62Being You

Anil Seth, Being You: A New Science of Consciousness, Faber, 2021

Linked to 30 chapters: supports 16 · contradicts 10 · mechanism 16 · same claim 7 · pairing 4

Seth's proposal is to stop asking why there is consciousness at all and to explain instead, one property at a time, why each experience is the way it is and not some other way. He calls this the real problem, and sets it between David Chalmers's hard problem, which asks why physical processing should be accompanied by experience at all, and the easy problems, which ask how brains perform functions. His model is the history of life: biologists never found the spark of life, they explained reproduction, metabolism and self-repair one at a time until vitalism had nothing left to do. Seth divides consciousness the same way, into level, content and self. Level yields to measurement — perturb the cortex, compress the echo, and a single number separates a dreaming brain from an anaesthetised one. Content yields to prediction, since what we perceive is the brain's best guess at the causes of its sensory signals, a controlled hallucination that keeps its grip on the world because the guesses are continually corrected by error. Self yields to the same account turned inward, onto a body whose states are predicted not in order to be known but in order to be regulated. From that comes the conclusion the book is named for. We are conscious because we are alive, not because we are clever.

The real problem

Seth takes his definition from Thomas Nagel's paper of 1974: a creature is conscious if there is something it is like to be that creature. What matters is phenomenology, the character of an experience, and that starting point lets him set aside the field's two leading theories, global workspace and higher-order thought, as foregrounding function where he wants to begin from experience. On metaphysics he is a physicalist who is agnostic and slightly suspicious about functionalism, the further claim that consciousness depends only on what a system does and not on what it is made of. A computer playing Go really is playing Go; a simulated rainstorm is not wet; nobody knows which case consciousness resembles.

He does dispose of one argument against physicalism. The philosophical zombie is a creature identical to a conscious one in every physical respect but with no inner life, and its conceivability is supposed to show that consciousness cannot be physical. Conceivability arguments, Seth answers, have a plausibility inversely related to knowledge: you can picture an Airbus A380 flying backwards until you know some aerodynamics, at which point you cannot. A zombie asks you to picture the capacities of eighty-six billion neurons and a thousand times as many synapses, and the failure of imagination is a fact about the imaginer. It was the same flaw that sank vitalism.

What replaces the hard problem is a research programme: explain, predict and control the phenomenological properties of experience in terms of mechanisms in the brain and body. Not to establish that a pattern of activity accompanies the experience of redness, but to say why redness is the way it is and not like blueness or like jealousy.

This is also his complaint about the method that has dominated the field since Francis Crick and Christof Koch set it going in 1990, the search for neural correlates of consciousness: find a case where the input holds still while experience changes, such as binocular rivalry, and compare brain activity across the two. Correlations are not explanations, and collecting more of them leaves the explanatory gap where it was. Build bridges instead, from properties of mechanism to properties of experience, and Seth expects the gap to narrow until the hard-problem intuition fades of its own accord.

Measuring consciousness

The chapter on level opens with the invention of the thermometer, whose fixed point was for a while a cool cellar under the Observatoire in Paris and, in one proposal, the melting point of butter. The moral is that measurement precedes understanding: thermodynamics did not merely correlate mean molecular kinetic energy with temperature, it said that this is what temperature is. Anaesthesia, meanwhile, is given to more than four million people a day, and the monitors in use combine electroencephalographic measures into a number without resting on any theory of what is being measured.

The first thing such a measure must do is separate consciousness from wakefulness. The two come apart both ways: a dreamer is asleep and having vivid experiences, while a patient in the vegetative state cycles through sleep and waking with no sign of awareness. Nor is the answer the number of neurons involved, since the cerebellum holds about three quarters of the brain's neurons and people born without one lead ordinary lives. What matters is how widely and how intricately the thalamocortical system talks to itself.

Marcello Massimini and Giulio Tononi made that quantitative. They used transcranial magnetic stimulation to push a sharp pulse through the skull into the cortex and electroencephalography to record the echo; the image Seth uses is banging on the brain with an electrical hammer and listening. Under anaesthesia and in dreamless sleep the echo dies away like ripples from a stone; in conscious states it travels widely and returns in complicated patterns. Compress that pattern with the algorithm that shrinks a photograph, and the length of the compressed version is the number: the perturbational complexity index.

