ICCM: Learning

People are capable of learning diverse functional relationships from data; nevertheless, they are most accurate when learning linear relationships, and deviate further from estimating the true relationship when presented with non-linear functions. We investigate whether, when given the opportunity to learn actively, people choose samples in an efficient fashion, and whether better sampling policies improve their ability to learn linear and non-linear functions. We find that, across multiple different function families, people make informative sampling choices consistent with a simple, low-effort policy that minimizes uncertainty at extreme values without requiring adaptation to evidence. While participants were most accurate at learning linear functions, those who more closely adhered to the simple sampling strategy also made better predictions across all non-linear functions. We discuss how the use of this heuristic might reflect rational allocation of limited cognitive resources.

In order to gain insight into how people acquire certain reference biases in language and how those biases eventually influence online language processing, we constructed a cognitive model and presented it with a dataset containing reference asymmetries. Via prediction and reinforcement learning the model was able to pick up on the asymmetries in the input. The model predictions have implications for various accounts of reference processing and demonstrate that seemingly complex behavior can be explained by simple learning mechanisms.

Often we find ourselves in unknown situations where we have to make a decision based on reasoning upon experiences. However, it is still unclear how people choose which pieces of information to take into account to achieve well-informed decisions. Answering this question requires an understanding of human metacognitive learning, that is how do people learn how to think. In this study, we focus on a special kind of metacognitive learning, namely how people learn how to plan and how their mechanisms of metacognitive learning adapt the planning strategies to the structures of the environment. We first measured people's adaptation to different environments via a process-tracing paradigm that externalises planning. Then we introduced and fitted novel metacognitive reinforcement learning algorithms to model the underlying learning mechanisms, which enabled us insights into the learning behaviour. Model-based analysis suggested two sources of maladaptation: no learning and reluctance to explore new alternatives.

In recent years, several models of human reinforcement learning have been proposed that balance rationality (maximizing expected utility) against cognitive costs. Lai and Gershman (2021) proposed a model in which the cognitive cost was assumed to be the policy complexity, defined in terms of information theory as the mutual information between the sensory input and behavioral response. Here, using evidence from a published data set (Collins & Frank, 2012), we show that this model fails to account for the ''set size effect'' in learning: humans’ learning efficiency decreases when the number of the presented stimuli increases. We therefore propose an alternative computational model, in which cognitive cost constitutes not only the policy complexity, but also the representation complexity---the amount of information conveyed from sensory inputs to internal representations. We quantify information processing cost as the combination of representation complexity and policy complexity. The resulting model captures the set size effect in an instrumental learning paradigm.