******************** 27-May-84 21:30:13-PDT,15114;000000000000 Mail-From: LAWS created at 27-May-84 21:28:54 Date: Sun 27 May 1984 21:22-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 #65 To: AIList@SRI-AI AIList Digest Monday, 28 May 1984 Volume 2 : Issue 65 Today's Topics: AI Tools - KS300 & MicroPROLOG and LISP, Expert Systems - Checking of NMOS Cells, AI Courses - Expert Systems, Cognition - Dreams & ESP, Seminars - Explanation-Based Learning & Analogy in Legal Reasoning & Nonmonotonicity in Information Systems ---------------------------------------------------------------------- Date: 23 May 84 12:42:27-PDT (Wed) From: hplabs!hao!seismo!cmcl2!philabs!linus!vaxine!chb @ Ucb-Vax Subject: KS300 Question Article-I.D.: vaxine.266 Does anybody know who owns the rights to the KS300 expert systems tool? KS300 is an EMYCIN lookalike, and I think it runs under INTERLISP. Any help would be appreciated. ----------------------------------------------------------- "It's not what you look like when you're doin' what you're doin', it's what you're doin' when you're doin' what you look like what you're doin'" ---125th St. Watts Band Charlie Berg ...allegra!vaxine!chb ------------------------------ Date: 25 May 84 12:28:22-PDT (Fri) From: hplabs!hao!seismo!cmcl2!floyd!whuxle!spuxll!abnjh!cbspt002 @ Ucb-Vax Subject: MicroPROLOG and LISP for the Rainbow? Article-I.D.: abnjh.647 Can anybody point me toward microPROLOG and LISPs for the DEC Rainbow 100. Either CP/M86 or MS-DOS 2.0, 256K, floppies. Thanks in advance. M. Kenig ATT-IS, S. Plainfield NJ uucp: ...!abnjh!cbspt002 ------------------------------ Date: 25 May 1984 1438-PDT (Friday) From: cliff%ucbic@Berkeley (Cliff Lob) Subject: request for info This is a request to hear about any work that is going on related to my master's research in expert systems: RULE BASE ERROR CHECKING OF NMOS CELLS The idea is to build an expert system that embodies the knowledge of expert VLSI circuit designers to criticize NMOS circuit design at the cell (<15 transistors) level. It is not to be a simulator, but rather it is to be used by designers to have their cell critiqued by an experienced expert. The program will be used to try to catch the subtle bugs (ie non-logic error, not shown by standard simulation) that occur in the cell design process. I will be writing the code in PSL and a KRL Frame type language. Is there any work of a similar nature going on? Cliff Lob cliff@ucbic.BERKELEY ------------------------------ Date: Fri 25 May 84 13:33:49-MDT From: Robert R. Kessler Subject: re: Expert systems course (Vol 2, #64) I taught a course this spring quarter on "Knowledge Engineering" using the Hayes-Roth text. Since we only had a quarter, I decided to focus on writing expert systems as opposed to developing expert systems tools. We had available Hewlett Packard's Heuristic Programming and Representation Language (HPRL) to use to build some expert systems. A general outline follows: First third: Covered the first 2 to 3 chapters of the text. This gave the students enough exposure to general expert systems concepts. Second third: In depth exposure of HPRL. Studied knowledge representation using their Frame structure and both forward and backward chaining rules. Final third: Discussed the Oak Ridge Natl Lab problem covered in Chapter 10 of the text. We then went through each of the systems described (Chapters 6 and 9) to understand their features and misfeatures. Finally, we contrasted how we would have solved the problem using HPRL. Students had various assignments during the first half of the quarter to learn about frames, and both types of rules. They then (and are right now) working on a final expert system of their own choosing (have varied from a mechanics helper, plant doctor, first aid expert, simulator of the SAIL game, to others). All in all, the text was very good, and is so far the best I've seen. Bob. ------------------------------ Date: Sat, 26 May 84 17:06:57 PDT From: Philip Kahn RE: Subject: cognitive psychology / are dreams written by a committee? FLAME ON Where can you find any evidence that "dreams are programmed, scheduled event-sequences, not mere random association?" I have never found any author that espoused this viewpoint. Per chance, I think that viewpoint imposes far too much conscious behavior onto unconscious phenomena? If they are indeed run by a "committee", what happens during a proxy fight? FLAME OFF ------------------------------ Date: Fri 25 May 84 10:13:51-PDT From: NETSW.MARK@USC-ECLB.ARPA Subject: epiphenomenon conjecture conjecture: 'consciousness', 'essence' etc. are epiphenomena at the level of the 'integrative function' which facilitates the interaction between members of the 'community' of brain-subsystems. Many