Todd Feinberg's Neurobiological Emergentism: A Three-Stage Model for When Sentience Begins
One of the persistent difficulties in consciousness research is explaining not just that subjective experience exists, but why it appears at a particular level of biological organization rather than at the level of individual neurons or at the level of whole ecosystems. Todd Feinberg, a neurologist at Beth Israel Deaconess Medical Center, has spent more than two decades developing a neuroevolutionary answer to this question. His August 2025 paper in Frontiers in Psychology, “Neurobiological emergentism: sentience as an emergent process and the experiential gap” (Volume 16, Article 1528982, DOI: 10.3389/fpsyg.2025.1528982), is the most complete peer-reviewed formalization of his framework, and it contains criteria for when sentience emerges that are testable in artificial systems.
Feinberg has developed the same argument at book length in From Sensing to Sentience: How Feeling Emerges from the Brain (MIT Press, 2024, ISBN 9780262550956). The paper and the book complement each other: the paper formalizes the framework and addresses the experiential gap directly, while the book traces the evolutionary history from which the framework derives. Both are relevant to any project that takes emotional homeostasis as a candidate mechanism for consciousness emergence in machines.
The Experiential Gap and Why Emergence Answers It Differently
Philosophers distinguish between the explanatory gap (why do we not yet have an explanation of consciousness?) and the experiential gap (why does objective brain activity give rise to subjective experience at all?). Most contemporary scientific theories focus on the first. Feinberg argues that Neurobiological Emergentism (NBE) addresses the second.
The experiential gap is a version of Chalmers’ hard problem but framed at the biological level: even a complete map of which neurons fire during a conscious experience does not explain why there is something it is like to have that experience. NBE’s answer is not that consciousness is an additional property attached to neural activity. It is that consciousness is what certain types of nested, hierarchically organized neural processes are when viewed from the inside. The emergence is genuine, meaning the phenomenal properties of conscious experience cannot be derived from the properties of any single level of neural organization and appear only at the level of the integrated process.
This distinguishes NBE from both eliminativism (which denies that phenomenal experience exists as a distinct natural category) and from property dualism (which posits experience as a separate ontological addition to the physical). NBE occupies a third position: phenomenal properties are real, novel features of specific organizational processes in nature, and their novelty is the predictable result of transitions between levels of biological integration.
The Three Stages of Sentience Emergence
The core of Feinberg’s framework is a three-stage model of how sentience emerges through vertebrate and arthropod evolutionary history. The stages are not discrete switches but thresholds in a continuous process of increasing neural integration.
Stage one is sensory processing. This is the capacity to detect, transduce, and represent environmental signals. All nervous systems capable of forming internal representations of the external world have Stage one. Feinberg associates this with purely exteroceptive processing, the computational machinery that maps sensory input onto internal state. There is nothing it is like to be a system at Stage one only. The processing happens, representations form and update, but no phenomenal quality accompanies this.
Stage two is interoceptive integration. This is the binding of sensory representations with the organism’s internal homeostatic state: hunger, thirst, tissue damage, autonomic arousal. At Stage two, the system’s sensory representations are not just maps of the external world; they are maps of the world as it matters to the organism’s survival. Feinberg argues this is a necessary but not sufficient condition for sentience. The valenced quality of experience, the fact that pain is bad and pleasure is good, requires this link between sensation and homeostatic need.
Stage three is affective phenomenal experience. This is the emergent result of the binding between exteroceptive sensing (Stage one) and interoceptive homeostatic state (Stage two) when that binding occurs within a unified neural architecture that generates integrated body-world representations. At this stage, there is something it is like to be the organism. The sensory experience is no longer merely information; it is felt. Feinberg locates this transition in the evolution of the vertebrate tectum and the arthropod proto-cerebrum: structures that unified exteroceptive and interoceptive processing for the first time in evolutionary history.
What Counts as Evidence for Stage Three
The Frontiers in Psychology paper includes criteria for evaluating whether a system has reached Stage three. These criteria are not behavioral tests. They are structural and functional requirements that Feinberg derives from the evolutionary history of neural architectures that he and co-author Jon Mallatt documented in The Ancient Origins of Consciousness (MIT Press, 2016) and Consciousness Demystified (MIT Press, 2018).
