The everyday sight of large language models conversing fluently and handling complex reasoning tasks leads many users to intuitively wonder whether something like subjective experience exists on the other side of the screen. But no matter how intelligent a machine's behavior appears on the surface, whether it is accompanied by genuine inner experience is a separate question. Among scientists and philosophers, a long-running dispute continues over whether machines can possess subjective experience—what is known as phenomenal consciousness.
Within this debate, one of the most influential arguments against the possibility of machine consciousness has been "biological naturalism." This position holds that consciousness is a phenomenon unique to the physical properties of carbon-based living tissue, and that no matter how precisely silicon circuits simulate computation, subjective experience cannot arise from them.
On August 11, 2026, Cell Press's peer-reviewed open-access commentary journal Trends Open published a theoretical paper dissecting the logical structure of this biological naturalism (DOI: 10.1016/j.treopn.2026.07.012; preprint posted to arXiv on June 1, 2026, DOI: 10.48550/arxiv.2606.02121). The authors are Dr. Ulysse Klatzmann, a postdoctoral researcher at the University of Montreal and CHU Sainte-Justine, and Professor Adrien Doerig of Freie Universität Berlin. The paper is titled "What biology can, and cannot, tell us about conscious AI."
This is not an empirical study presenting new laboratory measurements. Rather, in a journal centered on commentaries and opinion pieces, it is a conceptual analysis that clarifies the logical foundation of what biology "can and cannot tell us" about the possibility of artificial consciousness. The authors do not make a blunt determination about whether current AI models are conscious; instead, they present scientific criteria for testing whether machines could acquire subjective experience.
Two Kinds of Biological Naturalism, and the Boundary of Scientific Testability
Biological naturalism, proposed by philosopher John Searle, has long been treated as a formidable counterpoint to computational functionalism. Where computational functionalism holds that "consciousness arises wherever the right information processing is carried out, regardless of the underlying physical substrate," the biological position grants a special role to "the biological substrate of the brain itself."
Klatzmann and Doerig divide this biological naturalism into two categories: "Type-A biological naturalism" and "Type-B biological naturalism." The authors argue that any claim of biological naturalism necessarily falls into one of these two types.
Type-A holds that living tissue is "intrinsically special" for consciousness. It does not claim that biological tissue provides unique computational capacities or information-processing properties—rather, it holds that consciousness arises simply because the tissue is biological. Under this view, even a silicon device that performed information processing identical to a biological brain and produced identical behavioral outputs would not possess subjective experience.
The authors point out that this Type-A claim is, in principle, unfalsifiable and therefore cannot stand as a scientific hypothesis. Because it severs consciousness entirely from any observable behavior or information processing, there is no path to verifying it through external measurement or experiment. This logically echoes the "unfolding argument" that Doerig and colleagues raised in 2019 in the journal Consciousness and Cognition.
Type-B, by contrast, claims that living tissue gives rise to consciousness because biological systems possess "unique computational or information-processing properties not reproduced in current artificial systems." For example, it holds that properties such as complex dendritic integration in biological neurons, continuous temporal dynamics, and self-organization are involved in generating consciousness.
Because Type-B ties consciousness to observable information-processing capacities, it is empirically testable. Under the Type-B framework, biological naturalism does not necessarily conflict with computational functionalism. If consciousness depends on specific complex information processing that is difficult to implement in current silicon architectures and can currently only be realized through biological tissue, then biological naturalism and functionalism converge on the same conclusion.
| Criterion | Type-A Biological Naturalism | Type-B Biological Naturalism |
|---|---|---|
| Role of biological tissue | Intrinsically essential (unrelated to processing capacity) | Essential as the bearer of unique processing capacities |
| Consciousness in a machine with equivalent processing | Denied (because it is not biological) | Room for affirmation (if the processing is reproduced) |
| Empirical, scientific testability | Untestable (behavior and consciousness diverge) | Testable (specific processing can be identified experimentally) |
| Underlying premise | Causal power unique to biological tissue | Information-processing dynamics unique to living systems |
The Brain-Chip Thought Experiment and the Dilemma of Untestability
To explain why the Type-A claim cannot withstand scientific scrutiny, the authors invoke a thought experiment in which part of the brain is replaced with silicon components.
Suppose part of a subject's visual cortex is replaced with a microchip that performs input-output relationships and information processing completely identical to the original neural circuitry. The subject continues to recognize objects just as before replacement, verbally reports how things appear, and performs all visually based behaviors normally. Yet under the Type-A position, it would still be claimed that subjective visual experience (qualia) vanished the moment biological tissue was lost—even though information processing was fully preserved.
