Dr. Rasmus Bruckner: Decomposing the influences of uncertainty on learning: Normative computations, uncertainty biases, and lifespan differences

Abstract

Learning often takes place in environments in which it is impossible to know current and future outcomes exactly. To successfully behave in such uncertain environments, humans have to learn appropriate beliefs from past experiences to predict desirable and undesirable outcomes. Drawing on optimal inference models and behavioral learning tasks, I will illustrate how learning under uncertainty should be regulated from a normative perspective and how learning biases may emerge from deviations from these computations. I will show work on reward-based learning under perceptual uncertainty. This study suggests that humans dynamically combine uncertain perceptual and reward information but that categorical choices substantially modulate this integration. Moreover, I will present two lifespan studies on learning in dynamically changing environments. This line of work suggests that age-related learning differences arise from more liberal satisficing policies in children and older adults – that is, accepting predictions that achieve some sufficient level of accuracy. Finally, I will explore the possibility that these learning biases emerge from simplified learning strategies that might require lower cognitive resources than optimal solutions.