Figure 2. Bidirectional Influences of Information Sampling and Learning.

Note. In this example, a decision maker has learned to categorize stimuli, which vary according to three dimensions (one that is highly informative, one that is moderately informative, and one that is irrelevant), into two categories (denoted by blue crosses and red circles) by actively sampling information from the external world (110 stimuli, randomly drawn from this imaginary world, are illustrated at the right). Their knowledge of the world (depicted as two probability distributions at left) reflects the samples that have been observed. In this example, the decision maker has learned that the “highly important” dimension predicts the category label, but has not learned that the “moderately important” dimension mediates this relationship. As a result, this learner would be unable to classify all stimuli accurately. Characteristics of the external world (e.g., costs associated with sampling each dimension, or costs associated with incorrect choices), as well as characteristics of the learner (e.g., some learners might show a stronger bias for simple hypotheses) influence what is ultimately learned. See the online article for the color version of this figure.