UC Irvine

Normative models are powerful tools for advancing cognitive science. But where do their norms come from, and how well do they align with other constraints on the target system? This dissertation examines these questions within the family of sequential sampling models used to study the speed-accuracy tradeoff in decision-making. Across three chapters, we show how relaxing a common assumption generates normative solutions that require within-trial variability in the value of two key parameters (drift rate and decision threshold), and—critically—that the dynamics of this variability are governed by observers’ evolving belief about the reliability of evidence bearing on choice.We begin by specifying a novel model formalizing the role of memory retrieval dynamics in biasing decisions when observers have uncertainty about prior probabilities, and show that it can reproduce dynamic effects of prior probability observed across different tasks, species, and neuroimaging modalities. Then, we report empirical data from a novel task designed to test the key prediction of our model—that choices ought to be most strongly biased by prior probability at the moments in time when sensory evidence is maximally uncertain—and report direct evidence in support of this prediction. Finally, we present a “philosophical toolkit” forming the conceptual foundations of this work and demonstrate its utility by using it to re-analyze a long-standing debate about time-varying decision thresholds. Taken together, this work jointly illustrates how normative models offer principled starting points for developing and testing theories about neural and cognitive function, and how philosophy offers meta-scientific support toward that end.