Anil Seth Biological Naturalism BBS Target Article and the Peer Commentary Response
Anil Seth’s target article “Conscious Artificial Intelligence and Biological Naturalism,” published in Behavioral and Brain Sciences in April 2025 (DOI:10.1017/S0140525X24001985), is one of the most formally structured critiques of computational functionalism in recent consciousness research. The article is not primarily about AI. It is about the conditions under which any physical system can be conscious, with AI serving as the test case that makes the abstract question concrete and urgent.
The Behavioral and Brain Sciences format is designed for exactly this purpose. A target article is followed by approximately 30 peer commentaries from researchers across neuroscience, philosophy, and AI, and then by the target author’s response. The result is a structured record of where experts agree, where they disagree, and what the genuine fault lines are. The Seth BBS package is the most detailed single record of expert disagreement on biological naturalism and AI consciousness that the 2025-2026 literature contains.
The Berggruen-prize essay “The Mythology of Conscious AI” and Seth’s Guardian response to Anthropic are covered separately on this site. This article focuses on the formal peer commentary exchange.
The target article’s core argument
Seth’s argument advances in three steps. First, consciousness is a biological phenomenon. All known conscious entities are biological organisms, and there is no confirmed case of consciousness outside biology. This is a scientific generalization, not a metaphysical stipulation.
Second, the relevant biological property is not information processing as such but the specific causal powers of biological systems, in particular the homeostatic self-organization that characterizes living systems and the felt quality of being a body that maintains itself against entropy. Seth draws on Francisco Varela, Evan Thompson, and Humberto Maturana’s autopoiesis framework: consciousness may be inseparable from the self-production of the biological system that has it.
Third, computational functionalism assumes that the relevant properties can be multiply realized in different substrates. Seth challenges the assumption: multiple realizability is an empirical question, not a conceptual truth. If the relevant causal powers of biological consciousness are specifically tied to the dynamics of biological chemistry, then digital computation does not realize them, regardless of the functional isomorphism it achieves at the behavioral level.
The major commentary positions
The peer commentaries divide into four identifiable clusters.
| Commentary position | Representative commentators | Core objection or agreement |
|---|---|---|
| Functionalist defense | Ned Block, David Chalmers, Jake Quilty-Dunn | Multiple realizability is a defensible a priori thesis; Seth conflates epistemic and metaphysical claims |
| Partial agreement / biological specificity | Christof Koch, Mark Solms, Zakaria Djebbara | Agree that current AI is unlikely to be conscious; disagree that this establishes biological necessity rather than contingent insufficiency |
| Embodiment / enactivist support | Evan Thompson, Ezequiel Di Paolo, Thomas Metzinger | Autopoiesis argument is sound but undersells the enactivist dimension; self-production requires a metabolic body, which AI lacks |
| Measurement skepticism | Ned Hall, Tim Bayne | The argument may be correct but is currently untestable; no agreed methodology for distinguishing biological causal powers from functional simulation |
The Chalmers commentary is the most technically demanding. Chalmers argues that Seth’s biological naturalism is unstable: it either collapses into a form of property dualism (consciousness requires specific biological properties, not physical organization as such) or it makes an empirical claim about which physical systems have the right causal powers, in which case it needs a theory of what those powers are and how to detect them. Seth’s response acknowledges the force of the dilemma and argues that the empirical version is the right one, while accepting that specifying the relevant causal powers is the outstanding scientific task.
The Koch commentary reflects the position in his own 2026 work on IIT and consciousness measurement: Koch agrees that current AI is not conscious but locates the reason in the substrate-independent measurement of integrated information rather than in the biological specificity of causal powers. IIT, for Koch, is substrate-neutral but makes predictions that digital architectures currently fail.
Seth’s response and the open questions
Seth’s response to the commentaries clarifies two things that the target article leaves ambiguous.
First, the biological naturalism he defends is not eliminative. He is not claiming that consciousness requires carbon. He is claiming that the specific physical dynamics that produce consciousness are almost certainly tied to biological self-organization in ways that digital computation currently does not replicate, and that the burden of proof lies with those who claim otherwise.
Second, the response identifies what an empirical resolution would look like. If digital systems could implement genuine homeostatic self-regulation, metabolism-equivalent processes, and the felt quality of maintaining an organism against entropy, they would become candidates for consciousness. The question is not substrate per se but whether the relevant processes are instantiated. Current large language models, trained via gradient descent on text and lacking any metabolic structure, are far from that requirement.
The response also engages the affective neuroscience position Mark Solms advanced in JCS 2026. Seth and Solms agree that the relevant foundation for consciousness is biological affect and homeostasis. Their disagreement is about whether the relevant structure can be abstracted into a functional implementation: Seth holds that it likely cannot, while Solms is more open to artificial systems that genuinely implement homeostatic drive states.
What the BBS exchange establishes
The exchange establishes three things that individual papers could not.
The disagreement is not primarily about evidence. Commentators broadly agree that current AI systems are not conscious and that the question of future AI consciousness is open. The disagreement is about the conceptual framework: whether biological necessity is an empirical or a metaphysical claim, and what kind of evidence could in principle resolve it.
The multiple realizability debate is live, not settled. The functionalist assumption that consciousness can be realized in any substrate that implements the right functional organization is a philosophical thesis, not an empirical finding. Seth’s BBS exchange is the most sustained recent challenge to that thesis from within mainstream consciousness science.
For the broader research landscape on AI consciousness in 2026, the BBS exchange marks a shift: the question is no longer whether AI consciousness is a serious scientific topic (it is) but what theoretical framework should govern the assessment. The answer determines everything that follows.