Because interoception and Active Inference is the key to the whole system.
Visceral influences on precision-weighted inference. (a) Under the principles of Bayesian Inference, prediction errors are weighted by their statistical confidence (i.e. inverse variance or precision). Both “top-down” predictions and “bottom-up” prediction errors can be more or less precise. In this example, a precise prior belief (in red, relative to likelihood or sensory precision, in blue) has shifted a perceptual prediction error (dotted black line) towards the prediction. Note that the precision of the prediction error itself is the product of both prior and sensory precision; some models posit that an additional system must read out the precision of prediction errors to guide subjective sensitivity and/or confidence. (b) This examplle is expanded to consider how visceral information may enter a low- level multisensory “self-model” combining prior beliefs (red) with visual (green) and tactile (blue) sensory inputs. In this example, visceral signals are inherently more precise than other signals (see text), causing sensory signals to overwhelm prior expectations. (c) Here, highly precise visceral signals strengthen the influence of prior beliefs rather than feed-forward prediction errors. Note that in both examples, the hypothetical hyper-precision of visceral signals causes the prediction error to be highly stable, regardless whether prior or likelihood are weighted.