Researchers in the development of medical expert systems have Increasingly recognized the Importance of explanation capabilities in encouraging the acceptance of their programs. One survey of potential users of medical advice systems has suggested that explanation may be the single most important capability of an acceptable clinical decision tool (16). Good explanations serve four functions in a consultation system: 111 they provide a method for examining the program's reasoning if errors arise when the system is being built; 121 they assure users that the reasoning is logical, thereby increasing user acceptance of the system; 131 they may persuade users that unexpected advice is appropriate; and 141 they can educate users in areas where their knowledge may be weak.

The ONCOCIN Project is supported by research grants from the National Library of Medicine, the Division of Research Resources of the NIH, the Office of Naval Re"arch, and the Henry J. Kaiser Family Foundation.

Reprinted by permission of the author. Published in the Proceedings of a Symposium on Computers in Medicine, Annual Meeting, California Medical Association, Anaheim, CA., February 1984. Alt;iough computing technology is playing an increasingly important role in medicine, systems designed to advise physicians on diagnosis or therapy selection have remained largely experimental to date. Despite diverse research efforts, and a literature on computer-aided diagnosis that has numbered over 1500 references in the last 20 years, clinical consultation programs have failed to achieve wide acceptance. The reasons for attempting to develop such systems are self-evident.

"Coming to Terms With the Computer" by Edward H. Shortliffe reprinted with permission from The Machine at the Bedside, Eds. You are asked to assist a major teaching hospital in the assessment of a large computer system that was installed 3 months ago to help with doctors' orders, laboratory test reporting, nursing schedules, and bed control. Because of mixed reviews of the system's effectiveness, the hospital has decided to bring in outside experts to assess the computer's strengths and weaknesses. The computer system was installed by a vendor of large-scale hospital information systems (HIS). The company had developed the programs over several years, but this is its first major commercial installation.

Symbolic inference since the time of Aristotle has involved the combination of symbolic expressions.

Multiple processors can be used to achieve a speedup of a backward-chaining deduction by distributing or-parallel deductions. However, the actual speedup obtained is strongly dependent on the task allocation strategy. Also, communication cost can be a significant part of the overall cost of a deduction. For the multiple processor scenario used in this paper,, processors with replicated databases on a broadcast network, a variable supply model (VSM) is presented. VSM represents an infinite class of strategies with varying communication requirements.

A broad range of heuristic programs--embracing forms of diagnosis, catalog selection, and skeletal planning--accomplish a kind of well-structured problem solving called classification. These programs have a characteristic inference structure that systematically relates data to a preenumerated set of solutions by abstraction.

Loosely speaking, recursive inference is when an inference procedure generates an infinite sequence of similar subgoals. In general, the control of recursive inference involves demonstrating that recursive portions of a search space will not contribute any new answers to the problem beyond a certain level. We first review a well known syntactic method for controlling repeating inference (inference where the conjuncts processed are instances of their ruicestors), provide a proof that it is correct, and discuss the con- (Mims under which the strategy is optimal. We also derive more powerful pruning theorems for rases involving transitivity axioms arid cases involving subsumed subgoals. The treatment of repeating inference is followed by consideration of the More difficult prr)liIon of recursive inference Crat does not repeat. Here we show bow knowledge of the properties of the relations involved and knowledge about the contents of the system's database can be used to prove that portions of a search space will not contribute any new .az

COMPUTER SCIENCE DEPARTMENT Stanford University Stanford, California 94305 Commu licaticn and Cooperation Abstract Intelligent agents need to coordinate their actions in pursuit of common goals. When communication is possible, cooperating agents must decide what information to pass in order to agree on a single course of action. This paper outlines several communication strategies (under monotonic and nonmonotonic planning assumptions), proving that some are convergent while others are not. An analysis is also made of the advantages of passing false information. Introduction Recent years have seen increasing interest in Distributed Artificial Intelligence (DAI) systems, that is, in groups of intelligent agents whose members cooperate in carrying out tasks. Considerable work has gone on in this area, producing a number of tentative approaches to cooperation; notable among these research efforts are Smith and Davis' work on the Contract Net [1], Davis' investigations of Cooperative Problem Solving strategies [2], Georgeff's approach to assuring non-interference among distinct agents' plans [3, 41, and Lesser and Corkill's empirical analyses of distributed computation 151. Despite some genuine insights that these researchers have gained, however, DAI has lacked much of the formal foundation needed for progress. Recent work by Appelt 161, Moore [7, 8] and Icon lige 19, 10, 11, 121 has begun to develop the formal descriptions necessary for one agent to reason about another agent's knowledge and beliefs; this is a key step in the development of successful DAI systems. This paper begins to lay the groundwork for another aspect of Distributed Artificial Intelligence's foundation; it presents a description and analysis of information pass:ng strategies between intelligent agents. Through use of a formal descriptive language, certain information passing behavior is proven to be convergent. In addition, an analysis is made of the role that can be played by the passing of false information, i.e., information that is logically inconsistent with the beliefs of the sender. Consider, for example, two individuals who have lost contact with each other in a department store [131.

However, many 3pecialized Many languages have the property that when languages are limited in their expressive power. This some collections of facts are stated explicitly, additional paper presents methods for determining when a set of facts are stated implicitly. We call such languages facts is expressible in a language.