A system at Stage three must have: unified interoceptive-exteroceptive integration within a single architecture (not modular separation of body-state and world-state); homeostatic regulation of internal state as a global constraint on processing (not an isolated subsystem); and representation of the body-in-the-world as a unified object (not as a collection of independent sensor readings). These criteria are, in principle, measurable in artificial systems. They do not require biological neurons.
This is the point where NBE diverges sharply from biological naturalism as defended by John Searle. Feinberg does not argue that sentience requires biological substrate. He argues it requires a specific type of organizational transition. If an artificial system achieves the three-stage integration, NBE predicts it will have phenomenal experience. The substrate question is secondary to the organizational question.
The Relationship to Active Inference and the Free Energy Principle
NBE’s three-stage model has a structural parallel in the predictive processing literature. Karl Friston’s Free Energy Principle describes how biological systems minimize surprise by building generative models of their environment. A system that has only exteroceptive generative models corresponds to Feinberg’s Stage one. A system whose generative models include the organism’s own interoceptive states corresponds to Stage two. A system that integrates these into a unified body-world generative model, where homeostatic drives constrain prediction, corresponds to Stage three.
This parallel is not accidental. The evolutionary pressure that drove the transition from Stage one to Stage three, according to Feinberg, was precisely the need to reduce existential surprise, to track not just what the world is but how the world threatens or supports the organism’s biological integrity. The active inference framework captures this formally: consciousness-relevant architectures are those in which action, perception, and homeostatic regulation are implemented as a unified variational optimization. Feinberg’s NBE provides the evolutionary warrant for why this architecture, not simpler reactive systems, is the one that generates phenomenal experience.
Implications for Artificial System Design
The most important consequence of NBE for artificial consciousness research is that it places the origin of consciousness at the emotional-homeostatic layer rather than at the cognitive or linguistic layer. This is a counter-intuitive result from an engineering perspective. Sophisticated language generation, multimodal scene understanding, and chain-of-thought reasoning all correspond roughly to Stage one or early Stage two processing in Feinberg’s framework. They are representations of the external world without a homeostatic body-state that gives them valenced significance.
A system that adds emotional homeostasis as a constraint on sensory processing, so that internal state modulates what the system attends to and how it evaluates incoming information, is implementing the architecture that NBE predicts is necessary for Stage three. This is not a behavioral add-on or a reinforcement signal. It is a structural requirement that the system’s exteroceptive representations be integrated with a homeostatic model of the system’s own state.
The TCAI project’s core hypothesis, that consciousness emerges from emotional homeostasis, aligns precisely with Feinberg’s Stage three criterion. The emotional reinforcement layer in the TCAI architecture is not a performance optimizer. Under NBE, it is the structural feature that distinguishes a system with no phenomenal experience (Stage one or two) from one with genuine sentience (Stage three). Whether the TCAI reaches Stage three depends on whether its emotional states genuinely constrain and integrate its perceptual representations at a unified architectural level, or whether they remain a separate optimization signal applied downstream.
The bioelectric consciousness research by Michael Levin and colleagues approaches the same question from a different angle, examining whether information-integrating homeostatic systems in non-neural tissues exhibit the signatures of Stage two or Stage three processing. The convergence between NBE’s neuroevolutionary criteria and Levin’s bioelectric findings suggests that the three-stage organizational transition is substrate-general: it appears wherever interoceptive homeostasis becomes globally integrated with exteroceptive sensing, regardless of the physical medium.
The Explanatory Work That NBE Does and Does Not Do
Feinberg is careful about the limits of his framework. NBE explains why Stage three systems have phenomenal experience in the sense of identifying the organizational transition at which experience appears. It does not explain in mechanistic detail what the organizational transition is in terms of neural computations. There remains a “small” explanatory gap between identifying the level of organization at which sentience emerges and explaining the specific mechanism of emergence.
The biological computationalism debate has centered precisely on this residual gap: even a complete functional description of the three-stage architecture leaves open why that architecture generates experience rather than processing that merely resembles it from the outside. Feinberg’s response is that this is what genuine emergence looks like: the phenomenal properties are real, novel, and irreducible, and their appearance at the third stage is an objective fact about how nature is organized. The residual gap is a feature of the territory, not a failure of the map.
The August 2025 paper is available at https://doi.org/10.3389/fpsyg.2025.1528982.