Knowing how much you don't know: a neural organization of uncertainty estimates

Uncertainty or imprecision is present in all neural computations. It can arise from noise or incomplete information sensed from the environment, and from imprecision and noise in neural circuits.An estimate of such uncertainty can improve behavioural performance.Of the many variables we can be uncertain about, those in sensory perception and in economic outcome prediction have received the most empirical interest.Sensory and outcome uncertainty clearly influence behaviour, and there is evidence to suggest that rule uncertainty also influences behaviour. These influences on behaviour are often close to what optimal algorithms predict.Sensory uncertainty is represented by shallow slopes of firing rates in a neural evidence integrator. BOLD (blood oxygenation-dependent) functional MRI responses in several brain regions can be attributed to sensory uncertainty, but the neural code for sensory uncertainty is unknown.Outcome uncertainty influences prediction error signals generated in the midbrain, and is encoded in the firing rates of orbitofrontal cortex neurons, and possibly in the rate of change in firing of dopaminergic midbrain neurons.BOLD studies of outcome uncertainty await replication to confirm the spatial distribution of outcome uncertainty encoding in the brain.