Expected and unexpected uncertainties control allocation of attention in a novel attentional learning task.

In earlier theoretical work, we postulated that expected and unexpected uncertainties, possibly signaled by the cholinergic and noradrenergic systems, play critical roles in modulating cortical inference and learning. Here, we use a novel attention task to probe the representation and effects of expected and unexpected uncertainties on human inference and learning. The task is an extension of the Posner task, involving a visual discrimination task on a target stimulus, whose location has been indicated by a cue stimulus. As in the classical Posner task, we find that subjects respond more quickly and accurately on valid-cue trials than invalid-cue trials. This differential behavioral measure, termed Validity Effect (VE), is thought to reflect top-down, attentional modulation on visual processing. However, unlike the simple Posner task, which has a constant cue-target relationship, we explicitly manipulate both the identity of the predictive cue and its validity (probability of correctly predicting the target). Cue invalidity induces a form of expected uncertainty, while contextual switching induces a form of unexpected uncertainty. As a control condition, the subjects are told the cue-target rule, as it switches over time, but not the cue validity during each rule. The psychophysics results are consistent with a number of our theoretical predictions. First, VE is found to increase with increasing cue validity, and therefore inversely related to expected uncertainty. Also, VE is smaller for the experimental condition than the control condition, possibly reflecting the increased unexpected uncertainty in the control condition. More interestingly, the slope of VE decrease, as a function of increasing uncertainty, is steeper in the experimental condition than the control