Abstract
Bounded temporal accumulation of evidence is a canonical computation for perceptual decision making (PDM). Previously derived optimal strategies for PDM, however, ignore the fact that focusing on the task of accumulating evidence in time requires cognitive control, which is costly. Here, we derive a theoretical framework for studying how to optimally trade-off performance and control costs in PDM. We describe agents seeking to maximize reward rate in a two-alternative forced choice task, but endowed with default, stimulus-independent response policies which lead to errors and which also bias how speed and accuracy are traded off by the agent. Limitations in the agent’s ability to control these default tendencies lead to optimal policies that rely on ‘soft’ probabilistic decision bounds with characteristic observable behavioral consequences. We show that the axis of control provides an organizing principle for how different task manipulations shape the phenomenology of PDM, including the nature and consequence of decision lapses and sequential dependencies. Our findings provide a path to the study of normative decision strategies in real biological agents.
Competing Interest Statement
The authors have declared no competing interest.
