Speaker: Modesto Orozco Ruiz (ETH Zürich)

Title: Metacognition: between monitoring and interpreting self-efficacy

Abstract: Metacognition is the conscious awareness of one’s thoughts-thinking about thinking. This abstract concept embraces all the mental phenomena and estimations about self-control and reflective monitoring, whose final goal is maintaining a self-referential check and balance. Therefore, metacognitive processes are defined as second-order behaviours since they represent decisions derived from lower level behavioural outputs. Hence, metacognition pertains to the self-monitoring of one’s level of mastery in acting against perturbations. Recently, in computational neuroscience, level of confidence has been proposed as the measurable quantity for modelling metacognition computationally. Thus, confidence, or self-efficacy, should be a derived quantity from the performance on different first-order tasks. Generally, mental health and perceived self-efficacy are closely related. Thus, mental illnesses such as depression, bipolar disorder and other psychological alterations are thought to appear in agents (usually genetically predisposed) with an altered sense of self-efficacy. In this project we work on the assumption that healthy and unhealthy mental states are physical attractors, meaning that transitioning from one to another in either direction is not trivial. This reasoning has lead us to hypothesize that: (i) self-efficacy could be described by a bi-modal distribution where high self-efficacy and low self-efficacy should correspond to two attractive states. The more confident you are, the less probable it is to adopt low self-efficacy states and vice-versa; (ii) since metacognition is applied to judge all the first-order adaptations, multiple self-efficacy estimations are computed. Global self-efficacy results from a combination of that individual experiences. In this project we used a hamiltonian-based theoretical model to derive self-efficacy estimates, from several simulated exercises. Using bayesian inference, we studied how confidence behaves under different scenarios, from learning or non-adapting episodes to stressful situations. Finally, we explored how these scenarios affect our model’s self-efficacy belief and how it changes with time.