Control Limited Perceptual Decision Making

Periods of disengagement are generally observed during perceptual decision-making tasks, but a normative understanding of engagement is lacking. Here, we develop a theory that frames disengagement as a problem in cognitive control. Good performance through task engagement requires control, but control is costly, and this establishes a performance-control tradeoff. We derive decision policies that optimize this tradeoff as a function of the capacity of an agent for cognitive control. When their control ability is sufficiently low, agents lapse. For intermediate control limitations, a new decision-making regime appears where agents don’t lapse, but their behavior is nevertheless shaped by control. We identify hidden signatures of control-limited behavior at the level of accuracy, reaction time and decision confidence which are often observed experimentally, but had not been normatively explained. Our findings provide a path to the study of normative decision strategies in real biological agents.