In a study of brain-injured patients published in 2013 the index matched diagnoses neurologists had reached independently, separating the vegetative from the minimally conscious state and giving locked-in patients scores indistinguishable from healthy controls. Seth tells of a young man in Milan, unresponsive and headed for a diagnosis of vegetative state, whose index was as high as a healthy person's; an uncle flown in from North Africa spoke to him in Arabic and he smiled at jokes and gave a thumbs-up at a film. He had been conscious throughout, and unresponsive only in Italian. Estimates put covert consciousness in ten to twenty per cent of patients diagnosed as vegetative, and Seth thinks that low.

Measuring the complexity of ongoing activity rather than of an evoked echo, his own group found it falling in sleep and anaesthesia and rising, for the first time above ordinary waking rest, in people given psilocybin, LSD or low doses of ketamine. The result also embarrassed the measure, because pure randomness scores highest of all, and a brain with every neuron firing at will would presumably experience nothing; free-form jazz, as he puts it, stops at some point being music.

Complexity in the useful sense is the middle ground, which is where Tononi and Gerald Edelman placed consciousness in a 1998 paper in Science. Every conscious experience is informative, because having this one rules out the vast number of others you might have had, so that redness is redness because of all the things it is not; and every one is integrated, arriving as a single scene in which colours are not detached from shapes. If experience has both properties, the mechanism should have both too.

Integrated information theory turns that into an identity. Its measure, phi, captures how much a system generates as a whole beyond what its parts generate separately, and a system is conscious, on Tononi's claim, to exactly the degree that its phi is high. A camera sensor has none: the array carries a great deal of information, but each photodiode is causally independent, so the whole is no more than the sum — which is also why the cerebellum, wired much the same way, contributes so little.

Seth has been arguing with Tononi about this since an evening in Las Vegas in 2006, and his objection is not that the theory is strange but that its ambition costs it its testability. Phi must be computed from information intrinsic to the system rather than relative to an observer, since otherwise consciousness would be observer-relative too, and measuring intrinsic information means knowing not what a system does but everything it could do, which for a brain is unavailable. The consequences are odd to match: graft on a few neurons that would fire only if you ate a rare Japanese watermelon and, though they never fire, every experience you have changes, because the repertoire has changed. Seth's verdict is to keep the insight and drop the identity: integration and information are properties of experience to be explained, not axioms about what consciousness is. That is the difference between treating consciousness as temperature and treating it as life.

Perceiving from the inside out

Imagine being a brain: sealed inside the skull, no light, no sound, receiving electrical signals that arrive without labels saying which sense they belong to, let alone what caused them. The commonsense account has those signals analysed upward through a hierarchy, edges to object parts to faces and coffee cups, until a picture is assembled for a self behind the eyes. Seth reverses it with an exchange between Wittgenstein and Elizabeth Anscombe: it was natural to think the sun went round the earth because that is what it looked like, so what would it have looked like if the earth had been turning on its axis? Predictions flow down and out; sensory signals flow up and in only as prediction errors, the difference between what the brain expected and what it got. What we experience is the content of the predictions, never the signals.

The idea is old: Ibn al-Haytham's judgement and inference around 1030, Kant's noumenon behind the sensory veil, Hermann von Helmholtz's unconscious inference in the late nineteenth century. Seth's term for it, which he first heard from Chris Frith, is controlled hallucination, and the qualifier does half the work, since hallucination is uncontrolled perception.

Colour makes the case. A white page taken outdoors stays white although the light it reflects has changed completely, because the visual system discounts the illuminant and tracks an invariant property of the surface, the way in which it reflects light. The chair in the corner is not red; its surface has a property the brain keeps track of, and redness is how that tracking appears in experience. The Dress, which arrived in February 2015 and split Seth's laboratory roughly in half, works because people differ in the illuminant they assume for an overexposed photograph with no context; Adelson's checkerboard shows the same machinery running all the time, since two squares of identical grey look different because the brain knows what shadows do, and knowing they are identical does not help.

What makes the guess a best guess is Bayes. A wet lawn might mean overnight rain or a sprinkler left on, and the rule combines a prior with a likelihood to give a posterior. The brain's version runs continuously, each posterior becoming the next prior, on probability distributions rather than single numbers, so that each belief carries a confidence as well as a content.

Confidence is where the mechanism earns its keep, because the brain must estimate not only the causes of its signals but how reliable those signals are. This is precision weighting, and Seth's claim is that turning precision up on some signals is nothing other than paying attention — which turns it down on everything else, and is why Daniel Simons's viewers, counting passes between players in white, fail to see a person in a gorilla suit walk through the game. Action belongs to the same loop: a prediction error can be quenched by changing the prediction or by changing the world so that the signal fits, which Karl Friston named active inference. Moving the eyes across a cluttered desk until the car keys appear is a perceptual prediction making itself come true, and so is every movement, since actions are self-fulfilling proprioceptive predictions. Perception and action are not input and output flanking a mind but the same process with a different partner leading.

