Abstract
The world is largely stable and predictable. Humans and other organisms are sensitive to that stability, and use it to support cognitive processes. This work consists of a series of studies that explore how humans learn about such stability, use that information to generate predictions about forthcoming sensory input, and detect when such predictions are inadequate. First, I present a modeling study that quantified the distributions of errors that people make on a complex, visuomotor sequence learning task, and examine the serial position dynamics of several parameters describing short-term visual memory. Both precision and capacity for these sequences increases with familiarity, and the worst-represented items show the largest increases. Next, I present an experiment that used the same task to understand the effects of deviant items within familiar sequences. By measuring ERPs to new, familiar, and deviant items, I dissociate the neural activity associated with detecting a deviant from that associated with encoding task-relevant stimulus characteristics. Finally, I present an experiment investigating the role of prediction in a task that is stochastic, rather than sequential, and in which deviant events occurred among the distractors rather than among the task-relevant stimuli. Unexpected events among the distractors seem to obligatorily attract attention, enhancing or impairing performance. Further, I show that the individual differences in the neural response to such unexpected events is predicted by temperament. Together, these studies illuminate how the brain learns about predictability in a range of settings, and leverages such predictability to facilitate cognition.