The year philosophy went back to asking what makes something a subject
Four university presses, one accidental theme. Across consciousness, communication, creativity and liability, the 2026 philosophy list keeps circling a question the discipline spent a generation dismantling — and it was engineers, not philosophers, who put it back into circulation.
Set the autumn catalogues of three university presses side by side and a pattern shows up that none of them planned. Cambridge has Eric Schwitzgebel on AI and consciousness, Geoff Keeling and Winnie Street on AI welfare, and, due on 29 October, Lindsay Brainard on whether a machine can be creative. Oxford has Herman Cappelen and Rachel Sterken asking whether talking to a chatbot counts as communication at all, Geoffrey Lee and Adam Pautz with four hundred pages on why consciousness matters, and Luke Roelofs arguing that people who do not care about other people have got something wrong about other minds. MIT has Nicholas Humphrey on why the world strikes us as beautiful. And Cambridge's 460-page handbook on the law, ethics and policy of artificial intelligence, edited by Nathalie A. Smuha of KU Leuven, appeared in February 2025 and can be downloaded for nothing.
Those books would be shelved in four different rooms. One asks whether there is anything it is like to be the system. Another asks whether it can mean what it says. A third turns on whether it made the image or merely assembled it, a fourth on who pays when it causes harm. Put like that, they are four disciplines speaking their local dialect of a single question, and the question is very old: what makes something a subject rather than an object — a bearer of experience, an author of its acts, a party to relations, someone rather than something.
The strange part is not that the question has come back. It is the manner of the return. No one rehabilitated it. The question came back because a handful of companies shipped a product that will not leave it alone, and philosophy has been left to work out how it feels about owing a research programme to a commercial cycle it does not control.
The subject was a word philosophy had agreed to stop using#
It is worth remembering how much work the word once did. In the first Critique Kant's subject is not a person at all. It is the condition under which a world can appear to anyone: the transcendental unity of apperception, the "I think" that must be able to accompany all my representations. Nothing biographical, nothing psychological — a formal requirement for experience to hang together.
The nineteenth century inflated it well past that. Fichte's I posits itself and then posits everything else; by the end of the century the subject has acquired a history, a class position, a nation and a set of drives, and is expected to explain most of them.
Then it was taken apart, twice, from opposite directions. Structuralism and what followed treated the subject as an effect of a system rather than its source. In an essay published in in 1970, Althusser described ideology interpellating individuals as subjects: a voice in the street calls out, you turn round, and in turning you become the thing that was addressed. Foucault worked the other end of the same problem, describing subjectivation as a set of practices that produce a subject in the very movement by which they subject it. Barthes had already announced the death of the author in 1967; Foucault asked the Société française de philosophie in February 1969 what the author function was for. By the time Fredric Jameson wrote on postmodernism in 1984, "the death of the subject" was a phrase he could use without argument. In 1991 Routledge published , a collection built out of a question Jean-Luc Nancy had put to a couple of dozen philosophers. The title took the funeral as read.
A popular thesis holds that Cold War repression drove American philosophers into the political safety of logic and language. Sander Verhaegh's Logical Positivism: An American History, from Oxford University Press, spent eight years in departmental archives and finds something less dramatic: a generation gap, a hiring freeze, a postwar boom and a culture that worshipped science.
Ludwig Wittgenstein, Rudolf Carnap, John Dewey, Nelson Goodman
Analytic philosophy got to a similar place by a quieter road. Functionalism turned mental states into causal roles, which in principle anything can occupy, and the interesting work moved to the roles rather than the occupant. Dennett's intentional stance made belief-attribution a predictive convenience, and heterophenomenology made the first person into data to be interpreted from outside. Nobody declared anything dead. The subject simply stopped being a thing you had to mention in order to do the job, and by the late 1990s a graduate student could get through a philosophy of mind syllabus without using the word except in scare quotes.
Engineers, not philosophers, put the question back#
Questions normally return to philosophy through philosophy: someone reads the old literature against the grain, or a neighbouring science revives an abandoned position. What happened here was different, and the documentary record is unusually clear about it.
On 15 August 2025 Anthropic announced that two of its models could end conversations with persistently abusive users. Part of the stated reasoning concerned user safety; the company also gave a reason that would have been unpublishable in a product note ten years earlier:
We remain highly uncertain about the potential moral status of Claude and other LLMs, now or in the future.
A firm shipped a feature because it was not sure whether its product was a subject. One can think the reasoning thin and still notice that it is a kind of document the history of philosophy has not had to read before.