The other half of the argument comes from art history. In Vienna around 1900 Alois Riegl, and later Ernst Gombrich, named the beholder's share: the part of the experience of a painting contributed by the beholder and not present on the canvas. Gombrich's account of what happens when the brushstrokes of an Impressionist canvas come to life — that we have been led to project a scene into the dabs of pigment — is prediction error minimisation described from the inside. Pissarro's Hoarfrost at Ennery of 1873, dismissed as so much scraped paint, conveys a frosted field because it supplies the raw material for inference rather than its result. Keisuke Suzuki built the point the other way round, with a hallucination machine by running panoramic footage of the university square backwards through a neural network, so that predictions of dogs were projected into every part of the scene.

The same machinery handles the deep structure of experience, the way perception is arranged rather than what it contains. A tomato in the hand seems to have a back and a photograph of one does not, because objecthood is supplied by the brain's conditional predictions about what the signals would do if the object were turned. The sense that what we perceive is really out there belongs to the model too, on the pattern of Hume's argument that we project causality onto the world rather than finding it there, and it is useful, since an organism responds faster to what it takes to be really existing. We perceive the world not as it is but as it is useful for us, which is why, as he puts it, "when we agree about our hallucinations, that's what we call reality." Experience seems impossible to explain physically because its contents present themselves as mind-independent facts. "A chair has a mind-independent existence; chairness does not."

Expecting yourself

The self, on this account, is not the thing that does the perceiving. It is another perception, of a special kind and made for a different purpose. Seth opens with teletransportation: a machine scans Eva in London down to the arrangement of each molecule and rebuilds her on Mars, vaporising the original, until one day the vaporiser fails. Personal identity, he argues, turns on psychological rather than physical continuity, so that both Evas are the real Eva — which offends the intuition the exercise exists to expose, that a self is single, indivisible, and of a kind there can be only one of. Kant attacked that intuition, Hume replaced it with a bundle of perceptions, and it survives them both, along with the craniopagus twins in British Columbia, one of whom can feel the other drinking orange juice.

He then breaks the self into five. Embodied selfhood covers ownership of this particular body, with emotions, moods and arousal, and beneath them a formless feeling of being an embodied living organism at all. The perspectival self is the first-person point of view, apparently sitting somewhere behind the forehead, which Ernst Mach drew in 1886 by sketching exactly what he saw from his left eye. The volitional self is the experience of intending and of causing, which is what most people mean by free will. The narrative self is the thread of autobiographical memory that makes a personal identity possible, and with it emotions such as regret. The social self is how I perceive others perceiving me.

Each fails on its own terms, which is the evidence that they are separate constructions. Narrative selfhood erodes in dementia and warps in delirium; volition goes wrong in schizophrenia and alien hand syndrome; the perspectival self is dislodged in out-of-body experiences; body ownership fails in phantom limb syndrome, in somatoparaphrenia, where a limb is felt to belong to someone else, and in xenomelia, where a person wants a healthy limb amputated. The laboratory versions are milder. The rubber hand illusion, described by Botvinick and Cohen in 1998, strokes a visible rubber hand and a hidden real one at the same moment until the best guess arrives that the rubber hand belongs to the body; two papers in Science in 2007 extended it to the whole body using virtual reality.

Seth then turns on his own literature. The body illusions are weak: many people feel little, and nothing like the astonishment of Adelson's checkerboard ever happens. Worse, the set-up tells participants what they are supposed to feel, and a study led by Peter Lush and Zoltán Dienes, with Seth as an author, found that individual differences in illusion strength track hypnotic suggestibility. That fits the controlled hallucination view, since a suggestion is a top-down expectation, and it also raises the possibility that the rubber hand illusion is largely or entirely an effect of suggestion. Studies that have not controlled for it, which is most of them, can say little about mechanism. The clinical cases carry the weight instead.

For the narrative self the case is Clive Wearing, the musicologist whose herpes encephalitis in March 1985 destroyed the hippocampus in both hemispheres and left him with a present lasting between seven and thirty seconds. His diaries record one first awakening after another, each earlier entry angrily crossed out; his body ownership, his perspective and his voluntary action are intact, and he plays and conducts fluently. What has gone is the thread. The social self, for its part, requires perceiving other minds, and since my model of your mind contains your model of mine, we perceive ourselves refracted through other people. The chapter closes on the sentence that turns the book: we do not perceive ourselves in order to know ourselves, we perceive ourselves in order to control ourselves.

