Millions of base pairs worth of DNA sequences are now known and must be analyzed: hundreds of restriction enzymes and cloning vectors must be kept track of to carry out work in genetic engineering. Computational systems have become a necessary tool to acquire, retrieve, manipulate, and analyze this information. A myth of the early days of computing was that computers were excellent number manipulators, but poor at working with other forms of information.

Current subprojects include applications of the program Gerring.

Although computer-based instruction has become widely available as a learning aid in medical education, few physicians interact with educational programs after they have left medical school. Some notable exceptions occur when specially prepared computer programs are made available by vendors or program committees at annual clinical meetings. Yet this kind of learning tool is seldom used by practicing physicians at other times during the year. In this paper, I would like to consider ways in which computer-based education might be more effectively integrated into the clinical activities of the practicing physician, and to outline some of the technological and psychological barriers to their successful implementation.

J. J. Finger and Michael R. Genesereth Computer Science Department Stanford University, Stanford, California 94305 1. Introduction Automatic theorem proving methods, such as resolution or backwards-chaining, arc useful for verifying the correctness of a pre-cxisting design, but how might we use theorem proving methods to both generate and verify a design?

The process of circuit design is complex largely because the required knowledge takes many forms. We present a framework that contains such design descriptions as components, plans, goals, and tradeoffs. The design process is represented by tasks, which synthesize and revise descriptions, and principles that should be upheld by descriptions. Con, -)1 of the circuit design process involves sequencing the creation and execution of tasks and the.naimainence of principles. Control is knowledge-intensive: different design processes are represented by such control descriptions as strategies and metagoals. We provide examples of design tasks, principles, and control descriptions. Finally, we describe a computer program based on this framework.

Reprinted, with permission, from IEEE Acoustic, Speech and Signal Processing, Spring, 1984. ABSTRACT In the past fifteen years, artificial intelligence scientists have built several signal interpretation, or understanding, programs. These programs have combined "low" level signal processing algorithms with knowledge representation and reasoning techniques used in knowledge-based. HASP/SIAP is one such program that tries to interpret the meaning of passively collected sonar data. In this paper we explore some of the Al techniques that contribute in the "understanding" process. We also describe the organization of HASP/SIAP system as an example of a programming framework that show promise for applications in a class of similar problems.1 Using data from concealed hydrophone arrays, it must detect, localize, and ascertain the type of each ocean vessel within range. Tne presence and movements of submarines are of most interest, but there are strategic and tactical motives for monitoring all vessel types.

MARS is an experimental system that provides a framework for implementing hierarchical discrete event driven simulators. The program is independent of any domain.

The duraticn of the neuro'ogical signs is 7.2 hours.

Ihe control problem--which of its potential actions should an Al system perform at each point in the problem -solving process?-- is fundamental to all cognitive processes. To solve the control problem intelligently, Al systems should achieve (at least) the seven behavioral goals set forth in this paper. The paper proposes a blackboard model of control and shows how it achieves the goals. The pdper contrasts the model with three alternative control models and shows how it continues an evolutionary progression of control architectures.

To perform these diagnoses, DART must frequently determine how the hardware's primary inputs can be manipulated to produce desired test conditions at internal nodes. Especially when the system's behavior is time-dependent, this reasoning must be carefully controlled, or a combinatorial explosion may result. This paper contrasts two techniques for representing time-dependent digital system behavior and controlling reasoning to achieve desired hardware states. 2