Four books about the inside of a human life, in a year of books about machines

Cambridge University Press publishes Consciousness and Inner Awareness today, edited by Davide Bordini, Arnaud Dewalque and Anna Giustina: 288 pages, hardback now, paperback at the end of the month. Not one chapter of it is about artificial intelligence. After the year philosophy publishing has had, a book on the conscious mind that declines to mention machines starts to look like a position rather than an omission.

Yesterday this site ran a survey of eight titles from 2026 that all arrive, by different roads, at the same question about machines. This is the other list, and it is not conveniently bunched into one season: the Cambridge volume; Michael Pollan's A World Appears, from Penguin Press on 24 February; Improving Character, forty-five commissioned essays edited by Robert J. Hartman for Wiley-Blackwell in June; and Terry Eagleton's A Death of One's Own, from Yale at £18.99, reviewed by Kathryn Hughes in the Guardian on 3 September. Consciousness, character, mortality. Between them they cover most of what is involved in being a person.

None of this makes the four a refuge from the machine question. They are something more awkward for it: the place its vocabulary came from. The machine-consciousness debate runs on a notion of inner awareness it almost never defines. The alignment literature runs on a notion of character it has not examined since it borrowed the word. And the argument about whether a system can be wronged by being switched off runs on an account of why death is bad that nobody in that argument has written down.

The Awareness Principle is the load-bearing claim in a debate that never states it#

Cambridge's own description of the volume puts the question with unusual bluntness. It asks whether the following is true:

each conscious mental state is such that its subject is aware of it

Call that the Awareness Principle. It sounds close to a truism, and much of twentieth-century philosophy of mind treated it as one. If you are in pain, surely you know you are in pain; a pain nobody notices, including the person having it, is barely a pain at all. The trouble starts when you ask what the noticing consists in.

Two families of answer have been arguing since the 1980s. Higher-order theories, of which David Rosenthal's is the best developed, say that a mental state becomes conscious in virtue of a further, numerically distinct state that is about it. Your pain is conscious because a separate thought is directed at it; consciousness is a relation between two states, and the second does the work. Same-order theories deny the split. Uriah Kriegel's self-representational account, in Subjective Consciousness (2009), holds that the awareness is a constituent of the conscious state itself — the pain includes, as part of what it is, an awareness of itself. The ancestry runs back through Brentano's inner perception, which is presumably why the editors' introduction is called "Inner Awareness: Past and Present".

It matters which is right, in a place the contributors mostly do not go. If the higher-order theorists are correct, the condition on consciousness is architectural: a system needs states that represent its own states, and whether it has them could in principle be established by looking at how it is built. If the same-order theorists are correct, the condition is phenomenal rather than structural, and no architectural description will settle it, because what you are after is a feature of how an experience feels from within. Every recent attempt to reason from a system's wiring to a verdict about its experience has quietly assumed the first answer. Readers who followed the argument about reading consciousness off behaviour will recognise the difficulty.

The volume itself does not oblige. Daniel Stoljar's chapter, "Rationality and Acquaintance in Theories of Introspection", declines both options: introspective knowledge, he argues, is better explained by a rational agent's capacity to form beliefs about the states it is in than by any relation of acquaintance, and acquaintance theories carry commitments about non-rational creatures they cannot honour. Not a machine-friendly conclusion, but it moves the burden. What a system needs to know its own mind, on Stoljar's account, is rationality rather than a special inner glow — a lower bar, and a more contestable one.

Pollan is much better once he stops interviewing the scientists#

A World Appears came out on 24 February, ran to 320 pages, went onto the New York Times bestseller list and Barack Obama's summer reading list, and was a Kirkus Prize finalist. It promises a journey into consciousness by way of neuroscience, philosophy, literature, Buddhist practice and psychedelics, and it delivers all five, including an odd stretch in which Pollan looks for something like awareness in plants and another in which he asks whether a machine could have an inner life. So my claim that these four ignore artificial intelligence is not quite true here.

The journalist's book on the hard problem is a genre with a poor record. It interviews the scientists, reports them all with equal warmth, and ends on wonder. Writing in Commonweal on 26 May, Kit Wilson made the sharpest version of the complaint: the first half of A World Appears gives a platform to researchers who reduce consciousness to homeostasis, or inference, or computation, and Pollan lets them. Wilson found their explanations "laughable" and thought accepting them took "willful lack of introspection". He was hardest on the appeal to emergence, which re-describes the problem rather than solving it.

Wilson is right about the second half and slightly unfair about the first. The phenomenological chapters — spontaneous thought, ordinary perception, Husserl and Proust pressed into service without embarrassment — are the best popular writing on the subject in years, and Wilson says so, calling them humane and philosophically rich. But reporting the reductionists is a service too, since serious people hold that position and a reader deserves to see it stated by its holders. The failure is narrower and worse: the book never sets out the hard problem sharply enough for a reader to see why the emergence answer misses it. Get that framing right on page forty and the reductionist chapters read as an argument; leave it out and they read as opinions in a row. Christof Koch's report of a psychedelic experience in which, he says, "I accessed this universal mind", is the case in point. Without a framework it is anecdote.

