University College London | UCL · Institute of Neurology
Metacognitive signals for learning and decision-making
Optimal decision-making requires integrating expectations about external outcomes – positive and negative – with current internal states. However in many real-life decisions we do not receive external feedback or rewards, and have to estimate our successes or failures from internal, metacognitive signals. In my talk I will present work on how humans use these learning signals to guide behaviour. I will first present evidence from my PhD in which we found that in uncertain environments, the human prefrontal cortex weighs the pursuit of more reliable rewards against the prospect of larger ones, rather than computing and maximizing a subjective utility function. In my subsequent postdoctoral work, I have investigated whether the ability to internally monitor and evaluate our own decisions can act as a learning signal in the absence of external feedback. In a series of behavioural and neuroimaging studies, I show that human subjects can incorporate local decision confidence to form global self-performance estimates over time, while also pervasively underestimating their performance in the absence of feedback. I will suggest that studying these internal metacognitive signals holds particular promise for shedding light on disorders of mental health. Combining computational models of decision-making and metacognition in a large-scale general population sample, we found dissociable shifts in confidence level and metacognitive ability associated with transdiagnostic psychiatric symptom dimensions, in the absence of alterations in the decision process. These findings represent a first canonical step towards understanding how distortions in self-evaluation are formed and maintained in psychiatric disorders.