Abstract
A rich body of experimental data indicates that the neuromodulatory systems acetylcholine (ACh) and norepinephrine (NE) are crucially involved in a variety of cognitive tasks. However, there is little consensus on their independent and joint computational functions. We present a theory in which cortical ACh and NE report different aspects of uncertainty: ACh reports expected uncertainty, for instance coming from known variability or ignorance about the parameters of a task, and NE signals unexpected uncertainty, as when significant aspects of the task are unpredictably changed by the experimenter (Yu & Dayan, 2002). These different sorts of uncertainty should, according to statistical learning theories, interact in a specific way to control the integration of top-down and bottom-up information. Here, we apply these ideas to a new spatial attention task (Bentley, personal communication), which is an extension of the classical Posner Task to contextual cueing. In our model of the task, ACh and NE interact in a precisely specified and partly OPPONENT, partly SYNERGISTIC manner, as in examples of simulated pharmacological manipulations. The agreement between simulation results and existent data (Phillips et al, 2000) is remarkably good (Figure: a,b:data; c,d:model). The model makes specific, experimentally tractable predictions regarding trial-to-trial responses of the ACh and NE neurons, as well as certain psychophysical measures such as reaction time. If confirmed, these predictions could yield insights into the workings of these neuromodulatory systems, as well as a new perspective on spatial attention.