Being a beast machine

Descartes called animals bêtes-machines, beast machines: bodies that breathe, digest, perceive and move by the arrangement of their parts, with no mind directing them. In 1748 Julien Offray de La Mettrie extended the argument to us in l'homme machine, fled the Netherlands for Berlin, and three years later died of an excess of pâté. Seth's sympathies are with La Mettrie, and his thesis inverts Descartes exactly. Where Descartes made life irrelevant to mind, Seth makes mind a property of life.

The mechanism is interoception, perception of the body from within — heartbeats, blood pressure, blood chemistry, gastric tension, breathing — carried from the viscera through the brainstem and thalamus to the insular cortex. What these signals report is how well physiological regulation is going. In 1884 William James and, independently, Carl Lange proposed that emotions are perceptions of bodily change: we are sad because we perceive ourselves crying, not the other way round. The objection was that bodily states are not distinct enough to support the range of human emotion, and the answer of the 1960s was appraisal theory, in which a reading of the context does the discriminating — as in Dutton and Aron's bridge study of 1974, where men who had just crossed a high, rickety suspension bridge were far more likely to telephone the woman who had stopped them than men who crossed a solid one. Seth's objection is that appraisal theory needs a clean division between cognitive and non-cognitive processes, and the brain does not divide that way. His own proposal, interoceptive inference, needs one process rather than two: emotions and moods are the brain's best guesses about the causes of interoceptive signals, controlled hallucinations driven from the inside.

What makes this a theory of the self rather than of emotion comes from cybernetics, and from a theorem of Roger Conant and W. Ross Ashby published in 1970 whose title is the whole argument: every good regulator of a system must be a model of that system. Seth's illustration is a heating system asked to hold a house at nineteen degrees. A thermostat switches on below the target and off above it; a better regulator models how the house responds to heating and adjusts in advance, and with noisy sensors it must infer the temperature rather than read it. At no point is its goal to find out what is there. Its goal is to regulate. Ashby's earlier term for what is regulated is the essential variables: quantities such as core body temperature, which must stay between thirty-two and forty degrees or the organism dies.

From there the argument runs straight. The primary imperative of an organism is to stay alive, and this, Seth says, is why brains exist at all — not for poetry or crosswords but to hold physiological variables inside the ranges compatible with survival. Those variables are hidden causes like any other, so they must be inferred; the inference proceeds by prediction error minimisation; and the errors are quenched mostly by acting, through outward actions such as running from a bear and inward ones such as raising the heart rate or narrowing a blood vessel. Waiting for blood acidity to drift out of bounds before responding is a poor strategy, so the control is anticipatory, which physiology calls allostasis, stability achieved through change, as against homeostasis, a tendency towards equilibrium.

This explains the peculiar character of emotional experience. A coffee cup is perceived as having a back because the brain holds predictions about what would happen if it were turned; anxiety has no back. Interoception does not deliver the shapes and positions of the organs, so what is experienced instead is how well or badly things are going and are likely to go. The self feels stable, in turn, because regulation requires high-precision priors about the body's condition, priors that are largely self-fulfilling; effective regulation may depend on misperceiving the body as steadier than it really is. Seth is careful about what he has not shown. He is not claiming to have demonstrated that life is necessary for consciousness, and says nothing in the beast machine theory settles that either way.

Volition gets the same treatment. Benjamin Libet's experiments of the early 1980s found the readiness potential, a slow ramp in the electroencephalogram over motor cortex, beginning hundreds of milliseconds before people reported the urge to move, which they timed by noting the position of a rotating dot. Libet's own rescue was that room remained for a conscious veto, a free won't, which Seth dismisses: an inhibition is no less a miracle than an intention. The better answer came in 2012 from Aaron Schurger, who noticed that readiness potentials are found by averaging backwards from the moments when actions happened, and that nobody had looked at the moments when they did not — the error of studying only the fairground swings that ring the bell and concluding that a rising puck always rings it. Schurger showed that the same slow ramp precedes fast responses to a beep, when no voluntary action was being prepared, and not slow ones. The readiness potential is not the brain initiating an action; it is ongoing fluctuation that occasionally crosses a threshold.

