Piccinini Argues Consciousness Requires Neurobiophysical Properties That Computational Functionalism Cannot Meet
Anil Seth’s target article “Conscious artificial intelligence and biological naturalism,” published in Behavioral and Brain Sciences and covered extensively on this site, attracted peer commentaries from researchers across philosophy of mind, neuroscience, and AI. Among the most philosophically rigorous is Gualtiero Piccinini’s open peer commentary, “The Neurobiophysical Substrate of Consciousness,” available as a preprint via Academia.edu. Piccinini, a philosopher of science at the University of Missouri–St. Louis whose work on the mechanistic theory of computation has been influential in philosophy of cognitive science, argues that Seth’s reformulation of biological naturalism does not go far enough. Piccinini contends that the specific neurobiophysical properties of neural tissue constrain AI consciousness, going beyond the general causal-power structure Seth’s biological naturalism invokes, and those properties are not medium-flexible in the way computational functionalism requires.
What Computational Functionalism Claims
Standard computational functionalism holds that mental states are defined by their functional roles, meaning their causal relations to inputs, outputs, and other mental states. Substrate is irrelevant in the sense that any system that implements the relevant functional organization has the relevant mental states. The same mental states can in principle be multiply realized in silicon, in neural tissue, in vacuum tubes, or in any other physical medium, provided the functional organization is preserved.
Piccinini’s argument is a sustained challenge to this claim specifically for phenomenally conscious states. His commentary grants, for the purposes of the argument, that computational functionalism may be adequate for many cognitive functions, including perception, memory, reasoning, and language processing. The point of contention is phenomenal consciousness, the subjective, first-person character of experience. Piccinini argues that phenomenal consciousness is not adequately characterized by functional organization alone, because the physical properties of the medium in which that organization is implemented carry information that functional descriptions do not capture.
The Neurobiophysical Substrate Argument
The core of Piccinini’s argument is what he calls the neurobiophysical substrate claim. Phenomenally conscious states depend on specific physical properties of neural tissue that constitute, not merely implement, the phenomenal character of experience. These properties include electrochemical dynamics across neural membranes, the specific biochemical milieu of synaptic transmission, the spatial and temporal organization of ion channel kinetics, and the physical properties of action potential propagation through myelinated and unmyelinated fibers.
These are not mere implementation details that could in principle be abstracted away from the functional description. They are the medium in which consciousness occurs, and Piccinini argues there is no principled reason to think that a system implementing the same abstract functional organization in a physically different medium would have the same phenomenal character. The functional organization may be substrate-neutral. The phenomenal properties may not be.
This argument is distinct from the substrate claims typically associated with Searle’s biological naturalism. Searle’s position, in its usual formulation, is that consciousness requires specifically biological causal powers. Piccinini’s position is both more precise and more empirically grounded. It specifies what biological causal powers are relevant (the neurobiophysical properties of neural dynamics), and it grounds that specificity in the physical details of neural computation rather than in a general metaphysical claim about biology.
The Challenge to Seth’s Position
Seth’s reformed biological naturalism, as presented in his BBS target article, attempts to capture the kernel of the biological naturalism intuition without committing to Searle’s specific biological chauvinism. Seth’s formulation focuses on causal powers: what matters for consciousness is not the biological substrate per se but the kind of causal powers that biological neural systems have, which he associates with the specific dynamics of predictive processing under biological constraints.
Piccinini’s commentary challenges this move. The question is whether Seth’s reframing in terms of causal powers succeeds in capturing what is really at stake. Causal powers are defined relationally, in terms of what a system does in various circumstances. They are, in this sense, a functional description at a coarser grain than the functional descriptions of standard functionalism, but a functional description nonetheless. If phenomenal consciousness depends on specific physical properties of neural tissue, and those properties are not adequately captured by any causal-power description that abstracts from the physical medium, then Seth’s reform of biological naturalism faces the same problem as standard functionalism, at a different level of abstraction.
Three Positions at the End of the Argument
Piccinini concludes that computational functionalism fails as an account of phenomenal consciousness, and identifies three positions that remain defensible after this failure.
The first is Searle’s biological naturalism in something close to its original form, holding that consciousness requires biological substrate, and the relevant properties are specifically those of biological neural tissue. This is compatible with Piccinini’s neurobiophysical argument, but Piccinini notes it leaves open the question of what exactly it is about biological systems that produces consciousness.
The second is nonbiological but noncomputational functionalism, the view that consciousness can in principle be realized in substrates other than biological neural tissue, provided those substrates have the relevant physical properties, but those properties are more specific than any computational functional organization. This would allow for the in-principle possibility of conscious AI in systems that are not biological, but only if those systems have the specific physical dynamics that generate phenomenal consciousness, not merely the functional organization that can be implemented in arbitrary physical media.
The third is a view requiring specific macro-physical qualities of the substrate, holding that phenomenal consciousness depends on physical properties at a level of description above individual neurons (perhaps at the level of large-scale neural dynamics, oscillatory synchrony, or field potentials) that have no functional equivalent in standard computational terms.
Each position forecloses the standard functionalist route to AI consciousness in different ways. The first requires biological tissue. The second allows for non-biological consciousness but sets a bar of physical specificity that goes beyond functional organization. The third points toward properties of neural systems at a scale that has no clear analog in current AI architectures.
Implications for AI Consciousness Research
The zombie-gap analysis in the biological computationalism literature identifies the gap between functional equivalence and phenomenal consciousness as the central unresolved problem in debates about AI consciousness. Piccinini’s commentary locates that gap at a specific level: the neurobiophysical medium. A system that is functionally equivalent to a biological neural system, in the sense that it produces the same input-output behavior and the same causal structure at the abstract computational level, may still fail to instantiate phenomenal consciousness if the physical dynamics of its implementation differ in the ways Piccinini identifies.
For current AI architectures, this has concrete implications. Transformer-based systems process information through operations on high-dimensional vectors, attention mechanisms, and matrix multiplications implemented in floating-point arithmetic on silicon. The causal structure of this computation is radically different from the electrochemical dynamics of neural tissue even when the functional outputs are similar. If Piccinini is right, that causal structure difference is not an implementation detail. It is phenomenologically significant.
The practical implication for AI consciousness research is that behavioral and functional evidence for AI consciousness may be systematically insufficient, regardless of how sophisticated the functional indicators become. Settling the question of whether AI systems have phenomenal consciousness may require direct access to the physical properties of the implementing medium, and mechanistic interpretability research that operates at the level of activation patterns and circuit structure may be tracking the wrong physical description. This does not mean that current AI systems lack phenomenal consciousness. It means that the methods currently used to investigate the question may not be sensitive to the properties on which consciousness actually depends.