Theroy of Mind in Sigma
Abstract
A cognitive architecture is a hypothesis about the fixed structure underlying intelligent behavior, whether in natural or artificial systems. The diversity of intelligent behavior supported in such architectures may be due to either their comprising a relatively large number of specialized mechanisms or their supporting flexible interaction among a relatively small number of general mechanisms. Both of these strategies have been followed in building integrated models of central cognitive processes, although the choice generally embodies an implicit tradeoff between ease of achieving broad coverage and theoretical elegance.Over the past few years, in an attempt to better combine broad coverage with theoretical elegance, an approach to cognitive architectures has been under investigation that is based on combining a small set of general mechanisms at the implementation level beneath the architecture – based on the mixed (relational + probabilistic) and hybrid (discrete + continuous) capabilities of graphical models – with a diversity of capabilities at the architecture level. A fundamental challenge for any cognitive architecture built out of a small set of general mechanisms is whether the limited set of mechanisms can cover the broad range of human cognitive capabilities. We propose to consider this challenge from the perspective of a key human cognitive capability – Theory of Mind reasoning, the ability to ascribe and maintain beliefs about the mental states and processes of others as well as to exploit those beliefs. This ability plays a key role in social interaction from an early age, with beliefs about others informing decision making and profoundly impacting emotional and behavioral responses in interpersonal interactions.Theory of Mind calls on a wide variety of cognitive capabilities, making it an excellent challenge with which to test the representational and reasoning sufficiency of a cognitive architecture. To address this challenge, we will be building on existing work on decision theoretic control within the graphical cognitive architecture, but extending it to handle recursive multiagent modeling, belief revision upon mismatches between predicted and observed behavior, and Theory of Mind reasoning with communication. In the process we will be both evaluating the graphical architecture’s ability to handle these capabilities andexploring potential benefits of realizing Theory of Mind within the graphical architecture. Ultimately, the existence of a broadly functional cognitive architecture with an effective capability for Theory of Mind raises the possibility of automated systems that can: help in assessing complex multiagent situations and predicting what will come; engage in psychological operations while being less likely to be deceived by such attempts on the part of others; andform the basis for more socially sophisticated intelligent agents and virtual humans in simulation and training.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512468
Entities
People
- Paul Rosenbloom
Organizations
- Office of Naval Research
- United States Navy
- University of Southern California