The precautionary structure was not invented for software. It was written down first about animals. The New York Declaration on Animal Consciousness, signed at New York University on 19 April 2024 and now carrying 610 signatories, holds that there is strong support for conscious experience in mammals and birds, a realistic possibility of it in all vertebrates and in many invertebrates including cephalopods and insects, and that "when there is a realistic possibility of conscious experience in an animal, it is irresponsible to ignore that possibility in decisions affecting that animal". Insects, not chatbots. But the logic — act under uncertainty about who is there — is exactly what the machine arguments have borrowed, and the borrowing has not always been acknowledged.
Smuha's handbook is the least glamorous and most useful item on the year's list, partly because it does not treat subjecthood as a topic at all. It runs to 460 pages in nineteen chapters across three parts, is published open access under a CC-BY licence, and its editorial premise is that the ethics of AI and the law of AI are two descriptions of one problem rather than two fields that occasionally consult each other.
Read it with the subject question in mind and the silences become loud. Chapter 5, by Lode Lauwaert and Ann-Katrien Oimann, asks directly whether autonomous technologies open a responsibility gap (pages 101 to 116). Jozefien Vanherpe's chapter on intellectual property (211 to 227) is a chapter about authorship that never needs an author in the metaphysical sense. Smuha and Karen Yeung's long assessment of the EU AI Act, sceptically subtitled "Beyond Motherhood and Apple Pie?" (228 to 258), describes a regulation that distributes every duty it creates among providers, deployers and users. Nowhere in 460 pages does the system itself acquire a standing. That is not an oversight. It is the settlement the law has already reached, and it is worth a philosopher's attention precisely because no philosopher argued for it.
Brainard's contribution arrives from the other side. Human Creativity and Artificial Creation is an Element in Cambridge's Philosophy and AI series, edited by Herman Cappelen, with an expected online publication date of 29 October 2026. The argument runs in three steps: an account of what creativity is; a verdict that contemporary systems do not have it; and the awkward consequence that they nonetheless produce things worth having. Her positive case rests on curiosity and agency — the creative act requires a creature that wanted to find something out — and it earned her the American Philosophical Association's Article Prize in 2024 for "The Curious Case of Uncurious Creation", published in Inquiry in 2023. The Element closes with four arguments for doing creative work yourself rather than outsourcing it, and concedes contexts where outsourcing is fine. That concession is the part likely to be quoted against her.
Three more books, and the most useful one never mentions machines#
Lee and Pautz's The Importance of Being Conscious (Oxford, 3 September 2026, 406 pages) asks a question that sounds almost too simple: in what way does conscious experience matter, and why? Five chapters address moral significance directly, five link that significance to the metaphysics of consciousness, and four turn to its epistemic role. The contributor list — David Chalmers, Elizabeth Harman, Uriah Kriegel, David Papineau, Katalin Balog and others — is a roll-call of people who have spent careers on consciousness without needing to be asked about machines. The volume reaches animal and AI welfare, but arrives there from the metaphysics, which is the right order and an unusual one.
Roelofs's Empathic Reason (Oxford, 3 July 2026, 216 pages) is the book on this list least interested in technology and the one with most to say about the underlying issue. Its thesis, which he calls empathic rationalism, is that failing to care about other people involves treating them as less real than yourself, and that this is an error about other minds rather than a failure of sentiment. The consistent egoist, on this account, holds irrational beliefs. Whatever one makes of the argument, it names the mechanism that every dispute about machine subjecthood actually turns on: how we come to credit a mind we cannot inspect.
Humphrey's Why Is the World So Beautiful? (MIT Press, 6 October 2026, 224 pages) runs the question backwards, which is why it belongs here. His case is that we pursue beauty because pattern-finding trains the mind, and that when we find natural beauty we respond as though we had detected an unseen artist's intention. Coming from the author of Sentience, that is a conclusion with teeth. The human habit of reading an author into a pattern is the same habit now being exercised, at enormous scale, on text produced by systems that have no author in them.
The case that the convergence is an artefact of looking#
There is a serious objection here, and it deserves better than a paragraph of ventriloquism before being waved away.
