Affect, Learning, and Choice

Abstract

Affect is closely intertwined with learning and decision making, yet these topics tend to be studied separately: affective science considers affective processes, while decision science studies value-based choice and the learning processes that inform it. In this dissertation, I investigate the relationship between a foundational construct in affective science – affective valence – and key topics in decision science – value and RL – in order to bring insight into both fields, and into the many areas of psychological research they impact. First, I report a series of human behavioral studies that examined affective experience during RL tasks. Applying novel data analytic methods, I demonstrate that multiple distinct classes of RL-related computations make unique contributions to affective valence, and that valence makes a key contribution to RL: affect reinforces behavior alongside external reward, such that people prefer choice options that lead to more positive affect in addition to preferring those that lead to greater reward. Next, I review evidence on how affective valence emerges from valuation processes. Based on this review, I argue that the valence of the affective response to a stimulus depends on its utility (i.e., general desirability) relative to a comparison point, and that this dependence suggests that affective valence reflects either reference-dependent valuation or reinforcement processes. I then report a second series of behavioral studies that examined affective experience during classical conditioning. These studies provide initial evidence that affective valence corresponds to the (reference-dependent) value of the eliciting stimulus, rather than its reinforcing effects. Finally, I synthesize these findings into a general account of affect, value, and RL – one which frames affective valence as a mental representation of value that guides behavior alongside external rewards.