Figure 4.

Proposed Insula Function (in black). Insula receives interoceptive bottom–up signals as well as top–down predictive signals. The latter emerge as the output of a “reference model” that forecasts interoceptive signals. Anterior insula compares the bottom–up and top–down signals and produces a surprise signal when the discrepancy is larger than expected. Upon surprise, the decision-maker is made aware of the need to learn and adapt. This involves updating the reference model (black arrow pointing up), actively interfering in the environment (black arrow pointing to right) and emotional preparation for future changes in the environment (black arrow pointing down). Some time later, the environment (green) produces new stimuli, which in turn affect interoceptive signals. The resulting bottom–up signals are again compared to (top–down) predictions from the updated reference model. If the size of prediction errors is as expected, there is no more surprise, and no more learning and adaptation.