Adaptive decision making critically depends on agents’ ability to reduce uncertainty. To reduce uncertainty, agents need to perceive relevant environmental signals, extract useful information, and incorporate it into their knowledge and behavior. This is not a trivial task for two reasons. First, the environment is full of sensory stimuli, but most of them are irrelevant. Agents need to evaluate relevance of signals so that they can prioritize perception and processing of the most relevant ones. Second, perceived signals can be analyzed in countless ways, but most of them do not provide useful information. Agents need to be selective in how to process perceived signals before trying to make use of them in behavior.
This thesis reports three attempts to understand neurocognitive mechanisms of uncertainty reduction in value-based decision making. It is relatively well understood how human brain produces decision-making behavior under uncertainty. It is also well understood how human brain perceives and processes sensory stimuli. However, little is known how these systems cooperate to enable adaptive uncertainty reduction. Three chapters in this thesis try to fill this gap by behavioral experiments, functional magnetic resonance imaging (fMRI) experiments, and computational modeling.
Chapter 1 examines the way human brain processes perceived signals under conditions of reducible and irreducible uncertainty. Because not all uncertainty can be reduced in the same way, agents should take into account the nature of uncertainty they are facing. This challenges a widespread idea that uncertainty reduction is driven by the extent to which signals violate prior expectancy. It is behaviorally shown that subjects are sensitive to reducibility of uncertainty, and it can be quantitative characterized by a Bayesian model where agents ignore expectancy violations that do not update beliefs or values. Furthermore, fMRI results reveal that neural processes underlying belief and value updating are separable from responses to expectancy violation, and that reducibility of uncertainty in value modulates connection from belief- to value-updating regions. These results illustrate how human brain uses knowledge on uncertainty to process signals adaptively.
Chapters 2 and 3 examine the way human brain evaluates incoming signals’ relevance to economic choices. Human often seeks for information that is irrelevant to the task at hand. Although this challenges a normative account in which agents assign positive value only to instrumental information, it has been unclear how general and important such preference is in decision-making settings. Chapter 2 reveals that, behaviorally, subjective value of information exceeds the normative prediction not only when information is non-instrumental but also when it is instrumental. Overvaluation depends on outcomes at stake, rejecting a popular theory of internal drive for entropy reduction. Observed valuation bias can be explained by introducing recursive utility, which penalizes choices under the lack of information, to the normative account. Using this novel model and fMRI, Chapter 3 reports that activation in striatum represents value of information, unifying information’s relevance to the task and penalty for uninformed choice. These results propose a new neurocognitive account on how human brain acquires information with various degrees of uncertainty reduction.