a-i systems have been developed which model particular putative or likely brain-subsystems, what is the status of efforts allowing the integration of such systems in an attempt to model the consciousness as a 'community of a-i systems' ??? ------------------------------ Date: Fri, 25 May 84 10:09:44 PDT From: Scott Turner Subject: Dreams...Far Out Did the astronauts on the moon suffer any problems with dreams, etc? Without figuring the attentuation, it seems like that might be far enough away to cause problems with reception...since I don't recall any such effects, perhaps we can assume that mankind doesn't have any such carrier wave. Makes a good base for speculative fiction, though. Interstellar travel would have to be done in ships large enough to carry a critical mass of humans. Perhaps insane people are merely unable to pick up the carrier wave, and so on. -- Scott ------------------------------ Date: Sun 27 May 84 11:44:43-PDT From: Joe Karnicky Reply-to: ZZZ.V5@SU-SCORE.ARPA Subject: Re: existence of telepathy I disagree strongly with Ken's assertion that "There seems to be growing evidence that telepathy works, at least for some people some of the time." (May 21 AIlist). It seems to me that the evidence which exists now is the same as has existed for possibly 100,000 years, namely anecdotes and poorly controlled experiments. I recommend reading the book "Science: Good, Bad, and Bogus" by Martin Gardner, or any issue of "The Skeptical Observer". What do you think ? Joe Karnicky ------------------------------ Date: 23 Apr 84 10:51:01 EST From: DSMITH@RUTGERS.ARPA Subject: Seminar - Explanation-Based Learning [This and the following Rutgers seminar notices were delayed because I have not had access to the Rutgers bboard for several weeks. This seems a good time to remind readers that AIList carries such abstracts not to drum up attendance, but to inform those who cannot attend. I have been asked several times for help in contacting speakers, evidence that the seminar notices do prompt professional interchanges. -- KIL] Department of Computer Science COLLOQUIUM SPEAKER: Prof. Gerald DeJong University of Illinois TITLE: EXPLANATION BASED LEARNING Machine Learning is one of the most important current areas of Artificial Intelligence. With the trend away from "weak methods" and toward a more knowledge-intensive approach to intelligence, the lack of knowledge in an Artificial Intelligence system becomes one of the most serious limitations. This talk advances a technique called explanation based learning. It is a method of learning from observations. Basically, it involves endowing a system with sufficient knowledge so that intelligent planning behavior of others can be recognized. Once recognized, these observed plans are generalized as far as possible while preserving the underlying explantion of their success. The approach supports one-trial learning. We are applying the approach to three diverse areas: Natural Language processing, robot task planning, and proof of propositional calculus theorems. The approach holds promise for solving the knowedge collection bottleneck in the construction of Expert Systems. DATE: April 24 TIME: 2:50 pm PLACE: Hill 705 Coffee at 2:30 Department of Computer Science COLLOQUIUM SPEAKER: Rishiyur Nikhil University of Pennsylvania TITLE: FUNCTIONAL PROGRAMMING LANGUAGES AND DATABASES ABSTRACT Databases and Programming Languages have traditionally been "separate" entities, and their interface (via subroutine libraries, preprocessors, etc.) is generally cumbersome and error-prone. We argue that a functional programming language, together with a data model called the "Functional Data Model", can provide an elegant and simple integrated database programming environment. Not only does the Functional Data Model provide a richer model for new database systems, but it is also easy to implement atop existing relational and network databases. A "combinator"-style implementation technique is particularly suited to implementing a functional language in a database environment. Functional database languages also admit a rich type structure, based on that of the programming language ML. While having the advantages of strong static type-checking, and allowing the definition of user-views of the database, it is unobtrusive enough to permit an interactive, incremental, Lisp-like programming style. We shall illustrate these ideas with examples from the language FQL, where they have been prototyped. DATE: Thursday, April 26, 1984 TIME: 2:50 p.m. PLACE: Room 705 - Hill Center Coffee at 2:30 ------------------------------ Date: 3 May 84 16:21:34 EDT From: Michael Sims Subject: Seminar - Analogy in Legal Reasoning [Forwarded from the Rutgers bboard by