SEECS Colloquium

Decision making is influenced by a variety of factors. These include cognitive (general intelligence, specific skills), personality and temperament traits (aggressiveness, conscientiousness), and affective states (emotions and moods). Together, these factors are termed individual differences or behavior moderators. Modeling their effects within cognitive architectures is critical, both for the development of more realistic agents, and for the advancement of our understanding of perception, cognition and decision making.

In this talk I will describe research we have been conducting in the general area of individual differences modeling. First, I will describe a generic methodology for modeling individual differences within a symbolic cognitive architecture. The core component of the methodology is a highly-parameterized architecture, with parameters controlling both structure and processing. This parameter space then provides a means of encoding the effects of specific configurations of individual differences on a variety of perceptual and cognitive processes. Second, I will describe the cognitive architecture that implements the methodology, and associated representational and inferencing requirements. Third, I will discuss a computational modeling testbed and simulation environment we have developed, which supports the definition of multiple agents, whose behavior is controlled by instances of the architecture. Distinct agent profiles map onto distinct configurations of the architecture parameter space, which then produces variations in behavior. The testbed serves two purposes: First, it allows the creation of scenarios where distinct agent types are exposed to identical simulated situations, and the resulting variations in outcome can be observed and analyzed. These scenarios can be used for training and assessment purposes. Second, the testbed environment supports interactive construction and exploration of alternative models of the detailed mechanisms of individual differences effects. This supports the generation of specific empirical hypotheses regarding these mechanisms, and the evaluation of empirical data within a computational model. I will conclude with a brief example illustrating the methodology and the testbed functionality, in the context of a military training scenario: a Stability and Support Operations scenario.

EVA HUDLICKA is a Principal Scientist and President of Psychometrix Associates, Blacksburg, VA. Her research interests include cognitive modeling, affective computing, decision support system design, and human-computer interaction. She received her BS in Biochemistry from Virginia Tech, MS in Computer Science from The Ohio State University, and PhD in Computer Science from the University of Massachusetts-Amherst. Prior to founding Psychometrix Associates in 1995, Dr. Hudlicka was a Senior Scientist at Bolt Beranek & Newman, Cambridge, MA.