Humans learn by updating their expectations based on the feedback they have encountered in the past. For instance, we may update how good we think we are at our job following praise we receive from colleagues.

However, in many situations we do not receive regular feedback about our performance. For instance, when choosing a career, we might reflect on our skills and abilities to find the option that looks like a good fit, even when feedback and appraisals of our ability are unavailable. While we know a lot about how the brain learns from explicit feedback (e.g. rewards and punishments), we know much less about how it learns from self-evaluation.

During my postdoc at the Wellcome Centre for Human Neuroimaging at UCL, we set out to investigate how people learn about their abilities in the absence of feedback, with my co-authors Peter Dayan and Steve Fleming. Peter is an expert on models of learning and Steve is an expert on confidence and metacognition, the ability to evaluate and monitor our cognition. Working with them both gave me complementary perspectives on this question, and was very fruitful in creating an initial bridge between metacognition and mathematical models of learning, two research areas that were originally rather independent.

We thought that metacognition may be particularly useful when learning about our abilities because many real-life decisions lack immediate feedback. In our experiments, participants performed mini blocks of two tasks, and at the end they were asked to report on which task they think they did better. We found that human subjects indeed use fluctuations in confidence to learn about their task ability over time, confirming a link between metacognition and learning. Unexpectedly, however, we also found that they were often pessimistic about their performance in the absence of feedback.

Why is this important?

We believe that understanding the formation and maintenance of beliefs about our abilities – what psychologists call self-efficacy – is very relevant for thinking about mental health. Distortions in self-evaluation are pervasive in many psychiatric disorders. This altered self-evaluation can take many forms, such as overconfidence and underconfidence in one’s own abilities. In previous work [here], we have quantified how shifts in confidence (but not objective performance) are related to a large range of personality traits and mental health symptoms. 

For instance, people with anxious and depressive symptoms, who hold a generally lowered sense of confidence, may not update their global self-beliefs in the face of new positive experiences. In addition, they may be more likely to generalize low self-beliefs across multiple areas of life (as suggested here). We are currently testing some of these hypotheses in ongoing studies, in the hope that understanding how global beliefs are learned and updated will pave the way towards a better model of these disorders.

Dr. Marion Rouault

Postdoctoral Research Associate at UCL

I am a Postdoctoral Research Associate at the UCL Wellcome Centre for Human Neuroimaging. I am interested in the neurocognitive mechanisms underlying human metacognition, learning and decision-making.