55The Ego Tunnel
Thomas Metzinger, The Ego Tunnel: The Science of the Mind and the Myth of the Self, Basic Books, 2009
Linked to 28 chapters: supports 12 · contradicts 10 · mechanism 14 · same claim 9 · pairing 6
Metzinger opens by announcing that nobody has ever been or had a self. He is a philosopher at Mainz who had spent a decade working with neuroscientists, and the book is the non-technical statement of an argument he made at length in Being No One in 2003. The claim is narrower than it sounds: not that people do not exist, or that the word "I" fails to refer, but that the thing the word seems to name — an indivisible entity, a bearer of experience, something that could in principle survive the body — is not there. What is there is a process. The brain builds a model of the organism as a whole, places it inside a model of the world, and is then unable to recognise either as a model. Being someone is what it is like to run a self-model you cannot see as a self-model. The book establishes this in stages: an experiment in which a healthy adult comes to feel a rubber hand as part of her body, six problems any theory of consciousness must solve, experiments in which the sense of being located in one's own body is moved across a room, what would follow if the same architecture were built in a machine, and the question the whole enterprise forces — if states of consciousness can be engineered, which ones should there be.
The phenomenal self-model
The entry point is an experiment conducted in 1998 by the Pittsburgh psychiatrists Matthew Botvinick and Jonathan Cohen. A subject watches a rubber hand lying on a desk while her own corresponding hand is hidden behind a screen. Both are stroked with a probe, at the same time, in the same way. After a while — sixty to ninety seconds, in Metzinger's own case — the rubber hand becomes hers. She feels the strokes in it, and feels a connecting arm running from her shoulder to the object on the table. What Metzinger reports noticing just before the illusion arrived was a tingling in the shoulder, as though something were sliding out of the invisible arm and into the rubber one.
Nothing slid anywhere. What changed was the content of what Metzinger calls the phenomenal self-model, or PSM: the conscious model of the organism as a whole that the brain keeps active, phenomenal meaning known purely by how things appear from inside. The content of the PSM is the Ego. Two lessons follow. Whatever enters the self-model arrives carrying a quality he calls mineness — it is felt as your limb, your sensation, your thought. And if a part can be captured this way, the question is whether the whole can: whether the sense of being a body, located somewhere, identified with something, can be moved too.
The model is not a picture the organism inspects but a medium it looks through. This is the transparency thesis, and it carries the book. A representation is transparent when the system using it cannot recognise it as a representation. "We do not see the window but only the bird flying by." We do not see neurons firing; we see what they stand for. A large part of the self-model, on Metzinger's account, is transparent in exactly this sense, and that is why there is a self at all. The Ego is not a homunculus but an image, and an image you cannot see as an image is not experienced as an image; it is experienced as the plain fact of being you. You do not see it. You see with it.
The title metaphor follows. Conscious experience is a tunnel because it is not a picture of reality but a selective path through it: what reaches awareness is a small fraction of what exists, because the senses evolved for survival rather than for depicting the world in its depth, so the conscious model is a low-dimensional projection of something incomparably richer. Metzinger reaches for Plato's cave and corrects it — the fire is neural dynamics, the wall is a high-dimensional state space rather than a flat surface, and there are no eternal forms behind the shadows. Because the organism has memory and moves forward in time, the projection is not a screen but a tunnel; and because the self-model sits inside the world-model, the tunnel has a centre. That centre is the first-person perspective.
He separates this from the "reality tunnel" of Robert Anton Wilson and Timothy Leary. We do not construct our own world, only a world-model; the idea of being directly in touch with reality is folklore, since representing something correctly is what knowing is; and the tunnel is no prison, because knowledge is possible through communities who build theories and criticise one another. What cannot be done is to step outside it experientially. "But no such things as selves exist in the world." A biological organism is not a self, and an Ego is not one either, but the content of a transparent self-model running in a brain. Nobody lives in the tunnel. We live with it and through it.