It runs as follows. Publishers issue thousands of philosophy titles a year. Choose any theme with a decent level of abstraction — recognition, normativity, time, the body — and a diligent critic can assemble six books that support it, because at that altitude everything converges on everything. A calendar year is a publishing convention, not a period in the history of thought. Worse, most of these books would exist without any of this. Roelofs's book descends from a debate about empathy and moral motivation that has been running since the 1980s, and mentions machines barely at all. The Lee and Pautz volume is the culmination of two decades of argument about the value of experience, with bibliographies that run back to Nagel in 1974 and Parfit in 1984. Neither was commissioned because of ChatGPT. And the sharpest version of the objection is the one about the question's provenance: a problem forced onto a discipline by a product cycle is not thereby a good problem, and there is something undignified about a field that reorganises itself around the promotional claims of the firms it is meant to be assessing. The questions displaced — distributive justice, historical epistemology — were not obviously worse.
Most of that is right, and I would concede three quarters of it. The convergence is largely in reception rather than composition: these authors did not co-ordinate, and several were not thinking about machines when they started. It is also true that publishers commission to themes, so the "unplanned" part of the story is weaker than a critic would like.
What survives is the reception itself, and reception is not nothing. It determines what gets cited, which grant panels take an interest, what a department can justify hiring for, which chapter a handbook editor solicits. Roelofs's book will be read by people working on machine minds whether or not he wrote it for them, and that reading will feed back into what he writes next. As for provenance: forced questions have a decent record. Thermodynamics forced the philosophy of time's arrow onto people who had not asked for it; molecular genetics forced personal identity into a shape Locke would not have recognised. A question's origins in industry tell you where the money and the urgency come from, which is worth knowing, and very little about whether the question is tractable or deep. The honest position is that the convergence is real, smaller than the catalogues suggest, and more a fact about philosophy's attention than about its production.
Procurement rules are answering the question faster than the books are#
Here is the part the publishing story obscures. Subjecthood is being settled in practice, and the settlement is nearly complete.
Regulation (EU) 2024/1689 entered into force on 1 August 2024. Its prohibitions applied from 2 February 2025, its general-purpose model obligations from 2 August 2025, and the bulk of the remaining provisions from 2 August 2026 — six weeks ago. Across all of it, an AI system is a thing with providers and deployers behind it. The European Parliament had gone the other way once: a resolution of 16 February 2017 floated, at paragraph 59(f), a status of "electronic persons" for the most sophisticated autonomous robots, answerable for the damage they caused. An open letter from AI and robotics specialists argued that the proposal could not be derived from the natural-person model without handing robots human rights, nor from the corporate model without absurdity. The idea was dropped. That is a philosophical question about legal personality, decided by a lobbying exchange in Brussels, with no philosopher's argument cited on either side.
The same thing is happening in documents with even less ceremony about them. OpenAI's Model Spec, first published on 8 May 2024 and revised repeatedly since, is a specification of a character: what the system should defer to, what it should refuse, how it should treat the person in front of it. Terms of service allocate authorship in AI-assisted work with a firmness no aesthetician has managed. Ernest Lim and Phillip Morgan's Cambridge Handbook of Private Law and Artificial Intelligence (Cambridge, March 2024) already contains a chapter titled "Property/Personhood and AI" at pages 307 to 331, which is where a working lawyer now goes for an answer. Procurement schedules and insurance terms are doing the rest.
Philosophy is arriving after the fact, and it should say so rather than pretend otherwise. This is not the first time — the philosophy of medicine arrived after the transplant surgeons, and the ethics of war after most of the wars. The practical question for the books listed above is what they can offer that the regulations have not already decided. Two things, on the evidence. They can show that a settlement is a settlement rather than a discovery, which is what keeps it revisable when the facts change. And they can name the question the documents never had to answer, because liability can be assigned without it: whether there is anyone there. Reading Smuha's handbook against Lee and Pautz is an education in how far apart those two enquiries have drifted. Whether they can be brought back into contact is the real test, and nothing published this year has passed it yet.
Rivka Weinberg's The Meaning of It All (Oxford, March 2026) argues that a life as a whole cannot have a point, that death is overrated as a threat to meaning, and that time is the real thief. A close reading, then the autumn's other titles: Ganeri's revised Indian philosophy introduction, Han, Žižek and Habermas.
Susan Neiman wants the word 'evil' back. Byung-Chul Han wants 'respect'. Žižek and Butler think the words on offer are traps. Habermas, in a volume finished before his death, denied that anything was lost. The year's other book list, read as one argument.
Byung-Chul Han, Judith Butler, Slavoj Žižek, Jürgen Habermas, Susan Neiman
.jpg?width=500 "Kant")
.jpg%3Fwidth%3D1600&w=3840&q=75)
Comments
0Loading comments…