The Redundancy Effect in Human Causal Learning

Abstract

Everyday life requires us to make judgments based on indirect and fragmental evidence. This is especially true when we acquire causal knowledge, where potential causes are often correlated, redundant, and thus highly ambiguous. The complexity of ambiguous causal learning situations is epitomised by a recent human learning phenomenon known as the redundancy effect. It refers to the comparison between two types of redundant cue: a blocked cue, which predicts a meaningful outcome but supplies no new or unique information about its occurrence, and an uncorrelated cue, which is always less informative about the outcome than the cues with which it coincides. While the redundancy effect represents a novel empirical test of classic theories of associative learning, which appear to be broadly challenged by the different ways that people learn about redundant cues, proposition-based reasoning via higher order cognition could shed light on the causality judgment process and how it varies under different a priori assumptions. The current thesis took an integrated approach to understand differential learning about redundant cues in a range of ambiguous conditions. To explore the involvement of simple associative mechanisms, Chapter 2 conducted computational modelling of various candidate models. To examine the role of inferential reasoning, Chapter 3 manipulated prior assumptions necessary for deductive inference. Chapter 4 examined subsequent tendency to produce a judgment bias referred to as theory protection. Chapter 5 investigated the role of learning context. The results are consistent with a multi-process explanation for the redundancy effect, suggesting that learning and memory processes driven by prediction error and propositional reasoning are both important determinants in human learning under ambiguity.