Laws@SRI-AI.] machine learning brown bag seminar Title: Analogy with Purpose in Legal Reasoning from Precedents Speaker: Smadar Kedar-Cabelli Date: Wednesday, May 9, 1984, 12:00-1:30 Location: Hill Center, Room 423 (note new location) One open problem in current artificial intelligence (AI) models of learning and reasoning by analogy is: which aspects of the analogous situations are relevant to the analogy, and which are irrelevant? It is currently recognized that analogy involves mapping some underlying causal structure between situations [Winston, Gentner, Burstein,Carbonell]. However, most current models of analogy provide the system with exactly the relevant structure, tailor-made to each analogy to be performed. As AI systems become more complex, we will have to provide them with the capability of automatically focusing on the relevant aspects of situations when reasoning analogically. These will have to be sifted from the large amount of information used to represent complex, real-world situations. In order to study these general issues, I am examining a particular case study of learning and reasoning by analogy: legal reasoning from precedents. This is studied within the TAXMAN II project, which is investigating legal reasoning using AI techniques [McCarty, Sridharan, Nagel]. In this talk, I will discuss the problem and a proposed solution. I am examining legal reasoning from precedents within the context of current AI models of analogy. I plan to add a focusing capability. Current work on goal-directed learning [Mitchell, Keller] and explanation-based learning [DeJong] applies here: the explanation of how a the analogous precedent case satisfies the goal of the legal argument helps to automatically focus the reasoning on what is relevant. Intuitively, if your purpose is to argue that a certain stock distribution is taxable by analogy to a precedent case, you will know that aspects of the cases having to do with the change in the economic position of the defendants are relevant for the purpose of this analogy, while aspects of the case such as the size of paper on which the stocks were printed, or the defendants' hair color, are irrelevant for that purpose. This knowledge of purpose, and the ability to use it to focus on relevant features, are missing from most current AI models of analogy. ------------------------------ Date: 15 May 84 11:13:50 EDT From: BORGIDA@RUTGERS.ARPA Subject: Seminar - Nonmonotonicity in Information Systems [Forwarded from the Rutgers bboard by Laws@SRI-AI.] III Seminar by Alex Borgida, Wed. 2:30 pm/Hill 423 The problem of Exceptional Situations in Information Systems -- An overview We begin by illustrating the wide range of exceptional situations which can arise in the context of Information Systems (ISs). Based on this evidence, we argue for 1) a methodology of software design which abstracts exceptional/special cases by considering normal cases first and introducing special cases as annotations in successive phases of refinement, and 2) the need for ACCOMMODATING AT RUN TIME exceptional situations not anticipated during design. We then present some Programming Language features which we believe support the above goals, and hence facilitate the design of more flexible ISs. We conclude by briefly describing two research issues in Artificial Intelligence which arise out of this work: a) the problem of logical reasoning in a knowledge base of formulas where exceptions "contradict" general rules, and b) the issue of suggesting improvements to the design of an IS based on the exceptions to it which have been encountered. ------------------------------ End of AIList Digest ******************** 29-May-84 10:24:26-PDT,16580;000000000000 Mail-From: LAWS created at 29-May-84 10:22:41 Date: Tue 29 May 1984 10:13-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 #66 To: AIList@SRI-AI AIList Digest Tuesday, 29 May 1984 Volume 2 : Issue 66 Today's Topics: AI Courses - Expert Systems, Expert Systems - KS300 Response, Linguistics - Use of "and", Perception - Identification & Misperception, Philosophy - Identity over Time & Essence, Seminar - Using PROLOG to Access Databases ---------------------------------------------------------------------- Date: Tue 29 May 84 08:59:00-CDT From: Charles Petrie Subject: Expert Systems Course Gordon Novak at UT (UTEXAS-20) teaches Expert Systems based on "Building Expert Systems". The class project is building a system with Emycin. For details on the sylabus, please contact Dr. Novak. I took the course and found the "hands-on" experience very helpful as well as Dr. Novak's comments and anedotes about the other system building tools. Charles Petrie ------------------------------ Date: Mon 28 May 84 22:42:41-PDT From: Tom Dietterich