Browse Prior Art Database

Anecdotes: A Very Early Expert System Disclosure Number: IPCOM000129786D
Original Publication Date: 1993-Sep-30
Included in the Prior Art Database: 2005-Oct-07
Document File: 9 page(s) / 37K

Publishing Venue

Software Patent Institute

Related People

Herbert A. Simon: AUTHOR [+2]


Department of Psychology Carnegie Mellon University Pittsburgh, PA 15213 USA

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 11% of the total text.

Page 1 of 9


Copyright ©; 1993 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Used with permission.

Anecdotes: A Very Early Expert System

Herbert A. Simon

Department of Psychology Carnegie Mellon University Pittsburgh, PA 15213 USA

This article describes a seldom-mentioned set of expert systems that have some claim to be regarded as among the first artificial intelligence programs that were actually implemented and used to perform a professional task at an expert level, perhaps the very first. To make good this claim, I must first define what I mean by AI.

Definition of artificial intelligence.

A program is often said to exhibit intelligence if it does tasks that are thought to require intelligence if undertaken by a human being: "Man is the measure of all things." A successful expert system -- that is, a system capable of performing automatically one or more professional tasks -- is, by this definition, an example of AI. However, this simple definition is not quite adequate. We do not call programs that invert matrices, solve linear programming problems, or perform other number-crunching feats AI programs. These are surely "expert" performances, requiring intelligence (and patience!) of any human who would undertake them by hand. We do not generally regard them as AI because the computer is simply following its program. All of the intelligence has been provided by the programmer. Of course, the same charge may be, and has been, leveled against all AI programs. The computer only does what it is programmed to do. But when the computer is programmed to search selectively (heuristically), because it cannot find its answers in a direct, systematic, and guaranteed way without search, then we are more likely to pin the accolade "intelligent," or at least "artificially intelligent," on the program, especially if it is predominantly symbolic rather than arithmetic. Moreover, AI programs do not usually seek optimal solutions (although they may find local optima for subproblems), but look for solutions that are satisfactory, or "as good as possible," and discoverable within acceptable computational times and costs.

Symbolic heuristic search for satisfactory solutions to problems has been the hallmark of artificial intelligence, and when a system performs at an expert level, it is often dubbed an "expert system." In recent years, expert systems have often been described as programs capable of expert performance in some domain, comprising domain knowledge stored in an appropriately organized database, together with an "inference engine" capable of accessing knowledge when relevant and drawing some of the inferences implicit in it.

Since this form of program organization developed only after more than a decade of work in AI and there were a number of earlier expert systems (as defined in the previous paragraph) not organized in this way, we will use the sim...