From will@aristotle.ils.nwu.edu Fri Jul 8 19:04:42 EDT 1994 Article: 23012 of comp.ai Xref: glinda.oz.cs.cmu.edu comp.ai:23012 Path: honeydew.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!news.cac.psu.edu!news.pop.psu.edu!psuvax1!news.ecn.bgu.edu!anaxagoras.ils.nwu.edu!aristotle.ils.nwu.edu!will From: will@aristotle.ils.nwu.edu (William Fitzgerald) Newsgroups: comp.ai Subject: Cyc: An annotated bibiography Date: 7 Jul 1994 19:23:53 GMT Organization: The Institute for the Learning Sciences Lines: 168 Message-ID: <2vhko9$sc6@anaxagoras.ils.nwu.edu> NNTP-Posting-Host: aristotle.ils.nwu.edu X-Newsreader: TIN [version 1.1 PL9] An annotated bibliography on the Cyc project. Will Fitzgerald (will@ils.nwu.edu) The Institute for the Learning Sciences July, 1994 1984: The beginnings of Cyc. Lenat, D., Prakash, M., & Shepard, M. (1986). CYC: Using common sense knowledge to overcome brittleness and knowledge acquisition bottlenecks. AI Magazine, 6(4), 65-85. This is the original article introducing Cyc to the AI research community. It states that the overall goal of Cyc is the building, "over the coming decade, of a large knowledge base (or KB) of real world facts and heuristics and--as a part of the KB itself--methods for efficiently reasoning over the KB." The more specific goals and milestones include: - building a "documented, debugged representation language, adequate for representing the spectrum of material contained in--and assumed by--[a desk] encyclopedia." - building "an interface suitable for browsing and editing" Cyc - have Cyc be able to do analogical reasoning - building representations of each type of knowledge required of a common sense reasoner. - building a system that a "cadre of lightly trained knowledge enterers" could use successfully. - building the representations underlying a desk encyclopedia--as a start to representing "the world's most general knowledge, down to ever more detailed levels." - building hooks to other AI programs. 1989: Cyc at the midpoint. Guha, R. V., & Lenat, D.B. (1990). Cyc: A Midterm Report. AI Magazine, 11(3), 33-59. Lenat, D. B., & Guha, R. V. (1990). Building Large Knowledge- Based Systems: Representation and Inference in the Cyc Project. Reading, MA: Addison-Wesley Publishing Company, Inc. Lenat, D. B., & Guha, R. V. (1991). Ideas for applying Cyc (Technical Report No. ACT-CYC-407-91). Microelectronics and Computer Technology Corporation. The AI magazine article and the book present Cyc at midterm-- that is, five years after the beginning of the Cyc project. They describe the representation language they have developed, CycL, including the desiderata are for developing such a language, and why, for example, they no longer use numeric certainty factors. The also describe the top levels of their representations, such as stuff, things and events, time, scripts, causality and intentionality, intelligent beings, uncertainty, beliefs, space. They also (in the book) describe what Cyc's "final exam" will be in "late 1994." The goals are: - "To build an expert system in Cyc and show that it is better than building it from scratch. - "Build two expert systems in Cyc, on related topics, neither project knowing much about the other. Show that they can share each other knowledge productively. - "Communication with Cyc in English, by dint of a natural language understanding program. - "Demonstrate that, based on [the natural language interface], the Cyc project is no longer needed. The massive hand-coding knowledge representation effort can be wound down and eliminated by 1994. - "Demonstrate that Cyc can learn by discovery. - "Have Cyc be the major "consensus reality KB" for the world...we hope that by 1999 no one would even think about having a computer that doesn't have Cyc running on it." Early 1990's: Cyc reviewed and refined. Stefik, M. J., & Smoliar, S. W. (. (1993). The commonsense reviews. Artificial Intelligence, 61(1), 37-40. This journal number contains reviews of Building Large Knowledge-Based Systems by Charles Elkan and Russell Greiner; Drew McDermott; Robert Neches; Douglas Skuce; and John Sowa, totaling over 60 pages. Guha and Lenat reply in a 25-page rebuttal. Guha, R. V. (1990). Micro-theories and contexts in Cyc. Part I: Basic issues (Technical Report No. ACT-CYC-129-90). Microelectronics and Computer Technology Corporation. This paper describes the major architectural difference between Cyc circa 1990 and Cyc circa 1994: adding "contexts" which limit the extent to which knowledge in the KB is believed true, and how to build interfaces among contexts. 1994: Ten years of Cyc. Guha, R. V., & Lenat, D. B. (1994). Enabling agents to work together. Communications of the ACM, 37(7), 127-142. This is a good entry point for reading about Cyc. It describes the state of the art, as well as giving a bit of history. In some ways, this is the final exam report on Cyc, but perhaps a report later in the year is due. It gives a brief historical overview of the Cyc project, and describes its state as of March 1994. In terms of the 1990 final exam goals: - Expert system goals: There was a car selection expert system built around 1990. Guha and Lenat see Cyc as being more likely to succeed in information management programs requiring "person modeling" or "information access," rather than expert systems, - All of the projects listed in this article are prototypes: smart data bases and spreadsheets, that can do consistency checking and inference; image retrieval (e.g., retrieving a image labeled "a girl is lying on the beach" with the query "show me images of people at risk of getting cancer." - Communicating with Cyc in English. This isn't happening. There is a section describing the state of their art, CNL (Cyc natural language system). A quotation: "we expect a sort of crossover to occur in the next two years, by which we mean that most of the knowledge entry will take place by semiautomatic NL understanding, with humans able to take the role of tutors rather than brain surgeons." - Learn by discovery: Guha and Lenat don't discuss this. - Cyc as the major consensus KB: The Cyc project still has seven major sponsors (Apple, Bellcore, DEC, the Department of Defense, Interval, Kodak and Microsoft). It's unclear how any of them are applying Cyc. A large number of AI researchers have been associated with Cyc, but it's unclear that (m)any of them are using Cyc as a common-sense KB. But still Guha and Lenat say: "After almost a decade, the Cyc project is still on target..." with semiautomatic natural language understanding just around the corner, hardware costs plummeting to the point at which Cyc can run on standard low-end platforms (such as Sparc-10s and Mac IIs), and prototype applications available. --- Opinions expressed above are my own, and do not necessarily refect those of the Institute for the Learning Sciences or its partners