I Predict, Therefore I Am: Perturbed Predictive Coding Under Ketamine and in Schizophrenia

Predictive coding involves inferences about current sensations based on past experience. We infer what would need to be present (externally and internally) to cause the sensory information we detect (1). The stimuli whose causes are being inferred have transducers (e.g., photoreceptors in the eye, hair cells in the ear) that convert sensory energy into neural activity, which is processed hierarchically within streams that may be relatively encapsulated. However, ultimately these hierarchies converge on a coherent integrated percept (1). One crucial complexity, with relevance to the symptoms of psychotic illness, is that we change the sensations we receive from the world when we move through it (1). To accurately infer the causes of sensory inputs incident on us, we must take that into account (1). We do so by modeling ourselves as agents in the world, predicting and explaining away the sensory consequences of our actions (1). Failure to make such predictions can drastically change what we infer about ourselves and the world, manifest ultimately as psychotic symptoms—the profound departures from consensual reality that characterize serious mental illnesses such as schizophrenia (2). Kort et al. in this issue (3) shed important new mechanistic light on these consequences. They find that ketamine, the N-methyl-D-aspartate (NMDA) receptor antagonist that engenders psychotomimetic effects, perturbs the electrocortical signature of prediction emitted when one speaks in a manner redolent of that observed in schizophrenia. The extent of the perturbation correlated with the severity of ketamine-induced depersonalization, derealization, and aberrant salience that may model the earliest phases of psychotic symptom formation (2). In predictive coding, the task of inferring causal agency for some event involves conditioning the evidence (whether the event occurred) over the priors (whether there was an intention to act). Priors and evidence are weighted according to their precision (1). Imprecise priors produce prediction errors that render the consequences of actions indistinguishable from externally generated sensations, demanding odd conclusions such as one’s own movements, speech, or even thoughts being externally controlled. Tickling underlines the role of prior expectation in sensory attenuation. Usually we cannot tickle ourselves because we predict and cancel the sensory consequences of our own actions. However, patients experiencing passivity phenomena rate self-generated stimulation as ticklish. Likewise, for selfproduced forces, we cancel out the predicted sensory consequence such that, when matching a force that we have just experienced, we overcompensate to overcome the cancellation due to our priors. Patients with schizophrenia are more