Artificial intelligence (AI) is an extremely large and complex field technically, while at the same time it captures our imagination and prompts us to explore major philosophical and ethical questions concerning humanity and human intelligence. Teaching a course that does justice to all these aspects of the field is a big challenge. However, due to the increase in computational capability with a commensurate decrease in cost, a wealth of products and materials are available that can be used to provide students with rich, meaningful, and memorable experiences within the context of a primarily technical course in AI. Toys, articles, and movies can all be used to foster student exploration of key questions in the technical, philosophical, and ethical issues of AI.

The use of multi-agent architectures in knowledge-based systems is increasing with the recognition that complex, difficult tasks need to be decomposed into smaller sub-tasks and that each sub-task should be solved with the most appropriate reasoning technique. Complex tasks require complex architectures. This research took a preliminary look at planning and scheduling in a medical domain. The medical test planning system (MTPS) is a multi-agent software system for planning and scheduling medical tests. East of the main agents is responsible for solving a sub-task of the larger problem using an appropriate reasoning technique. While some of the issues addressed in the system stemmed from the domain of medical testing, the approach to planning and scheduling operations within a domain is general.>

Current expert system technology tends to rely on the use of shallow empirically based experiential knowledge. With only this type of knowledge available, expert systems have been capable of reaching a high level of agreement with human experts in a limited area of expertise. However, due to the nature of their knowledge, such systems fall short of human expertise in many ways. The human diagnostic process is examined as it relates to the malfunction of mechanical and electrical devices. An expert system design is presented, called the integrated diagnostic model (IDM), that attempts to address some of the issues involved in bridging the gap between human and computer expertise. The IDM contains two different types of knowledge, one based on experience and one based on how the device to be diagnosed functions. These two types of knowledge are used together during a diagnostic session to determine what is wrong with the device. To demonstrate how the IDM works, an interaction with a prototype system that was built using the IDM is described; then research on extensions to the IDM is discussed.

Existing expert systems have a high percentage agreement with experts in a particular field in many situations. However, in many ways their overall behavior is not like that of a human expert. These areas include the inability to give flexible, functional explanations of their reasoning processes, and the failure to degrade gracefully when dealing with problems at the periphery of their knowledge. These two important shortcomings can be improved when the right knowledge is available to the system. This paper presents an expert system design, called the integrated diagnostic model (IDM), that integrates two sources of knowledge, a shallow, reasoning-oriented, experiential knowledge base and a deep, functionally oriented, physical knowledge base. To demonstrate the IDM's usefulness in the problem area of diagnosis and repair, an implementation in the mechanical domain is described.

A natural language processor is described for control of a machine in task-oriented situations. Particular emphasis is given to issues related to flow-of-control statements in dialogue. These include branching constructs, as in `if row 1 contains a positive entry, then . . .' and looping constructs, as in `repeat for all other rows'. Special problems are discussed concerning the processing of deeply nested control structures, pronoun resolution, and the handling of conjunctions. An experiment is described in which the robustness of the conditional feature was tested with a group of computer naive subjects. It was found that subjects could discover and use the feature effectively in solving problems even though the fact of its existence was systematically withheld during the training session.