UC Irvine

Humans often fail to accord with the predictions of optimal decision making models. How can we be such poor decision makers in simple task environments and yet also be deft navigators of the real world? This dissertation presents work suggesting that these two observations are likely related: Our decision making strategies have been shaped by the complexity and uncertainty of real-world decision problems, and we apply these strategies even in much simpler decision contexts. In each of the dissertation’s chapters, we consider a presumed suboptimal behavior and demonstrate how it can emerge from rational responses to uncertainty. In chapters 2 and 3, we focus on the case of patch foraging. We show that both adults’ over-exploitation and children’s over-exploration can stem from rational inference of the environment’s underlying structure. Our results reveal that these two opposing behaviors emerge from different structural priors. Finally, in chapter 4, we focus on the reward learning deficits that characterize anhedonia. Using a reinforcement learning model, we show that simulated agents who have rationally adapted to an unpredictable early life environment produce anhedonia-like behavior when later placed in a predictable environment. Collectively, this work demonstrates how multi-scale learning processes work to mitigate the many forms of uncertainty present in real-world decisions. By taking these learning processes into account, we are able to rationalize multiple “suboptimal” behaviors.