Department of Computer Science Carnegae-Mellon Unaverszty P&burg, PA 15213 Editors' Note: Many expert systems require some means of handling heuristic rules whose conclusions are less than certain Baysian techniques and other numerical scoring methods have been developed to combine and propagate certainty measures as the expert system draws inferences in solving different problems. Doyle's paper argues that it is difficult for a human expert to produce reliable probabilities or numerical scoring factors for an inference rule, and that a radically different approach to the problem should be considered He essentially suggests that the expert be encouraged to think in terms of specific instances which would conflict with the general rule and to encode this knowledge explicitly. Methodologically this seems to be very appealing, and helps to make both explicit and rigorous some of the techniques currently used by knowledge engineers whm they encode and refine the expert's knowledge We would welcome comments and criticisms of this approach from those steeped in the practical issues of constructing large rule-based expert systems. Probabilistic rules and their variants have recently supported several successful applications of expert systems, in spite of the difficulty of committing informants to particular conditional probabilities or "certainty factors," and in spite of the experimentally observed insensitivity of system performance to perturbations of the chosen values Here we survey recent developments concerning reasoned assumptions which offer hope for avoiding the practical elusiveness of probabilistic rules while retaining theoretical power, for basing systems on the information unhesitatingly gained from expert informants, and reconstructing the entailed degrees of belief later @

In fact, there are enough opinions for four men. That is, the views advanced are contradictory. I agree with one of the A fifth answer is also advanced, but is immediately withdrawn. Roger Schanks, and disagree with the other three. Schank hoped that his article would start a debate on As & hank points out, this is unsatisfactory because it leads the issues he raised.

One of the major problems in developing so-called intelligent computer-aided design (CAD) systems (ten Hagen and Tomiyama 1987) is the representation of design knowledge, which is a two-part process: the representation of design objects and the representation of design processes. We believe that intelligent CAD systems will be fully realized only when these two types of representation are integrated. Progress has been made in the representation of design objects, as can be seen, for example, in geometric modeling; however, almost no significant results have been seen in the representation of design processes, which implies that we need a design theory to formalize them. According to Finger and Dixon (1989), design process models can be categorized into a descriptive model that explains how design is done, a cognitive model that explains the designer's behavior, a prescriptive model that shows how design must be done, and a computable model that expresses a method by which a computer can accomplish a task. A design theory for intelligent CAD is not useful when it is merely descriptive or cognitive; it must also be computable.

Studies in design methodology provide various structured approaches to the design process. Many books provide definitions and elaborations of the design process: In the structural engineering field, such books include Holgate (1986) and Lin and Stotesbury (1981). More generally, various design methods and techniques are described in Alexander (1964) and Jones (1970). These design methods share the characteristic of prescribing a general set of tasks to be performed by the designer. One problem with design methodologies is that such approaches prescribe what a designer should do but not how.

I should like to lodge a complaint about your editorial standards in the article "An Assessment of Tools for Building Large KB Systems," by William Mettrey, in the winter 1987 [volume 9 number As a primary architect of CRL-Ops and a former KnowledgeCraft class instructor, I had to deal with the general public's misconceptions about forward versus backward chaining systems. Mr. Mettrey's article, in my opinion, is the type which generates the confusion that forward chaining rule systems cannot "backwards chain." This nonsensical view was held by the vast majority of our customers in the KC class. The section on Rule-Based inference implies that backward chaining is done only by Prolog in KC with its statement "by contrast, Knowledge-Craft implements backward chaining by supporting a version of Prolog." Any forward chaining rules system can efficiently implement constrained backward chaining by simply using a goal structure to search for the required knowledge.

I report on my experience over the past few years in introducing automated, model-based diagnostic technologies into industrial settings. In particular, I discuss the competition that this technology has been receiving from handcrafted, rule-based diagnostic systems that has set some high standards that must be met by model-based systems before they can be viewed as viable alternatives. The battle between model-based and rulebased approaches to diagnosis has been over in the academic literature for many years, but the situation is different in industry where rule-based systems are dominant and appear to be attractive given the considerations of efficiency, embeddability, and cost effectiveness. Traditionally, industrial diagnostic systems have been handcrafted to reflect the knowledge of a domain expert. They take the form of if-then rules that associate certain forms of abnormal system behavior with faults that could have caused this behavior.

In 1980, Digital Equipment Corporation began to use a rule-based system called Rl by some and XCON by others to configure VAX-11 computer systems In the intervening years, Rl's knowledge has increased substantially and its usefulness to Digital continues to grow. This article describes what is involved in extending Rl's knowledge base and evaluates Rl's performance during the four year period. "Rl: the formative years" described how a A large number of people have played critical roles in Rl's development. Among those who deserve special mention are John Barnwell, Dick Caruso, Ken Gilbert, Keith Jensen, Allan Kent, Dave Kiernan, Arnold K&t, Dennis O'Connor, and Ed Orciuch. We want to thank Allen Newell, Dennis O'Connor, and Ed Orciuch for their helpful comments on earlier drafts of this article Briefly, given a customer's purchase order, Rl determines what, if any, substitutions and additions have to be made to the order to make it consistent, complete, and produce a number of diagrams showing the spatial and logical relationships among the 50 to 150 components that typically constitute a system.

In this editorial we introduce the articles published in this special AI Magazine issue on innovative applications of artificial intelligence. Discussed are a pick-pack-and-ship warehouse-management system, a neural network in the fishing industry, the use of AI to help mobile phone users, building business rules in the mortgage lending business, automating the processing of immigration forms, and the use of the semantic web to provide access to observational datasets. Pop culture often shows AI in clearly visible forms like a futuristic robot attempting to take over the earth. However, current applications of AI are often more difficult to detect because they are imbedded in existing products, processes, and services. We interact with and benefit from these AI applications in ways that most people never even notice.

Interactive intelligent systems often suffer from a basic conflict between their computationally intensive nature and the need for responsiveness to a user This paper introduces the Interviewer/Reasoner model, which helps to reduce this conflict This model partitions an intelligent system into two asynchronous components The Interviewer's primary function is to gather data while providing an acceptable response time to the user The Reasoner does most of the symbolic computation for the system This paper describes the implementation of the model in both timesharing and personal workstat,ion environments, and uses the ONCOCIN system as an example The work described in t,his paper was carried out at Stanford University and was partly supported by the National Library of Medicine under program project grant LM-00395. The original idea for splitting the tasks of information gathering from reasoning in order to improve system response time was suggested by Ted Shortliffe and Chuck Clanton for the ONCOCIN project Thanks are due to Eric Schoen and Bill van Melle for help with the implementation, to Mark Stefik and Harold Brown for help in writing this paper, and to the rest of the ONCOCIN project members, including Carli Scott, Miriam Bischoff, Charlotte Jacobs, and Craig Tovey. An acceptable response time is needed both during system testing and to help insure end-user acceptability. During the normal course of development of an AI system there is substantial t,esting on real problems under the guidance of human experts whose time is usually valuable. Moreover, many end users (e.g., physicians) will simply refuse to use a system if they have to wait for a response.

J. Sainsbury has extensive assets, with subsidiaries such as Shaws in the United States and the Savacentre and Homebase chains in the United Kingdom. Given J. Sainsbury's position in the retail market, the efficient and effective running of the supply chain for J. Sainsbury is critical to the mission of the organization. The J. Sainsbury logistics purpose statement is to manage the flow of goods from supplier to shelf, ensuring that the customer has the right product in the right place at the right time. To these ends, J. Sainsbury's Logistics Group is committed to being world class. The group's direction principle is to be seen as the world's best logistics team.