Can consciousness be fully understood in computational terms, or must an adequate theory make reference to fine-grained, substrate-level features? This is a central question for contemporary cognitive science, one whose resolution has been made more urgent by the arrival of sophisticated anthropomimetic systems (Shevlin, Reference Shevlin, Rangeley and Fairfax2025), able to emulate the coarse-grained behavioural capacities of human intelligence. In this landmark article, Seth appropriately frames the debate around the duelling approaches of biological naturalism and computational functionalism. However, as I will argue, there are nuances in how we apply these theories, and it is not necessarily the case that all forms of consciousness must be analysed using the same scientific-explanatory lens.

As an initial analogy, consider the phenomenon of flight. Over the course of evolution, flight has emerged independently at least four times, in insects, pterosaurs, birds, and bats. A proper understanding of flight in each case will make essential reference to fine-grained biological characteristics, from the demands it makes on metabolism to its associated and varied changes in muscles, tendons, and skeletal structure. However, we also know that flight is not restricted to biological organisms: human artefacts such as planes, helicopters, and rockets demonstrate that flight can be achieved in quite different forms.

The point of this analogy is hopefully clear: even if a proper account of biological consciousness cannot be given in purely computational terms, and must make reference to fine-grained, substrate-dependent features, this does not foreclose the possibility of different kinds of consciousness that lack these features. Even though AI systems may never instantiate the same kinds of conscious states as humans and animals, it is premature to say that they will never instantiate consciousness in any form.

The astute reader may recall the famous quip of Edsger Dijkstra that the “question of whether a computer can think is no more interesting than the question of whether a submarine can swim” (Dijkstra, Reference Dijkstra1986), and point out that the multiple realisability of flight is primarily a semantic matter, thus disanalogous to the scientific concept of consciousness. While I agree that there is more semantic flexibility in how we use the term “flight” than the term “consciousness,” the analogy illuminates the possibility of a coarse-grained, functional account of consciousness encompassing quite different realisations, some biologically grounded, and others potentially realised computationally.

One concern with this pluralistic approach is that it seems theoretically disunified. If the category of flight has “powered motion through the air” at its core, what might provide a similar unifying function for capturing seemingly disparate phenomena like autopoietic biological consciousness and putative computational machine consciousness?

One answer is provided by the historically significant and recently resurgent family of views about consciousness known as analytic functionalism (Lewis, Reference Lewis and Cheng1975). Unlike computational functionalism, analytic functionalism does not aim to understand the mind exclusively within the austere framework of Turing computation, but rather in the wider vocabulary of folk psychology. For the analytic functionalist, we should understand our lexicon of mental-state terms as folk-theoretical posits to help us better understand, predict, and explain the behaviour of a special class of physical system, namely intelligent goal-directed agents (Dennett, Reference Dennett1987). (As Seth rightly notes, of course, the intentional stance can go awry, in leading us to make naïve ascriptions of humanlike mentality to AI systems. However, this arguably reflects current quirks of Large Language Models (LLMs) such as hallucination, rather than any fundamental inapplicability of the stance to artificial systems.)

Analytic functionalism is appealing for a wide range of mental-state concepts. To give a toy example, what it is for a system to have preferences might be the tendency to seek out goals in a manner analysable through the lens of means-end folk psychology. Seth himself seems sympathetic to such a high-level account of intelligence.

But even if some mental states can be analysed through this lens, can we really approach consciousness in a similarly interpretative manner? Many will balk at this suggestion, taking it as given that doing justice to consciousness requires us to identify a singular underpinning biological or computational kind.

However, analytic functionalism need not commit us to eliminativism: consciousness may be a “real pattern” with distinctive explanatory power, even while lacking a singular biological or computational basis (Dennett, Reference Dennett1991). Moreover, “deep realism” about consciousness (Shevlin, Reference Shevlin and Strasser2024) should itself not go unquestioned, not least because of the lurking hints of dualism it encourages. As Shanahan (Reference Shanahan2024) puts it, “We must resist the temptation to ask whether [a system is] conscious as if consciousness were something whose essence is out there to be uncovered by philosophy (or neuroscience) while simultaneously having an irreducibly private, hidden aspect.” What if the best way to analyse our concept of consciousness (and the ethical implications that go with it) is in broader ecological terms, as a way of understanding how intelligent and self-aware systems move through the world, interact with others, and conceptualise their own status within it?

The science of consciousness remains in its infancy, and Seth’s article provides a powerful reminder of our limited but growing understanding, even of biological consciousness. However, even as we refine our theories of human and animal consciousness, we should be cautious not to extrapolate too quickly from contingent features of its realisation in beings like us to the wider and more inscrutable space of possible minds.