16.871 SPRING 2006 READING LIST: Installment #1Lecture 1: Course Organization; Spirit of Undertaking1. Feigenbaum, E.A., Knowledge Engineering. The Applied Side of Artificial Intelligence, An-nals New York Academy of Sciences, 91-107.A call to arms that lays out the spirit of knowledge based systems, some of its promise, andgoals.2. Newell, A., The Knowledge Level, Artificial Intelligence Magazine, Summer 1981, pp. 1-20+.A somewhat theoretical statement whose intention is to demonstrate that it makes sense to talkabout the performance of a program a t the “knowledge level,” i.e., in terms of what it knows,rather than in terms of what computations it carries out. This becomes particularly clear whenNewell explains that we can talk about what an agent knows (i.e., its knowledge level description)without any idea about how that knowledge is represented inside the agent (i.e., its symbol-leveldescription).This is a useful point to keep in mind as we go through the course: we want you to understandboth what knowledge is needed and how to represent it, but we also want you to be able todistinguish carefully and know which one mat ters when.3. Topolski A S, R eece D K, Packaging Advisortm: An expert system for rigid plastic foodpackage design, in Proc Innovative Applications of AI, 1989, Schoor and Rappaport (eds.),pp. 348–357.This is one small, well-described application, to give you a feeling for what expert systems canbe, and why they can matter. Note in particular what the company got from building the system;pay attention t o the wide variety of benefits that resulted.Lecture 2: Tell it what to know; Search1. Korf, R. E., Search: A survey of recent results. Exploring Artificial Intelligence, Shrobe, H.E. (ed), Morgan Kaufmann, 1988, pp. 197-237, and ACM Computing Surveys, vol. 27, No. 3,Sept 1995, pp. 337-339 [an update to the previous paper].If you are somewhat lacking in AI background, you would d o very well to read these two papers.We will often rely on concepts of search in the rest of t h e course, so you ought to know thebasics, at least. You should read these closely enough to have a feel for wha t’s hard and what’snot. Instant recollection of part icular techniques is not the goal.2. Rowley, S., Shrobe, H., Cassels, R. and Hamscher, W., Uniform Access to hetrogenous knowl-edge structures, or Why Joshing is better than Conniving or Planning. In AAAI-87, pages48-52.This is about a language which tries to support hybrid reasoning, i .e., an approach in which youcan mix together many components supporting different paradigms. We use it as one exampleof what it means for a language to provide direct support f or a representation and/or reasoningmetho d.Lecture 3: Origins of KBS: macsyma and dendral1. Slagel J R, A heuristic program that solves symbolic integration problems in freshman cal-culus, in Computers and Thought, pp. 191–203.Pay particular attention to the table on page 201. What does it tell you about the effect of theheuristics on solving the problem? In p articular, how us eful were they?6.871, Spring 2006 22. Moses J, PhD thesis, Chapter 2: How SIN differs from SAINT.In this very short piece of text from his PhD thesis Prof. Joel Moses (former MIT Provost) explainshow his approach is substantially different in method and mindset from SAINT.3. Feigenbaum, E.A., Buchanan, B.G., Lederberg, J., On Generality and Problem Solving: ACase Study Using the DENDRAL Program. in Meltzer and Michie (eds) Machine Intelligence6, pp. 165-189.This article is in part about the balance of power in a program using generate and test. Considerwhat makes t h e generator gets smarter and what consequences this has for the tester.Lecture 4: Application Analysis Case Study: Case Introduction1. Sviokla J, Smartwave: The Wave-Soldering Expert System, Harvard Business School Cases#0–187–062 and #0–187–063This lecture will set the background for your analysis of the wave soldering system as a knowledge-based system. Use the notes from lecture as a guide to the kinds of things to look f or in theproblem description.This is a real problem, h ence there is a lot of inf ormation in these descriptions; some of it isirrelevant to your needs; some of what you need may be miss ing. Your job is to find what youneed, to ask about what you need but don’t have, and to do the best analysis you can.You should concentrate on the first case, Smartwave (A), t rying to determine to what extent itmeets the criteria outlined in the lecture. Then look bri efly at Smartwave (B), to see what sortsof problems they faced in the next step of development.Lecture 5: Application Analysis Case Study: Class Discussion1. Written Assignment: Prepare a brief written report on the case analysis to use in the classdiscussion. Your write-up should concentrate on Smartwave A and explain what aspects ofthe task made it a good candidate and what aspects made it a poor candidate.Your write-up should also provide a list and description of the problems Smartwave encoun-tered at the next stage (case B). What problems did they face and why?6.871, Spring 2006 3Lecture 6: Rule Based Systems1. Davis, R., Pro duction Rules as a representation for a kn owledge-based consultation Program,Artificial Intelligence Journal, 8:15-45, 1977.What are the basic strengths and weaknesses of rule-based systems, as reported here?2. Duda, R., Gaschn ig, J., and Hart, P., Model design in the prospector consultant systemfor mineral exploration, Expert Systems in the Microelectronic Age Michie (ed.), EdinburghUniversity Press, 1979, pp. 154–167.Read this to see another way to implement a rule-based system, wi th a different approach tohandling uncertainty.3. Campbell A. N., Hollister, V. F., Duda, R. O., and Hart, P. E., Recognition of a hiddenmineral deposit by an artificial Intelligence program, Science, 217:927-929, Sept. 3, 1982.This short piece is interesting documentation of the significant real-world impact knowledge basedsystems can have.Lecture 7: Semantic Nets1. Quillian, M. Ross, Semantic Memory. In M. Minsky (ed.) Semantic Information Processing,The MIT Press, 1968, pp. 227–270.This is the grand-daddy of semantic nets. Al most all the detail s are wrong if not outright silly.Don’t worry about trying to commit this stuff to memory. Instead concentrate on spirit of thi sidea and the insights that lead in this di rection.2. MacGregor R, The evolving tech nology of classification-based knowledge representation sys-tems, in