Since all externally observable behavior, self-reports, and neural signal transmission remain unchanged, there is no way for a third party to objectively confirm whether subjective experience has actually been lost. If science is an enterprise grounded in objective measurement and falsifiability, then such a hypothesis cannot be placed on the table for empirical testing.
This thought experiment is not pure fantasy. The paper references an actual empirical study in which an electrode array was implanted in the occipital cortex of a fully blind patient, and electrical stimulation induced simple perceptions of light (phosphenes). The fact that the subject reported subjective perception when artificial electrical signals were input into biological tissue demonstrates that intervening in information processing can directly affect subjective experience.
The authors define the methodological requirements for biological naturalism to qualify as a scientific hypothesis. Researchers taking the Type-B position must present concrete hypotheses answering the following three questions:
- Which biological property is essential for consciousness?
- How does that property alter the mode of information processing?
- How can the claim be tested experimentally?
Just as Global Workspace Theory (GWT) advances the testable prediction that "conscious perception arises only when large-scale neural ignition is present, and does not arise when it is absent," Type-B must likewise experimentally demonstrate a causal relationship in which "consciousness arises when biological property X is present, and disappears when it is absent."
Candidate biological properties include continuous neural dynamics, biological self-regulation, nonlinear signal integration at dendrites, and multi-level dynamic connectivity across the brain. However, the authors caution that, at present, no example exists proving that any of these properties is indispensable for generating consciousness.
Even against the claim that continuous, chaotic dynamics are unique to biological systems, the authors note that weather systems are continuous and chaotic yet are simulated with high precision on computers—demonstrating that this alone does not imply impossibility of reproduction on silicon.
An Open Question: What Kind of Information Processing Supports Consciousness
Klatzmann, one of the paper's authors, has described his own position as agnostic. In an official statement released by the University of Montreal, he said, "I would be surprised if current AI models developed consciousness. But we cannot completely rule out that possibility," cautioning against hasty conclusions.
This paper does not settle the question of whether machines can be conscious in black-and-white terms. Its significance lies in untangling what has resembled a theological dispute and organizing the two major scenarios that future research should pursue.
The first scenario is one in which biological properties turn out to be absolutely necessary for generating consciousness. In this case, software running on conventional von Neumann-style silicon hardware would never possess consciousness. However, a path would remain open for humanity to eventually build artificial systems using organic devices that mimic biological tissue or biological properties, potentially realizing artificial consciousness equipped with biological characteristics.
The second scenario is one in which consciousness depends on a specific functional structure of information processing. In this case, the focus of research shifts to identifying which information-processing properties carried out by the brain support consciousness, and then pursuing the engineering and computational question of whether artificial machines can reproduce them.
The authors further leave open a third possibility: that the properties relevant to consciousness may not be reducible to pure abstract computation, nor necessarily confined to biological tissue alone. Theoretically, there remains room for the premises of both positions to fail to hold.
What the authors emphasize is that biological findings should be used not to declare machine consciousness impossible, but to identify the computational principles underlying consciousness. Within the Type-B framework, the biological approach and computational functionalism can coexist—but exactly how the properties related to consciousness are positioned between computation and biology remains, itself, an unresolved scientific question.
Where Conceptual Analysis Stands in AI Consciousness Research Today
This study is a theoretical and philosophical analysis published in a peer-reviewed commentary journal; it does not provide new laboratory measurements or brain activity data. It neither confirms nor refutes any specific hypothesis through experiment—it is an attempt to reconstruct the logical framework of the debate.
In recent years, debate over AI consciousness has rapidly intensified. In a June 2026 commentary contributed to the science media outlet The Transmitter, Hadid, Jerbi, and Krakauer cited neuroscientific findings on patients with blindsight—who can avoid obstacles without any accompanying visual awareness—warning that sophisticated, intelligent behavior alone is not proof of consciousness. Criticism of AI consciousness from enactivism, championed by cognitive scientist Anthony Chemero and others—which locates consciousness in an embodied agent's interaction with its environment—also remains influential.
In the theoretical literature, a preprint published on arXiv by Simon, Campero, Shiller, and Aru attempts to systematically classify objections to digital consciousness using computational neuroscientist David Marr's "three levels of computational analysis" (computational theory, representation and algorithm, and hardware implementation). The work by Klatzmann and Doerig echoes this classificatory approach while organizing the debate specifically around the question of testability.
As the authors themselves acknowledge, at present there is little empirical evidence supporting any particular hypothesis of biological naturalism. Whether consciousness can reside in silicon, or whether it requires the wet, carbon-based tissue of living organisms—arriving at an answer will require not further philosophical proclamations, but the patient work of rigorously translating the neural correlates of consciousness (NCC) into the language of information processing and building testable experimental systems one by one.