Fig. 4.

Predictive processing account of alien limb. A) Normal behaviour relies on hierarchical brain networks encoding predictions (blue) and prediction errors (red). Prediction units are ensembles of neurons whose activity encodes the probability distribution of anticipated action outcomes. The predicted outcome is encoded in increasing level of complexity and abstraction in the posterior-anterior brain axis. At each level of the hierarchy, predictions from a higher level are compared with predictions from the current level by the error units. Prediction errors are forwarded up the hierarchy, while predictions propagate down the hierarchy. In movement, prediction errors are minimised by changing bottom-up sensory evidence. Specifically, predictions that are afforded sufficient precision propagate down the hierarchy where they induce movements by the motor neurons through a mechanism similar to the classical reflex arc, where predictions about proprioceptive signals are compared with actual proprioceptive input. B) We propose that networks in medial brain regions encode predictions of the precision of predictions in lateral regions (light blue). The principal mechanism in alien limb is then an impairment in the lateral connections between medial and lateral regions, either because of white matter deficits or damage to grey matter in medial frontal regions (we do not formally distinguish between damage to grey matter or white matter). Similarly, damage to lateral regions such as posterior parietal cortex, can afford certain beliefs about action outcomes (i.e., affordances) with high precision. This leads to unmoderated, abnormally high precision of certain outcomes that are not consistent with motivational and/or contextual relevance.