Subject: Re: KS300 Inquiry KS300 is a product of Teknowledge, Inc. Palo Alto, CA ------------------------------ Date: 23 May 84 17:31:36-PDT (Wed) From: hplabs!hao!seismo!cmcl2!philabs!sbcs!debray @ Ucb-Vax Subject: Re: Use of "and" Article-I.D.: sbcs.640 > No person would have any trouble at all understanding "people > in Indiana and Ohio", so why should a natural language parser > have trouble with it??? The problem is that the English word "and" is used in many different ways, e.g.: 1) "The people in Indiana and Ohio" -- refers to the union of the set of people in Indiana, and the set of people in Ohio. Could conceivably be rewritten as "the people in Indiana and the people in Ohio". The arguments to "and" can be reordered, i.e. it refers to the same set as "the people in Ohio and Indiana". 2) "The house on 55th Street and 7th Avenue" -- refers to the *intersection* of the set of houses on 55th street and the set of houses on 7th Avenue (hopefully, a singleton set!). NOT the same as "the house on 55th street and the house on 7th Avenue". The arguments to "and" *CAN* be reordered, however, i.e. one could as well say, "the house on 7th Ave. and 55th Street". 3) "You can log on to the computer and post an article to the net" -- refers to a temporal order of events: login, THEN post to the net. Again, not the same as "you can log on to the computer and you can post an article to the net". Unlike (2) above, the meaning changes if the arguments to "and" are reordered. 4) "John aced Physics and Math" -- refers to logical conjunction. Differs from (2) in that it can also be rewritten as "John aced Physics and John aced Math". &c. People know how to parse these different uses of "and" correctly due to a wealth of semantic knowledge. For example, knowledge about computers (that articles cannot be posted to the net without logging onto a computer) enables us to determine that the "and" in (3) above refers to a temporal ordering of events. Without such semantic information, your English parser'll probably get into trouble. Saumya Debray, SUNY at Stony Brook uucp: {cbosgd, decvax, ihnp4, mcvax, cmcl2}!philabs \ {amd70, akgua, decwrl, utzoo}!allegra > !sbcs!debray {teklabs, hp-pcd, metheus}!ogcvax / CSNet: debray@suny-sbcs@CSNet-Relay ------------------------------ Date: Fri 25 May 84 12:10:32-CDT From: Charles Petrie Subject: Object identification The AI approach certainly does not seem to be hopeless. As someone else mentioned, the boat and ax problems are philosophical ones. They fall a bit out of our normal (non-philisophical) area of object recognition: these are recognition problems for ordinary people. The point we should get from them is that there may not be an objective single algorithm that completely matches our intuition about pattern recognition in all cases. In fact, these problems may show such to be impossible since there is no intuitive consensus in these cases. The AI approach aspires to something more humble - finding techniques that work on particular objects enough of the time so as to be useful. Representing objects as feature, or attribute, sets does not seem hopeless just because object's features change over time. Presumably, we can get a program to handle that problem the same way that people do. We seem to conclude that an object is the same if it has not changed too much in some sense. Given that the values of the attributes of an object change, we recognize it as the same object if, since the last observation, either the values have not changed very much, or most values have not changed, or if certain high priority values haven't changed, or some combination of the first three. To some extent, object recognition is subjective in that it depends on the changes since the last observation. When we come home after 20 years, we are likely to remark that the town is completely different. But what makes it the same town so that we can talk about its differences, are certain high importance attributes that have not changed, such as its location and the major street layout. If we can discover sufficient heuristics of how to handle this kind of change, then we succeed. Since people already do it, even if it involves additional large amounts of contextual information, feature recognition is obviously possible. Charles Petrie ------------------------------ Date: 23 May 84 11:18:54-PDT (Wed) From: ihnp4!ihuxr!lew @ Ucb-Vax Subject: Re: misperception Article-I.D.: ihuxr.1096 Alan Wexelblat gave the following example of misperception: ------------------- A more "severe" case of misperception is the following. Suppose that, while touring through the grounds of a Hollywood movie studio, I approach what, at first, I take to be a tree. As I come near to it, I suddenly rea