Hartman's contributors answer the situationists by declining to argue with them#

The argument the book is really answering is twenty-five years old and has never been settled. In 1999 Gilbert Harman told the Aristotelian Society that ordinary attributions of character traits are systematically mistaken, a product of the fundamental attribution error; John Doris made the fuller case in Lack of Character (2002). Both leaned on social psychology in which trivial situational variation swamps supposed dispositions. The famous item is Isen and Levin's 1972 study: subjects who found a dime in a phone box's coin return went on to help a stranger who dropped a sheaf of papers; those who found nothing did not. Eighty-eight per cent against 4 per cent. Darley and Batson's Princeton seminarians, hurrying to a talk on the Good Samaritan and stepping over a slumped man on the way, made the point with better irony. If a ten-cent coin can do that, what is compassion?

Hartman's volume, published by Wiley-Blackwell in June, does not take that argument head on. Forty-five newly commissioned essays run across four parts: the basics of moral virtue and the reasons for wanting it; individual virtues from practical wisdom and courage through to love and hospitality, with Confucian and Buddhist material alongside the Aristotelian default; strategies for cultivation, among them habituation and scripts for honesty, friendship for practical wisdom, reading literature for love; and a closing set of harder questions, including whether virtue signalling can be a virtue and whether bad luck can block the development of character. An appendix on A Christmas Carol is either charming or a sign that the book knows exactly which market it is for. Probably both. At 448 pages, aimed frankly at undergraduates, it will not satisfy a reader who came for metaphysics.

The strategy is real, not an evasion. If specific interventions reliably produce more honest or more courageous behaviour, the situationist finding becomes a claim about baselines rather than possibility. Doris never denied that character could be built; he denied that most people have the globally consistent traits virtue ethics presupposes. A book that treats improvement as a practical problem concedes the first point and gets on with the second. Whether the strategies work is an empirical question forty-five short essays cannot answer, and readers who want the evidence assessed rather than assumed should go to Christian Miller's survey work. What interests me is the shift of register. Virtue ethics spent forty years after MacIntyre arguing about whether it was a coherent theory. This is what it looks like when it decides to be a curriculum.

Eagleton will not let his book be used as a resource#

Eagleton is eighty-three. A Death of One's Own is what a lifetime of Marxist criticism and enormous reading produces when pointed at mortality: erudite, unsystematic, funny in a way that keeps catching the reader off guard. Kathryn Hughes, reviewing it in the Guardian on 3 September, picked out its impatience with sloppiness about death. Dylan Thomas's injunction not to go gentle into that good night gets short shrift, on the reasonable ground that if the night really is good there is nothing to rage against. She also picked out the insistence that death is indifferent to the circumstances of its arrival; whether you go in ordinary clothes or, in Eagleton's example, a giraffe outfit, the difference stops mattering quickly. What he calls the erotic bliss of yielding to an impregnable power is mischievous rather than consoling.

Hughes ends by recommending the book be taken slowly, a bit each night before sleep, and reports Eagleton's own doubt that living in constant awareness of death is worth the effort. I part company with the review here, on framing rather than fact. Prescribing a book on mortality as bedtime reading in small doses domesticates it, and the concession that death-awareness may be more trouble than it is worth is the one move that lets a reader off. The interesting claim is the opposite one. A life is intelligible as a whole only because it ends, and the shape of a finished thing is not available to anything that does not finish.

Which brings the argument round. To know whether shutting down a model wrongs it you need an account of why death is a harm at all, and the standard modern account is Thomas Nagel's, in "Death" (1970): death is bad because it deprives its subject of the goods a continued life would have contained. Deprivation accounts presuppose a subject with a life that was going somewhere. The AI welfare literature has largely skipped this step, treating the badness of termination as self-evident and arguing instead about consciousness. Both questions need answering, and the second does not settle the first.

What the machine debate has been quietly taking#

Put the three borrowings together and a pattern shows. Machine-consciousness arguments need the Awareness Principle to be true and need the higher-order reading of it, because only that reading makes the question tractable from outside. They rarely say either thing aloud. The alignment vocabulary of helpful, honest and harmless dispositions instilled by training is virtue-ethical to its bones, and it inherits the situationist problem in aggravated form: a system whose behaviour swings with the phrasing of a prompt is the most vivid demonstration of situational sensitivity anyone has built. Isen and Levin needed a dime.

There is a fair objection. Nobody is obliged to cite the sources of an idea they reached independently, and philosophy has always worked by half-remembered borrowing. But these borrowings are load-bearing, and an unstated premise is an untested one, which is how a literature ends up confident and unmoored at once. The fog around machine consciousness is not entirely a fog about machines.

The judgement: Cambridge, and not by a small margin#

This is my own view and I mark it as such. Pollan will sell more copies than the other three combined, Eagleton's is the book people will press on friends, and Hartman's will be assigned in more classrooms than the rest of the list together. On reach, the Cambridge volume finishes last: expensive, technical, published into a market looking elsewhere.

It will still matter most, because it is the only one of the four attempting to fix a definition. The other three enrich a subject; this one tries to settle what a term means. A debate can survive without good popular exposition or a decent textbook. It cannot survive indefinitely on a concept nobody has pinned down, and inner awareness is now doing more work in more arguments than any other undefined term in philosophy of mind. If the machine-consciousness literature is still incoherent in 2031, it will be because it never read the book that came out while it was busy.