CASOS Working PAPER

"Generating Realistic Heterogeneous Agent; Computing Confidant-based Base Interaction Probabilities (PDF file)
Author: Alex Yahja & Kathleen M. Carley

Abstract
Organizational theory often assumes classes of agents, for example classes of the operating core, the middle line, the strategic apex, the techno-structure, and support staff. While these classes are adequate for conventional analysis of organization, they are inadequate for more sophisticated analysis using computer modeling and simulation. Furthermore, in reality, organizational agents - humans, software agents, webbots or robots - are very complex and exhibit a plethora of behaviors. Modeling such agents with sufficient accuracy is a challenge. We take a statistical approach to model the behavior of these agents in this paper. We are modeling the interaction probabilities between two categories of agents so that the number of confidants (the number of people each agent interacts with most) matches the empirical population data about confidants. A gradient descent approach is used and the results are presented indicating the efficacy of the approach. This work represents the first step in generating heterogeneous organizational agents based on empirical data and its use in enhancing the evaluation of organization theory by computer-enabled theorizing.