Increasing Realism of Human Agents by Modeling Individual Differences: Methodology, Architecture, and Testbed

A key challenge in creating simulated human agents is to produce sufficiently realistic behavior. A critical component of such realism is the range of variations in behaviors exhibited by humans. These variations are due to a variety of factors, including varying levels of intelligence and skill, differences in cognitive and decision making styles, personality differences, and different affective states: collectively termed individual differences. In this paper we describe an approach for increasing the realism of simulated human agents by explicitly modeling these individual differences. The core component of this approach is a generic methodology for modeling individual differences within symbolic cognitive architectures, via parametric manipulations of the architectural structures and processes. Individual differences profiles are first translated into specific architecture parameters, which then bias the architecture output in corresponding directions, causing variations in performance. The architecture is embedded within a modeling and analysis testbed environment, which supports the simulation of multiple agent interactions within the task simulation; modeling of a variety of agent types by specifying their distinct individual differences profiles; and modeling and analysis of alternative mappings of distinct individual differences onto the cognitive architecture parameters. Prototype versions of the cognitive architecture and testbed have been implemented and evaluated in the context of a military simulation training scenario, representing the interaction of a number of commanders in the context of a specific mission. The simulated mission demonstrated distinct differences in individual commander behavior, as a function of their individual profile, and consequent significant differences in the final mission outcome. The testbed environment was effective in facilitating the rapid selection of alternative individual differences profiles and observation of resulting individual behavior variations and subsequent mission outcome. These preliminary results indicate the ability of the modeling methodology to effectively represent individual differences, and the ability of the testbed to support flexible exploration in the broad research area of individual differences modeling.