The simBorg Approach to Modeling a Dynamic Decision-making Task

Understanding decision-making processes within dynamic task environments via embodied computational cognitive models proves to be a challenge for the modeling community (see Gonzalez, Lerch, & Lebiere, 2003 for an example). Decisions made by an agent may be the result of explicit, strategic moves, or result from implicit, costbenefit tradeoffs occurring at the level of 1/3 of a second. Understanding and properly modeling decisions occurring at these different levels is a challenge for the modeler. This paper proposes a novel incremental modeling approach (see Byrne, 2001 and Gonzalez et al., 2003 for examples of other incremental approaches) that promises to inform decision theory as well as limitations within the chosen modeling architecture. The simulated cyborg (or, simBorg) approach blends computational embodied-cognitive models of interactive behavior with artificial intelligence based components in a simulated task environment (Gray, Schoelles, & Veksler, 2004). simBorgs combine human and machine components. This combination of high fidelity cognitive modeling (human) and AI (machine) facilitates the development of families of models that allow the modeler to hold components (memory, vision, etc) at different levels of expertise without concern for cognitive plausibility. For example, rather than modeling human problem solving, the modeler can rely on various black-box techniques (i.e., cognitively implausible AI), thereby focusing on predicting how subtle differences in costs and benefits in interactive methods affect performance and errors. The current modeling endeavor adopts the simBorg approach in order to build a family of interactive decisionmaking agents. The following section will present the task of interest followed by a fraction of the family of simBorgs under development. Finally, we conclude with future directions we plan to lead the family of simBorgs.