Metacognitive Control of Linear Dynamic Systems with Self-confidence Adaptation

Metacognitive control is a salient capability of the human cognitive processes that allows them to regulate their executive functions and improve their performance and learning. Here, we contribute towards a mathematical model of metacognitive control for linear dynamic systems with three key features i) control the system, ii) evaluate a second order judgement about the decisions made (confidence in control) and, iii) self-monitor the control confidence and adapt it to improve the controller. We show that adapting the control confidence can improve the controller performance, particularly when the agent is biased with over confidence on its prior beliefs, i.e., it is too confident about its control.