Expected value of information overlaps with reward circuits in humans

Abstract: The ability to identify useful information, and selectively extract it, is critical for perception and cognition alike. For instance, to discriminate between object categories, the visual system must direct the eyes' gaze to useful features. What are the neural substrates of assessment of features' relative expected information value? We trained 10 subjects on a probabilistic categorization task in which simulated plankton stimuli are classified as either species A or B, using experience-based learning. Subjects classified randomly sampled specimens, and received immediate feedback. Each plankton specimen has two two-state features (open or connected claw; dotted or non-dotted eye) that probabilistically predict the species. The moreand lessuseful features, if viewed individually, lead to 85% and 60% accuracy, respectively. Since all combinations of features occur in both species, 100% accuracy is not possible. Following behavioral training, subjects participated in an event-related fMRI experiment in which they were cued to anticipate moreor less-useful information. Anticipation of information was induced by first presenting an obscured version of the highor low-usefulness feature, such that subjects could not yet categorize the stimulus or prepare a specific motor response. The specific form of the feature (e.g. dotted eye) was then revealed, allowing categorization. Finally, subjects received feedback on whether their classification was accurate, which was probabilistic Abstract Print View http://www.abstractsonline.com/Plan/AbstractPrintView.aspx?mID=255...