Suboptimal Criterion Learning in Static and Dynamic Environments

Elyse H. Norton, Stephen M. Fleming, Nathaniel D. Daw, Michael S. Landy

Abstract: Author Summary Understanding how humans make decisions based on uncertain sensory information is crucial to understanding how humans interpret and act on the world. Signal detection theory models discrimination and detection decisions as a comparison of “stimulus strength” to a fixed criterion. In a world that is constantly changing a static criterion makes little sense. We investigate this as a problem of learning: How is the decision criterion set when various aspects of the context are unknown (e.g., category means and variances)? We examine criterion learning in both static and dynamic environments. In addition to a more traditional discrimination task in which the criterion is a theoretical construct and unobservable, we use a novel task in which participants must explicitly set the criterion before being shown the stimulus. We show that independent of environment and task, observers dynamically update the decision criterion, even after prolonged training in a static environment. Our results provide evidence against an assumption of stability and have implications for how psychophysical data are analyzed and interpreted and how humans make discrimination decisions under uncertainty.