CASOS Working PAPER

"Talent on Task: Core Decisions and Agents"(PDF file)
Authors: Hunter, K., Larkey, P. D.

Abstract
Almost fifty years ago Herbert A. Simon, Allen Newell and J.C. Shaw produced the Logic Theorist, a procedural program capable of solving a modest range of problems in formal symbolic logic (Psychological Review, 65: 151-66, 1958). The research strategy was novel and direct: pose a problem and build a symbolic agent capable of solving the problem.
A few years later, Richard M. Cyert, James G. March and others adapted this strategy to model price and output determination in a business firm (A behavioral Theory of the Firm, Prentice-Hall, 1963).
In the intervening years the strategy has been utilized by many researchers in many domains including laboratory puzzles, games, language acquisition, physics, chemistry, and resource allocation processes in public and private organizations. Some of this work is grounded in the careful study of subjects solving the posed problems. Other work is grounded more in the researchers' introspections on how they would solve the problem. Some of the work, especially the organizational work, is very abstract with no clear empirical referents. Other work never gets around to building the agents; once the problem is sharply posed, why fiddle around trying to understand how lesser mortals might work the problem when you can move directly to developing powerful heuristic solutions, perhaps even optimal solutions?
For understanding the behavior of individuals and organizations, the strategy has much to recommend it. Simon's original interest, for example, was in the bounds on rational behavior. Such bounds become very clear with the discipline of building an explicit model that must acquire and process information to achieve a particular result.
The purpose of the work reported here is to model agents performing an organizational task common to virtually all organizations that produce goods and services. A large portion of the initial effort is devoted to developing a computerbased architecture that will permit rich experimentation with agents and context.