Very often, humans are remarkably good at judging how well they performed - even when nobody told them how they actually did. This ability varies from person to person and from domain to domain. Some people are very confident about their memories and are often correct. Other people are very good at predicting how they will perform in an exam. Some people are just not so good in either of them. The question is: How do people generate this sense of confidence? How can they judge their performance without feedback? And can they learn without external feedback?

In our view, people can generate a perceptual confidence signal by using the amount of visual information that they have gathered from the environment. Some researchers believe that the perceptual decision variable already represents perceptual confidence. For example, in their view the amount of information about whether it is Robert on the other side of the street is identical to our confidence. However, this view does not take into account that this information is still specific to the actual perceptual choice and is in contrast to our understanding of confidence as being choice-independent. Put simply, we can also be confident that it is not Robert, but both times we talk about our confidence.

In our research, we showed that humans indeed carry a choice-independent confidence signal in the ventral striatum. Importantly, we could show that the perceptual decision variable in the cerebral cortex is translated to choice-independent confidence in the ventral striatum. This result is important, because it suggests that the signal in the ventral striatum is not a mere by-product of perceptual confidence (for example that being confident just makes you happy) and that humans indeed have a choice-independent representation of perceptual confidence.

What then is the role of this signal in the ventral striatum? We addressed this question in a recent perceptual learning study where we hypothesized that humans use this confidence signal to learn in the absence of external feedback. We devised a computational model that successfully explained behavioral choices and learning without explicit feedback by using confidence as a learning signal. We showed that this learning signal is also found in the ventral striatum. Importantly, the strength of the striatal modulation explained learning success. Together, this provides an explanation of how humans can learn in the absence of external feedback. Our model extends reinforcement learning models to cases where no external feedback is provided.