An interoceptive predictive coding model of conscious presence

Abstract

We describe a theoretical model of the neurocognitive mechanisms underlying conscious presence and its disturbances. The model is based on interoceptive prediction error and is informed by predictive models of agency, general models of hierarchical predictive coding and dopaminergic signaling in cortex, the role of the anterior insular cortex (AIC) in interoception and emotion, and cognitive neuroscience evidence from studies of virtual reality and of psychiatric disorders of presence, specifically depersonalization/derealization disorder. The model associates presence with successful suppression by top-down predictions of informative interoceptive signals evoked by autonomic control signals and, indirectly, by visceral responses to afferent sensory signals. The model connects presence to agency by allowing that predicted interoceptive signals will depend on whether afferent sensory signals are determined, by a parallel predictive-coding mechanism, to be self-generated or externally caused. Anatomically, we identify the AIC as the likely locus of key neural comparator mechanisms. Our model integrates a broad range of previously disparate evidence, makes predictions for conjoint manipulations of agency and presence, offers a new view of emotion as interoceptive inference, and represents a step toward a mechanistic account of a fundamental phenomenological property of consciousness.

Keywords: agency; consciousness; depersonalization disorder; insular cortex; interoception; predictive coding; presence; virtual reality.

Figure 1

An interoceptive predictive coding model of conscious presence. Both agency and presence components comprise state and error units; state units generate control signals (A_out, P_out) and make predictions [A_pred, P_pred, A_pred(p)] about the consequent incoming signals (A_in, P_in); error units compare predictions with afferents, generating error signals [A_err, P_err, A_err(p)]. In the current version of the model the agency component is hierarchically located above the presence component, so that it generates predictions about the interoceptive consequences of sensory input generated by motor control signals.

Figure 2

The human cingulate (red) and insular (blue) cortices. Image generated using Mango (http://ric.uthscsa.edu/mango/).

Figure 3

A schematic of hierarchical predictive coding across three cortical regions; the “lowest” (R1) on the left and the “highest” (R3) on the right. Light blue cells represent state units, orange cells represent error units. Note that predictions and prediction errors are sent and received from each level in the hierarchy. Feed-forward signals conveying prediction errors originate in superficial layers and terminate in deep (infragranular) layers of their targets, are associated with gamma-band oscillations, and are mediated by GABA and fast AMPA receptor kinetics. Conversely, feedback signals conveying predictions originate in deep layers and project to superficial layers, are associated with beta-band oscillations, and are mediated by slow NMDA receptor kinetics. Adapted from (Friston, ; see also Wang, 2010).

Figure 4

Predictive coding applied to interoception. Motor control and autonomic control signals evoke interoceptive responses [intero(actual)] either directly (autonomic control) or indirectly via the musculoskeletal system and the environment (motor control). These responses are compared to predicted responses [intero(pred)], which are generated by hierarchically organized forward/generative models informed by motor and autonomic efference copy signals. The comparison, which may take place in AIC, generates a prediction error which refines the generative models. Subjective feeling states are associated with predicted interoceptive signals intero(pred). The figure is adapted from a general schematic of predictive coding in (Bubic et al., 2010).