of Illinois National Test Bed Facility at Falcon Airforce installation Cray YMP Ohio State University Ohio Supercomputer Center Shell Research B. V., Exploration Lab, Netherlands Saxpy will be selling its technology and assets. Sequent will be joining with Franz and Quintus to offer versions of LISP and Prolog for their system. Parasoft will be reselling DEfinicon's add-in Transputer boards. Floating Point System reported a quarterly loss of 7.3 million. Multiflow layed off 25 person in manufacturing. MASSCOMP reported $250,000 revenues. ETA announces "native UNIX" System V on its system. Encore announced a fully parallel ADA for the Multimax configuration. It received the fastest completion time for the Ada Validation Suite on record. Sequent announced a version of X-Windows for its systems. Pacific Cyber/Metrix announced a 250 MIPS VMEBus Data flow machine. The basisc system consisiting of four processors costs $20,300. Japanese government has announced a 640 MFLOP dataflow system. It has 128 processors and 128 microprocessors to data read/store. ------------------------------ Date: 3 Aug 88 18:17:44 GMT From: Martin-Charles@yale-zoo.arpa (Charles Martin) Subject: Re: Will computers dominate chess? (was Re: Computers & Chess) In article <35187@aero.ARPA>, srt@aero (Scott R. Turner) writes: Lenat's EURISKO program was innovative enough in Starfleet Battles that it was eventually barred from tournament play - after having invented a new winning strategy two years running. Surely you mean /Trillion Credit Squadron/. Also, my impression is that EURISKO was not solely responsible for the strategies; considerable editing was required by Lenat, adjustment of weights, etc. I believe he cited some figure such as 60/40 Lenat/EURISKO, which if nothing else at least reflects his own estimation of the limitations of this program applied to this task. TCS, while requiring large amounts of data for the various weapon and defensive systems, is an extremely simple game. It is the large amount of data which makes it difficult for humans to "grasp" the game. The TCS system was designed to be simple---as the previous three-dimensional game of maneuver was too complex for people to play with more than a couple of ships. Concepts at the level of "fork," basic to tic-tac-toe and chess, do not play a role in TCS. The EURISKO line of research was not pursued (as far as I am aware) into more complex games with less human intervention. Charles Martin // INTERNET: martin@cs.yale.edu // BITNET: martin@yalecs UUCP: {cmcl2,harvard,decvax}!yale!martin ------------------------------ Date: 3 Aug 88 19:47:36 GMT From: att!alberta!jonathan@bloom-beacon.mit.edu (Jonathan Schaeffer) Subject: Re: computer chess In article <376@ksr.UUCP>, richt@breakpoint.ksr.com (Rich Title) writes: > There's a Carnegie Mellon PhD thesis by Carl Eberling, > that was published (by MIT press > I think) under the title "All the Right Moves". It describes HiTech, > the current world computer chess champion. That thesis in turn > points to other papers on computer chess. Hitech is NOT the World Computer Chess Champion. In the last championship in 1986, there was 1 4-way tie for first place between Cray Blitz, Hitech, Bebe, and Phoenix. Cray Blitz was awarded first place on tiebreak. "All the Right Moves" is a good thesis, but is not the best place to look for references. The International Computer Chess Journal is published quarterly with the latest in research results, tournaments, games, etc. That is the best place to look. Also, several computer chess bibliographies have been published. Perhaps the most comprehensive, albiet slightly out of date, is Tony Marsland's (available as a technical report from the University of Alberta). > Carnegie Mellon seems to be *the* place for computer chess. > Hans Berliner, former postal chess champion, is a comp sci > professor there. CMU is only one of a number of places with active computer chess groups. Others include University of Alberta, McGill University, University of Limburg, Bell Labs, Los Alamos National Lab, etc. > The techniques used in the top machines such as HiTech represent > impressive engineering, but aren't what most people think of > as "AI". Very fast searching, aided by hardware that generates > and evaluates moves in parallel and evaluates positions > in parallel. True, but that is not all the things people are doing in computer chess. As it stands right now, the strongest chess playing machines are more engineering than science. But do not underestimate the scientific component of computer chess. A lot of this work may not be high profile unless it is incorporated as part of a winning chess program, but it is still important, core AI research. > - Rich - Jonathan ------------------------------ End of AIList Digest ********************