Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!ukc!strath-cs!expya!jtr From: jtr@cs.exeter.ac.uk (Jason Trenouth) Newsgroups: comp.ai Subject: Re: Doug Lenat's "cyc" project Message-ID: Date: 21 Aug 90 09:36:43 GMT References: <1990Jul31.034417.19350@nixtdc.uucp> <1990Aug19.145236.1@csc.fi> <44091@apple.Apple.COM> Sender: news@cs.exeter.ac.uk Organization: Computer Science Dept. - University of Exeter. UK Lines: 73 In-reply-to: lantz@Apple.COM's message of 19 Aug 90 21:43:06 GMT I've just read the following article: D. B. Lenat, R. V. Guha, K. Pittman, D. Pratt, and M. Shepherd (1990) "CYC: towards programs with common sense" CACM 33(8) 30-49, August 1990. Cyc is a very large knowledge base whose language CycL has two levels: the epistemological level (EL) and the heuristic level (HL). The EL is first order predicate calculus augmented with reification (statements about statements) and reflection (assumptions support beliefs). This level exists to provide a simple semantics and an easy medium of communication with users. The HL is an optimisation layer that uses special purpose data structures and procedures do most of the inferencing. Cyc is being built by focusing on senarios (e.g. "buying something") and adding the knowledge that allows the Cyc system to answer common sense questions about them. *** I tend to agree with the philosophy behind Cyc for 3 reasons: <1> In addition to studying the individual pieces of the AI puzzle, the research establishment should also be funding work on the "big picture". Of course funding should probably be biased (for the moment) towards piecemeal projects since we don't seem to fully understand any of the components of AI. However, regular "stocking taking" is a good thing to do: can the current techniques/technologies fit together? The Cyc project (and the basic-agent projects) is (are) justifiable as _integration_ research. <2> The scale of Cyc also makes it unique. The largest knowledge based systems around have O(100000) assertions or rules, and the majority have only O(100)-O(1000). Lenat aims to keep building on Cyc until it has O(100000000) such objects. Clearly, this represents a _new_ test for each of the component technologies. So, the Cyc project is justifiable as _very large scale_ research. <3> The result of the Cyc project will be a resource that the whole AI community can benefit from. A very large public-domain knowledge-base can save researchers from perpetually re-inventing the wheel. Instead they will be able to concentrate on their particular interest e.g. planning, faster inference, etc. A common knowledge-base will allow better comparisons between these parasite projects. A common sense knowledge base can also be used to remove the bittleness from the current generation of expert systems. So, the Cyc project is justifiable as a _resource_ provider. *** Ultimately, the Cyc project is justifiable purely because it takes a _different approach_. In this sense I feel it is a bed-fellow of (:) Connectionism. What kind of self-respecting knowledge-base-system designer would entertain building a inference mechanism that only searched from one direction?! Equally, for efficiency at the methodological level, a certain amount of pluralism is "a Good Thing": Research Table TOP-DOWN BOTTOM-UP DEPTH-FIRST Mainstream Symbolic AI Current Connectionism BREADTH-FIRST Cyc Project Future Connectionism? In other words: there is room for Cyc. Ciao - JT -- ______________________________________________________________________________ | Jason Trenouth, | JANET: jtr@uk.ac.exeter.cs | | Comp. Sci. Dept., Exeter Univ., | UUCP: jtr@expya.uucp | | Devon, EX4 4PT, UK. TEL: (0392) 264061 | BITNET: jtr%uk.ac.exeter.cs@ukacrl|