Attention and Prediction Error as Mechanisms for Theory Protection?

Abstract

Recently, a principle known as theory protection has been proposed to explain theway people bias the updating of their beliefs when they encounter new information about ambiguous cues. This principle presents an alternative to the proposal that a combination of individual and summed prediction error contributes to learning in situations where combinations of predictive cues are presented as potential causes of an outcome. Here, we discuss similarities between the notion of theory protection and attention shifting models of learning that assume attention is guided by individual prediction error. We report simulations using a prominent attention shifting model in the category learning literature and show that it accounts for several of the key examples of theory protection. The basis of these learning biases, hypothesized to be determined by either cue uncertainty or prediction error, is yet to be determined and requires further tests that dissociate these factors more clearly. It may be the case that theory protection is better understood as an organizing principle for knowledge updating rather than a single psychological mechanism.