The low cost of microprocessors today, and the future trend in both cost and performence, makes large microprocessor-networks very interesting. A net of a thousend processors or more can be built using present techniques. However, there are several problems in utilizing such a horee of processors in a rcesonebly efficient way. There also seem to be restrictions to the kinds of opplications that can be mapped on to such a system. One control mechanism that seems to be very useful, at least for some Artificial Intelligence type problems, is the CONTRACT NET, ISmith77]. In this paper we will first look at some of the desirable characteristics of a lerge multi-microprocessor net. Than we will describe several different organizations together with their advantages and problems. We are also discussing broadcasting in lattices using circuit switched minimum spanning trees.

A new predicate calculus deduction system based on production rules is proposed. The system combines several developments in Artificial Intelligence and Automatic Theorem Proving research including the use of domain-specific inference rules and separate mechanisms for forward and backward reasoning. It has a clean separation between the data base, the production rules, and the control system. Goals and subgoals are maintained in an AND/OR tree to represent assertions. The production rules modify these structures until they "connect" in a fashion that proves the goal theorem. Unlike some previous systems that used production rules, ours is not limited to rules In Horn Clause form. Unlike previous PLANNER-like systems, ours can handle the full range of predicate calculus expressions including those with quantified variables, disjunctions and negations.

Interpretation expected by the referring The technical issues of acquiring this knowledge, physician.

Each computing facility, which we will call a node, includes a process which is in charge of the management of some section of the data base. Each node is connected to the other nodes to allow the sharing of data. We will assume that there is neither shared memory nor global data in the system. Except for this restriction, our model will not specify the interconnection architecture of the network. Au important objective of our studies is in fact the development of measures of computational activities which will lead to evaluation criteria for architectural alternatives [10].

C. Richard Johnson, Jr. provided very helpful comments on adaptive control systems. We received many valuable suggestions from members of the Heuristic Programming Project at Stanford. 2 Supported by the Research and Development Branch of the Department of National Defence of Canada.

Computer facilities were provided by t"e SUMEX-AIM facility at Stanford University under National Institutes of health grant RR-00785. The author is supported by the Research and Development Branch of the Department of National Defence of Canada.

Both tasks are concerned with the interpretation of large quantities of digitized signal data. The task of SU/X is to understand "continuous signal that is, signals which persist over time. The task of SU/P is to interpret protein x-ray crystallographic data. Some features of the design are: (1) incremental interpretation of data employing many different pattern-invoked sources of knowledge, (2) production rule representation of knowledge, includiR high level strategy knowledge, (3) "opportunistic" hypothesis formation using Dr,

It is generally accepted that problem solving systems require a wealth of domain specific knowledge for effective performance in complex domains. This report takes the view that all domain specific knowledge should be expressed in a knowledge base. With this in mind, the ideas and techniques from problem solving and knowledge base research are reviewed and outstanding problems are identified. Finally, a task domain is characterized in terms of objects, actions, and control/strategy knowledge and suggestions are made for creating a uniform knowledge base management system to be used for knowledge acquisition, DD "" 1473

ABSTRACT A computer program that models an expert in a given domain is more likek to be accepted by experts in that domain, and by non-experts seeking its gavice. An explanation capability not only adds to the system's credibility, but also enables the non-expert user to learn from it. Furthermore, clear explanations allow an expert to check the system's "reasoning", possibly discovering the need for refinements and additions to the svstem's knowledge base. In a developing system, an explanation capability can be used as a debugging aid to verify that additions to the system are working as'hey should. The explanation facility in MYCIN is discussed as an illustration of how the various problems might be approached.

Methods of Approach At the outset it might be well to distinguish sharply between two general approaches to the problem of machine learning.