ABSTRACT
One particularly fascinating aspect of the mind/brain is the question of how it enables us
to make decisions. I am approaching this question through the use of computational
modeling. In particular, I am developing a neurally plausible connectionist model that is
capable of assimilating information, making inferences, simulating and performing
actions, seeking to fulfill goals, and learning useful behaviors. My immediate goal is to
build a working system that is strongly constrained by the requirement of neural
plausibility. Although I am not trying to model any particular set of psychological data
per se, I do expect ultimately to produce a model that correctly predicts a good range of
psychological data, and more importantly, one that also explains some of the brain
circuitry behind it. The foundation of this connectionist model is SHRUTI [Shastri and
Ajjanagadde, 1993], which demonstrates how a system of neuron-like elements can
encode a large body of relational causal knowledge and provide a basis for rapid
evidential inference. The overall aim of this project is to develop a decision-making
system, SHRUTI-agent. Within this framework, there are several important questions that
are being addressed: (1) How can the existing connectionist representation of belief and
utility be extended to support decision-making? (2) What kinds of control mechanisms
are needed, on top of the existing spreading activation model, in order to enable effective
decision making, and how can these control mechanisms be implemented in a neurally
plausible manner? (3) How can a SHRUTI-agent with a limited world model learn the
right concepts and rules to make decision-making more efficient?