ABSTRACT
In this paper we present an initial specification of a general, robust, and efficient computational framework for tracking cognitive processes — that is, inferring a persons’ thoughts from their actions. Our framework, which we call the mind-tracking architecture, centers on two core processes: generating predicted cognitive and action sequences using computational cognitive models, and tracking observed actions through robust matching with predicted actions. In essence, the mind-tracking architecture “thinks along” with the person in predicting a set of possible thoughts and actions, and then matches these to the person’s observed actions to infer their most likely thoughts. In the paper we provide a background of related work (e.g., for intelligent tutoring systems), outline the basic components of the architecture, and demonstrate its usefulness for a sample real-world application — real-time inference of driver intentions.
