Localization of fMRI signal correlating with risk prediction errors in anterior insula (bilateral)
Localization of fMRI signal correlating with risk prediction errors in anterior insula (bilateral)
Our team was the first to identify reward risk prediction errors in fMRI signals of the human brain. These are mistakes that the brain makes in predicting the size of the reward prediction error. We localized these in anterior insula, which raises interesting issues about the role of emotions, because anterior insula at the same time is known to help humans make aware of their emotions (i.e., to make "feelings" out of emotions).
Unlike reward prediction errors (the mistake in predicting the rewards themselves), risk prediction errors have therefore first been observed in the human brain, and not in (individual) neurons of animals (such as monkeys). Reward prediction errors have been associated with the neuromodulator dopamine. There may be an analogous link for risk prediction errors, because of their relation with unexpected uncertainty, which is thought to be encoded in the phasic firing of neurons releasing norepinephrine (noradrenergic neurons). A former member of our team, Kerstin Preuschoff, has recently found some indircet evidence for this link, by studying pupil dilation.
Uncertainty is not monolithic. At a minimum, it consists of irreducible risk (even in the best of circumstances, one does not have perfect foresight), estimation uncertainty (you know what you don't know, and can learn to predict better), and unexpected uncertainty or jump risk (you realize that things have changed – you actually know less than you thought you did). Humans are quite capable of tracking these different facets of uncertainty (sometimes to their detriment, as when humans become ambiguity averse). With Elise Payzan at the University of New South Wales and with Johyn O'Doherty's team, we have been studying the neurobiological foundations for this capacity.