この座談会は,昨年9月に北海道大学で開かれた文部省の知的処理機構の研究会に参加した方々にお集まりいただいて,研究会後に行われたものである.ここに掲載したものは人間の知能と機械の知能について,2時間にわたって自由に話していたたいたものを約半分に縮めて,編集したものである.

It is very important for many rule-based systems to specify certainties of rules and to compute certainties of inferred conclusions. One of the promising logical frameworks for it is logic programs with uncertainties (LPU). This paper proposes a best first search (BFS) algorithm for LPU. It allows us to make useful interpreter for LPU that can give, to a given query, the current best solution in the remaining solutions (or tell us the failure of finding it) each time we ask the interpreter for another solution. Important is that each solution is found with the minimal computation required at each time. Using BFS algorithm we implemented an extended Prolog which has the BFS interpreter avoiding the infinite computations and dealing with the negation as failure similar to standard Prolog.

A story understanding system using belief revision mechanism is described. The system uses TMS, written by Doyle <(1)>. The system maintains multiple belief spaces for each person in a story, along with one real space which stands for "absolute truth" about the story. Especially, a new method to deal with contradiction between each person's belief spaces is proposed. In addition, a script can handle multiple paths and branches at any point, and such branches can cause revision of multiple states wholly over the script and belief spaces. To achieve this, when one knowledge requested invocation of some relative knowledge accompanying it (superordinate knowledge, etc.), a daemon process may be invoked to construct dependency links between related events

This paper presents a learning method for automating knowledge acquisition in an arithmetic problem solver. Our arithmetic problem solver with learning mechanism called LAPS can solve many different arithmetic problems given in natural language without reprogramming. LAPS consists of a natural language processor, equation extractor, equation solver, rule generator, rule modifier and knowledgebase. In such a problem solver, a huge amount of knowledge is required such as, common knowledge to understand given problems and domain specific knowledge to extract equations. When a given problem cannot be solved because of a shortage of such knowledge, LAPS can acquire the knowledge through interaction with an outside teacher. Knowledge obtained from a teacher is represented in the form of rules. Such obtained knowledge is usually rather specific and specific knowledge should be generalized and refined in order to use it for solving problems. Since a problem solver generally does not have enough knowledge to explain a given problem, explanation based generalization is not applicable. On the other hand, the similarity based learning requires a negative instance or some other constraint in order to avoid overgeneralization. In our problem solver, generalization is made by using the constraint that the given problem should be solved. Our learning method is based on empirical similarity but can avoid overgeneralization without giving negative instances. That is, the validity of the generalized rules are confirmed by solving problems, although the similarity-based learning cannot assure it. Furthermore, in this generalization process, LAPS can improve its performance at problem solving by synthesizing the applied rules.

Medical knowledge-base systems are expected to become useful tools in practical clinical environments. To attain this goal, the system should integrate medical record management utility. POS is one of most representative scheme for medical record management, and it is quite easily adaptable to object-oriented or frame system. Integrating knowledge information processing utility and POS on a frame system, really practical medical knowledge-base systems can be implemented.

There are two important parts in the study of the artifitial intelligence. One is a knowledge representation, and the other is a program transformation. Till now, they were studied independently. But we thought if they amalgamated, they might obtain some new results. So we applied the program transformation to the multiple world mechanism which is used for the knowledge representation. In this paper, we introduce an example of the transformation on the multiple world mechanism and examine its validity. As the result, the method described here may be a hopeful approach for the AI programming.