What survives is free will as a competence rather than an intervention. Seth lists three features of the experience: that I am doing what I want to do, that I could have done otherwise, and that the action comes from within rather than being imposed. The first is bounded by Schopenhauer's observation that a man can do what he wills but cannot will what he wills. The second is false as a claim about physics and useful as a claim about the future, since circumstances never repeat exactly and the feeling that I could have done otherwise is part of what makes it likely that next time I will. The chapter's title names the capacity involved. A rock has no degrees of freedom, a train on a track has one, an ant has several and a person a great many; voluntary action is the competence to control them in line with one's own beliefs and goals. That competence is real because it can be lost: a tumour pressing on the amygdala turned Charles Whitman into the Texas Tower sniper.

Beyond human

Absence of language, Seth insists, is not evidence for absence of consciousness. Pre-Cartesian Europe was readier than we are to credit animals with minds, to the point of putting pigs on trial, and what replaces both that credulity and Descartes' denial is Jakob von Uexküll's Umwelt, the world as it is for a creature given its own senses and needs. The question becomes not how conscious an animal is on a single scale but what kind of consciousness it has. Even among mammals the variation is wide: a koala sleeps twenty-two hours a day, a giraffe fewer than four.

The octopus is his case for how far a mind can differ. Much of its nervous system lies in its arms; its skin changes colour through chromatophores although the animal is colour-blind, so it controls an appearance it cannot see; and a severed arm will grip a skinned arm from another octopus but not an intact one, which suggests a chemical rather than spatial sense of what belongs to it. His verdict on the literature is characteristic: all of it is suggestive, none conclusive. For the harder question, which animals can suffer, he prefers evidence of the beast machine kind — zebrafish that pay a cost to obtain a painkiller, fruit flies that develop something like chronic pain — and the principle that you do not have to be clever to suffer. His intuition stops somewhere: a nematode with three hundred and two neurons, he doubts, is anyone.

The machine chapter rests on the same separation. Much of what was called artificial intelligence in 2021 is, he writes, sophisticated pattern recognition with some planning attached, and whether or not it is intelligent it does what it does without being conscious of anything; making computers cleverer will not make them sentient, though that does not make machine consciousness impossible. His test case is a silicon beast machine: a robot whose controller implements predictive processing and active inference, with artificial interoceptive channels reporting battery level and the integrity of its actuators, generating synthetic emotional states that guide its behaviour and giving every outward impression of sentience. Would it be conscious? He does not know, he says, but probably not, and the reason is that self-maintenance in living systems does not stop at the whole organism. Every cell is continually regenerating the conditions of its own persistence, which no computer does and no silicon beast machine would. The ethics follow from an asymmetry: a conscious machine would acquire moral status, and we might not recognise the forms its suffering took. In June 2019 Thomas Metzinger proposed a thirty-year moratorium on research aimed at synthetic phenomenology, and Seth, who was in the room, thinks the letter of it unworkable and the spirit right. The nearer worry is not sentient machines but machines we feel to be sentient, and what living among them will do to us.

Since 2021

The clinical half of the book has aged best. A multicentre study published in the New England Journal of Medicine in 2024 found that around a quarter of patients with severe brain injury who showed no observable response to commands nonetheless produced brain responses to them, a higher rate than the ten to twenty per cent Seth reports, and detecting covert consciousness has moved towards a standard of care faster than any theory of consciousness has settled.

The theories have not settled. The adversarial collaboration Seth mentions being involved in ran, and its first large result, published in 2025, tested integrated information theory against global workspace theory and left neither intact: content could be decoded from posterior cortex for as long as it was experienced, as integrated information theory expects, but the sustained synchronisation it predicts was absent, and so was the frontal ignition global workspace theory predicts when a stimulus ends. Meanwhile the dispute turned sharp. In September 2023 more than a hundred researchers signed a letter calling integrated information theory pseudoscience, largely on the ground Seth had put more politely in this book: that the quantity at its centre cannot be measured for any real brain. The suggestibility challenge he raised against his own body-ownership experiments has likewise not been resolved.

The controlled hallucination account has held up the way frameworks do, by organising work rather than by surviving a decisive test, which is also the standing complaint against it and against Karl Friston's free energy principle, whose unfalsifiability Seth concedes. The machine question moved fastest. Within a year of publication, large language models made systems that seem conscious a matter of practice rather than thought experiment, and in 2022 an engineer publicly claimed that one of them was sentient. Seth joined a 2023 report that drew indicators of consciousness from several competing theories and applied them to existing systems, and in 2025 he set out the biological case at length in an essay for Noema, sharpening this book's last chapter into a direct argument that computation is not enough and that life may be necessary.