Information Integration in Perception and Cognition : Exploring the Perception-Cognition Gap

Abstract: Much research in both perceptual inference and cognitive inference has focused on how well human performance fares against what is considered the most optimal or rational thing to do. The verdict, however, differs between the fields. In cognitive inference tasks, people are generally considered to be poor at probabilistic reasoning and prone to irrational biases. In stark contrast to this, human performance on perceptual inference tasks has often been found to be close to that of an optimal agent. This dichotomy is sometimes referred to as the perception-cognition gap. A common explanation of the gap is that perception and cognition operate on different architectures. However, this explanation is problematic, because it is not only the findings that differ between the fields but also the experimental practices and the very definition of rationality. In this thesis I identify five factors on which studies in the two fields deviate from each other: 1) the criterion of rationality/optimality, 2) the complexity of the normative integration rule, 3) implicit or explicit reasoning with probabilities, 4) the origin of the sources of uncertainty, 5) the level at which data are analysed (group vs. single subject). The thesis investigates in three studies to what extent each factor contributes to the gap, by using a combination of experimentation and modelling as main tools, applied to a selection of classical perception and cognition tasks. The findings show that some of the discrepancies between cognition and perception can be explained by the different practices regarding experimental designs and theoretical assumptions. Therefore, part of the gap can be closed by recognizing that cognition is not as irrational as is sometimes claimed and perception not as optimal. However, the findings also suggest that there is a real difference in how humans integrate information in perception and cognition. In the perceptual tasks, participants generally seemed to use more sophisticated integration methods. Moreover, individual differences in strategy selection were substantially smaller in perceptual tasks than in cognitive ones, which suggests that performance on cognitive tasks is more sensitive to the idiosyncrasies of the person. Overall, the results of this thesis suggest that the perception-cognition gap is real and related to different information-processing mechanisms, particularly concerning how information is integrated, in these two systems.