DISCTPLE-1: INTERACTIVE APPRENTICE SYSTEM IN WEAK THEORY FIELDS YVES KODRATOFF LRI, Inference et Apprentissage, Bit. 490, U.A. 410 du C.N.R.S. & University Paris-Sud, F - 91405 Orsay OHEOROHE TCCUQ Research Institute for Computers and Informatics, 71316, Bd. Miciurin 8-10, Sector 1, Bucharest, Romania ABSTRACT rj A WEAK THEORY DOMAIN The paper presents an interactive approach to learning apprentice sys- tems for weak theory domains. The approach consists of a combination of teaming by analogy and learning by generalizing instances. One main point of this approach is that it uses the explanations drawn from an example, both to reduce the version space of me rules to be learned, and to generate new examples, analogous to the given one. Another im- portant point is that it demonstrates not only that over-generalization is harmless but also useful and necessary, when interacting with a user. It allows to use the theory of the domain, though incomplete as it is, in order to extract the missing knowledge by asking "clever" questions to its user. This paper presents a first prototypical version of DISCIPLE and its use to the design of technologies for the manufacturing of loudspeakers. I INTRODUCTION If Expert Systems have proven useful in many domains, their appli- cations are limited by their inability to acquire and to update their knowledge. This problem is largely recognized as the knowledge ac- quisition bottleneck of Expert Systems (Feigenbaum 1977), (Mitchell & Al. 1985), (Kodratoff 1986), etc... Recent Machine Learning achieve- ments ((Mitchell, Carbonell & Michalski 1985) offer new solutions to the knowledge acquisition problem and open a new area in the evolution of Expert Systems, that is, Expert Systems able of automatic knowledge acquisition and learning, such as Learning Apprentice Systems (LAS). A LAS is an interactive knowledge-based consultant that directly assimi- lates new knowledge by observing, analyzing and questioning about the problem solving steps contributed by their users through their normal use of the system. The user gives to the system a problem to solve and the expert sub-system starts solving this problem by showing the user all the problem solving steps. The user may agree or reject them. Therefore, in its Expert System mode, a LAS may encounter two situations. Either the current problem-solving step (which we shall further call partial solution) is accepted by the user. Then, the current state of the knowledge base is judged as satisfactory, and no learning will take place. Or it is unable to propose any partial solution (or the solution it pro- poses is rejected by the user). Then, the user is compelled to give his own solution. Once this solution is given, a learning process will take place. The LAS will try to learn a general rule so that, when faced with problems similar to the current one (which it has been unable to solve), it will become able to propose a solution simitar to the solution given by the user to the current problem. We are developing a LAS, called DISCIPLE, spe- cialized for weak theory domains. In this paper we describe the learn- ing mechanisms of DISCIPLE. To this purpose we use examples from Technology Design. The next section is a brief description of this domain. The following sections present the learning problem and the learning method of DISCIPLE We have chosen, as a first domain to test our approach to interactive LASs, the domain of designing technologies for the manufacturing of loudspeakers. Before presenting in more details this domain we stress two of its important features. Firstly, the domain is usually too complex for an autonomous system. Secondly, small improvements in technology have important outcomes since a technology is usually used for a large number of products. Therefore the best solution is searched. A conse- quence of these features is that such a domain is most appropriately han- dled with an interactive system as the expert (consultant) sub-system of a LAS, where the user and the system cooperate in finding the best solu- tion to the current problem. Technology Design might well be viewed as successive decompositions of complex operations into simpler ones, and successive specializations of these simpler operations by choosing tools, materials or verifiers, which are in turn successively specialized. To design a technology, DISCIPLE needs some knowledge about the components of the loudspeakers, about the technological solutions for the manufacturing of loudspeakers, about the tools and the materials one can use to manufac- ture loudspeakers. All this knowledge constitutes the domain theory. This domain theory is inherently incomplete since we can not suppose that DISCIPLE knows all the objects of the domain, all the properties of a given object, all the actions that can be performed for manufacturing loudspeakers, all the properties of the known actions (preconditions, effects), all the ways of decomposing or specializing a given action, etc... III THE LEARNING PROBLEM In the domain we have chosen, DISCIPLE acts as an aid to a tech- nology designer. The problem to be solved is that of planning the manufacturing of a certain loudspeaker. The solution to this problem is a plan of actions for manufacturing the loudspeaker. The problem-solving paradigm is problem-reduction. That is, DISCIPLE will successively decompose an action into simpler actions or specialize an action to a better defined one. In this way, DISCIPLE will build a problem-solving tree. This process continues untill the leaves of this tree are elementary actions. They represent the solution to the original problem (the top of the problem-solving tree). Let us suppose that, during planning the manufacturing of a loudspeak er, DISCIPLE encounters the following problem ATTACH sectors ON chassis-membrane-assembly for which it is unable to propose a satisfactory solution. Let us further suppose that the user indicated the following solution to DISCIPLE: APPLY mowicoll ON sectors, PRESS sectors ON chassis-membrane-assembly Note that APPLY and PRESS may be actions previously unkown to the system and, in such a case, it knows nothing about them except that the are means of ATTACHing. Now DISCIPLE knows a solution of the current problem Kodratoff and Tecuci 271