Learning in Amnesia

We compared learning by amnesic and control participants in the “Weather Prediction” task, a widely-used probabilistic categorization task. Earlier studies with the WP task (e.g., Knowlton et al., 1994) found no difference in early categorization performance between amnesic and control participants. This has been taken as evidence for a multiple memory systems view, in which categorization learning is dependent on implicit memory, which is unimpaired in amnesia. However, earlier research used measures of explicit knowledge of questionable validity, and focused only on the outcome of learning, rather than the learning process itself. To overcome these problems, we tapped two relevant aspects of explicit knowledge: task knowledge (knowledge of the relations between cues and criterion) and self-insight (knowledge of the relations between cues and response). If learning depends on implicit memory at least one of these should yield minimal evidence of awareness. By using a dynamic lens model approach, we conducted a fine-grained analysis of learning. The lens model offers a general framework to study information utilization and its relation to the contingencies in the learning environment. It does so by fitting two models simultaneously, one to the cues and criterion, and one to the cues and responses. To study learning, past research has used “rolling regression” models (e.g., Lagnado, Newell, Kahan & Shanks, in press). In the present study, we used associative and Bayesian learning models. These are models for on-line learning, in which cue validities are estimated on a trial-by-trial basis.