Information and value during multi-attribute learning and decision making

Abstract

Dissecting the computational components of the explore-exploit dilemma is critical to our understanding of how the mind works. A core component of the dilemma is understanding the contexts where option informativeness is either appetitive or irrelevant. In the present thesis, this computational problem was investigated using a novel multi-attribute bandit task and Bayesian model analyses, observing two critical results. First, a behavioral task was used to probe whether informativeness can defined as a quantifiable variable, as opposed to paradigms in the literature that use a categorical operational definition. Indeed, subjects considered this quantifiable definition of informativeness alongside value. Specifically, analyzing the behavioral experiment with traditional statistics demonstrated signature patterns of exploratory behavior that was consistent with the literature. Second, Bayesian modeling allowed further investigation of potential hypotheses underlying these patterns of exploration – namely, the modulatory role of uncertainty in the deliberation of value and informativeness. There are further questions about informativeness to explore, but this thesis presents a means of investigating and exploring this critical construct on more mathematical grounds.