Active interoceptive inference and the emotional brain

Abstract

We review a recent shift in conceptions of interoception and its relationship to hierarchical inference in the brain. The notion of interoceptive inference means that bodily states are regulated by autonomic reflexes that are enslaved by descending predictions from deep generative models of our internal and external milieu. This re-conceptualization illuminates several issues in cognitive and clinical neuroscience with implications for experiences of selfhood and emotion. We first contextualize interoception in terms of active (Bayesian) inference in the brain, highlighting its enactivist (embodied) aspects. We then consider the key role of uncertainty or precision and how this might translate into neuromodulation. We next examine the implications for understanding the functional anatomy of the emotional brain, surveying recent observations on agranular cortex. Finally, we turn to theoretical issues, namely, the role of interoception in shaping a sense of embodied self and feelings. We will draw links between physiological homoeostasis and allostasis, early cybernetic ideas of predictive control and hierarchical generative models in predictive processing. The explanatory scope of interoceptive inference ranges from explanations for autism and depression, through to consciousness. We offer a brief survey of these exciting developments.This article is part of the themed issue 'Interoception beyond homeostasis: affect, cognition and mental health'.

Keywords: cybernetics; emotion; interoception; neuromodulation; predictive coding; self.

Figure 1.

Inference and perception across different modalities. Green arrows represent exteroceptive predictions and prediction errors underlying perception of the external world. Orange arrows represent proprioceptive predictions (and prediction errors) generating action through active inference. Blue arrows represent interoceptive predictions (and prediction errors) underlying emotional processing and autonomic regulation. Integrated experiences of embodied selfhood emerge from the joint hierarchical content of self-related predictions across all these dimensions, including—at hierarchically deep levels—multimodal and amodal predictions. Adapted from Seth [7]. (Online version in colour.)

Figure 2.

This figure summarizes the hierarchical neuronal message passing that underlies predictive coding. The basic idea is that neuronal activity encodes expectations about the causes of sensory input, where these expectations minimize prediction error. Prediction error is the difference between (ascending) sensory input and (descending) predictions of that input. This minimization rests upon recurrent neuronal interactions between different levels of the cortical hierarchy. Current interpretations suggest that superficial pyramidal cells (red triangles) compare the expectations (at each level) with top-down predictions from deep pyramidal cells (black triangles) of higher levels [22,23]. On the left: this schematic shows a simple cortical hierarchy with ascending prediction errors and descending predictions. This graphic includes neuromodulatory gating or gain control (blue) of superficial pyramidal cells that determines their relative influence on deep pyramidal cells encoding expectations through modulation of expected precision (see below and text for details). On the right: this provides a schematic example in the visual system. It shows the putative cells of origin of ascending or forward connections that convey prediction errors (red arrows) and descending or backward connections (black arrows) that construct predictions. The prediction errors are weighted by their expected precision that we have associated with the activity of neuromodulatory systems—here projections from ventral tegmental area (VTA) and substantia nigra (STN). In this example, the frontal eye fields send predictions to primary visual cortex, which it projects to the lateral geniculate body. However, the frontal eye fields also send proprioceptive predictions to pontine nuclei, which are passed to the oculomotor system to cause movement through classical reflexes. These descending predictions are also passed to the lateral geniculate body and constitute corollary discharge. Every top-down prediction is reciprocated with a bottom-up prediction error to ensure predictions are constrained by sensory information. The resolution of proprioceptive prediction error is particularly important because this enables descending predictions—about the state of the body—to cause movement by dynamically resetting the equilibrium or set-point of classical reflexes. Resolving sensory prediction errors through action is known as active inference (see the text). Adapted from Friston [26]. (Online version in colour.)

Figure 3.

(a) Giuseppe Arcimboldo, The Vegetable Gardener (ca 1590). Oil on panel. Our percepts are constrained by what we expect to see and the hypotheses that can be called upon to explain sensory input [27]. Arcimboldo, ‘a 16th century Milanese artist who was a favourite of the Viennese, illustrates this dramatically by using fruits and vegetables to create faces in his paintings. When viewed right side up, the paintings are readily recognisable faces’ [, p. 204]. Adapted from Friston [26]. (b) Faces are probably one of the most important (hidden) causes of our sensations. While in Arcimboldo's image, viewing right side up is needed for the configuration of features to appear as a face, when images are already recognizable faces, viewing right side up (by rotating the page) reveals that these faces might in fact be more different than they appear (this is the so-called ‘Thatcher illusion’). These examples illustrate the complex interplay between prior expectations and stimulus features that shape perceptual content (adapted from Little et al. [29]). (Online version in colour.)

Figure 4.

A (simplified) neural architecture underlying the predictive coding of visual, somatosensory and interoceptive signals. The anatomical designations, although plausible, are used to simply illustrate how predictive coding can be mapped onto neuronal systems. As in figure 2, red triangles correspond to neuronal populations (superficial pyramidal cells) encoding prediction error, while blue triangles represent populations (deep pyramidal cells) encoding expectations. These provide descending predictions to prediction error populations in lower hierarchical levels (blue connections). The prediction error populations then reciprocate ascending prediction errors to adjust the expectations (red connections). Arrows denote excitatory connections, while circles denote inhibitory effects (mediated by inhibitory interneurons). In this example, recurrent connections mediate innate (epigenetically specified) reflexes—such as the suckling reflex—that elicit autonomic (e.g. vasovagal) reflexes in response to appropriate somatosensory input. These reflexes depend upon high-level representations predicting both the somatosensory input and interoceptive consequences. The representations are activated by somatosensory prediction errors and send interoceptive predictions to the hypothalamic area—to elicit interoceptive prediction errors that are resolved in the periphery by autonomic reflexes. Oxytocin (in green) is shown to project to the hypothalamic area, to modulate the gain or precision of interoceptive prediction error units. One hypothesis for autism rests on a failure to attenuate the precision of autonomic prediction errors, thereby precluding expectations about visual and somatosensory information (e.g. a mother's face or affiliative touch) that is not accompanied by autonomic input (see the text). FFA, fusiform face area; AIC, anterior insular cortex; ACC, anterior cingulate cortex; OFC, orbitofrontal cortex; PAG, periaqueductal grey; PBN, parabrachial nucleus. (Online version in colour.)