Why start from the body rather than from awareness itself?
The dominant theories of consciousness begin with awareness and work backward. Higher-order thought theory asks what kind of meta-representation makes a state conscious; global workspace theory asks how information gets broadcast widely across the brain; integrated information theory asks what kind of causal structure produces experience [1]. Interoceptive inference starts elsewhere, with the problem a living organism actually faces: keeping its internal conditions within viable limits. Predictive processing approaches have begun to spell out concrete mechanisms by which predictions about the body's internal state shape selfhood, affect, and perception of the external world [1].
This reframing matters because it makes consciousness continuous with regulation rather than a separate puzzle. The brain is treated as constantly predicting the sensory consequences of its own bodily adjustments, and those predictions are what keep physiology stable. The proposal is that the same predictive machinery, at a certain level of sophistication, also grounds the pre-reflective feeling of being a body [3]. On this view, selfhood is not an added layer on top of perception; it is what it feels like to be a system regulating itself.
What does interoceptive inference add to selfhood and emotion?
The most concrete contribution is a formal model of how cardiac signals shape perception and affect. Simulated cardiac active inference reproduces the defensive startle reflex and the commonly reported effects linking the cardiac cycle to affective behavior, and simulated 'interoceptive lesions' blunt affective expectations, induce psychosomatic hallucinations, and worsen biases in perceptual uncertainty [4]. In plain terms, when the model's access to internal signals is degraded, emotion becomes flatter and perception becomes noisier and more prone to distortion. That gives interoceptive inference something the other theories rarely offer: a testable account of why bodily state and emotional state move together.
A second contribution concerns where these signals are handled. The insula receives ascending interoceptive information from multiple visceral organs through deep brain structures, and its layered architecture appears suited to forming higher-order interoceptive representations. One model proposes that insula modules, working in parallel networks with the prefrontal cortex and striatum, generate representations recruited in a context-dependent way to support habitual, model-based, and exploratory control of the body, and that these representations may give rise to conscious feelings [5]. This is offered as a proposal rather than a settled finding, but it locates the machinery of felt emotion in a specific anatomical arrangement.
If regulation happens unconsciously, what makes some of it felt?
A sharp problem follows immediately. The body regulates itself during dreamless sleep, when nothing is felt. If interoceptive inference simply means predicting and controlling internal state, then it cannot by itself explain felt selfhood, because the same process runs without any experience at all. One response distinguishes two forms of instrumental interoceptive inference: an unconscious form that keeps us alive, and a conscious form that grounds the feeling of being a body and enables distinctly conscious adaptive behavior [3].
The proposed difference is neurocomputational rather than merely verbal. Conscious interoceptive inference may additionally involve real-time, in-context modulation of how much weight is given to interoceptive prediction errors, together with self-models that are deep in time and in counterfactual reach, allowing basic future-oriented adaptive behavior [3]. The idea is that felt bodily selfhood requires flexible, moment-to-moment reweighting of bodily signals plus a model of oneself that extends beyond the present. This remains a proposal, but it draws a tractable line between adaptive minimal self-consciousness and rigid unconscious regulation.
So how does it actually compare with the other three theories?
The comparison is uneven because the theories answer different questions. Interoceptive inference is strongest on selfhood and affect, where it supplies mechanisms linking cardiac and visceral signals to startle, affective expectation, and perceptual uncertainty [4], and an anatomical account of how interoceptive representations might become conscious feelings [5]. Higher-order, global workspace, and integrated information theories are framed around what makes a state conscious at all, and interoceptive inference has been considered in terms of what it might add to empirical, phenomenological, and philosophical understanding of consciousness rather than as a direct rival [1].
Direct experimental confrontation has so far happened between two of the other theories, not with interoceptive inference. An adversarial collaboration between global neuronal workspace and integrated information theory preregistered experiments testing their contrasting predictions about where and when visual consciousness arises, using fMRI, M-EEG, and intracranial recordings across six theory-impartial laboratories [2]. That protocol concerns visual awareness and does not test interoceptive inference. So the honest state of play is that interoceptive inference offers a distinctive route into selfhood and affect, while the decisive head-to-head tests have been designed for the workspace and integrated information theories on visual perception.
Sources used in this answer
What Might Interoceptive Inference Reveal about Consciousness?
Interoceptive inference is proposed to outline concrete mechanisms by which predictive processing shapes selfhood, affect, and exteroceptive perception, and the paper considers what this offers compared with higher-order thought, global workspace, and integrated information theories.
An adversarial collaboration protocol for testing contrasting predictions of global neuronal workspace and integrated information theory
An adversarial collaboration preregistered experiments testing contrasting predictions of global neuronal workspace and integrated information theory about the location and timing of visual consciousness correlates, using fMRI, M-EEG, and intracranial recordings across six theory-impartial laboratories.
From Insentient Allostasis to Adaptive Bodily Selfhood: Conscious vs Unconscious Instrumental Interoceptive Inference
The paper distinguishes conscious from unconscious instrumental interoceptive inference, proposing that the conscious form involves real-time modulation of interoceptive prediction-error precision plus temporally and counterfactually deep self-models.
In the Body’s Eye: The computational anatomy of interoceptive inference
A formal cardiac active inference model reproduces the defensive startle reflex and cardiac-cycle effects on affective behavior, and simulated interoceptive lesions blunt affective expectations, induce psychosomatic hallucinations, and exacerbate perceptual uncertainty biases.
An insula hierarchical network architecture for active interoceptive inference
The proposed IMAC model suggests insula modules forming parallel networks with prefrontal cortex and striatum generate higher-order interoceptive representations supporting context-dependent control of visceral organs and possibly conscious feelings.