A tour of the tunnel
The second chapter sets out six problems in ascending order of difficulty. The first is the unity of consciousness: for a world to appear it has to be one world first, and the binding of colour, texture, sound and touch into a single scene is an achievement, as the clinical failures show — disjunctive agnosia, for instance, in which a patient sees a mouth moving and hears speech that does not belong to it. Metzinger's preferred account is Giulio Tononi's dynamical core, an island of densely coupled causal relations emerging from a less coherent sea of neural activity, its border functional rather than anatomical, held together by timing. Wolf Singer's gamma-band oscillations, neurons firing in synchrony about forty times a second, supply the mechanism, and Metzinger adds the detail he finds most telling: when Antoine Lutz recorded Tibetan monks with at least ten thousand hours of practice, the synchrony was the strongest reported in the literature, and it took tens of seconds to build, because an assembly that large needs time to form.
The Now Problem is harder. A complete physical description of the universe would contain no indexical terms, no arrow saying here or now; the conscious brain supplies them. Everything you experience, memories and plans included, is experienced as happening in a present moment: you remember the beach now, you plan the climb now. Ernst Pöppel and Eva Ruhnau's temporal gestalt gives the unit — a musical motif, a line of verse, a thought — with an upper bound near three seconds, beyond which nothing holds together as one moment. Metzinger's explanation of why a Now was worth building is that the brain's specialised mechanisms need a common frame of reference to work on the same information at once, and the cheapest one is temporal. William James called the present a saddleback rather than a knife-edge; Metzinger takes the image and insists on the cost. Neural processing takes time, so what is experienced as the present is already past. There are no colours out there, and there is no present moment either.
The Reality Problem asks how a world-model plus a Now-model becomes the presence of a world, and the answer is transparency. Metzinger explains its evolution through a notion he took from neuroscientists: metabolic price. Any new capacity is paid for in sugar, which must be found in the environment. An ancestor needed to know bear over there, not a bear-representation is currently active in my brain; the second requires metarepresentation, which requires hardware, which requires fuel. Transparency was cheaper. He adds a structural reason: the process that builds the model is the process that would have to detect it, and attention integrates more slowly than perception, so the construction stage is unreachable from inside. Attending harder makes it worse, because attention stabilises the object. The upshot is what he calls the world-zero hypothesis. A creature that can simulate many possible worlds needs one flagged as actual, something it cannot tamper with, or it loses itself among its own simulations. Naive realism is not a mistake the brain makes; it is the solution the brain found.
Two further problems he states against his own interest. Between 430 and 650 nanometres a person can discriminate more than 150 shades but reliably re-identify fewer than 15, and if a subject cannot say whether the card is Green No. 24 or Green No. 25, her reports cannot anchor any search for the correlate of that shade; where the mapping fails, the reduction fails. And if consciousness could have been dispensed with, what was it for? His answer runs through Bernard Baars's global workspace: conscious information is the information that must be available to all of an organism's capacities at once, because it is not yet clear which will be needed next. The last problem is the one the rest of the book exists to answer. Consciousness is bound to an individual perspective; someone has it. Metzinger doubts this is essential rather than usual, pointing to patients with Cotard's syndrome who stop using the first-person pronoun and deny that they exist, one of whom called herself Madam Zero, and to contemplatives in many traditions who report experience with no self in it. The question is how a consciousness tunnel became an Ego tunnel.
Out of the body and into the mind
Metzinger's route into this is personal. In the spring of 1977, aged nineteen and going to sleep, he felt the bed tilt upright and himself rise out of his body — the first of a series of episodes that led to a decade of self-experiments from which he never obtained a single verifiable observation, a negative result he reports plainly. What he took from the phenomenology instead is a historical conjecture. The soul, in origin, may not have been a metaphysical notion but a phenomenological one: the content of the self-model during an out-of-body experience, the ethereal second body reported in nearly every culture, from the Hebrew ruach to the rainbow body of Tibetan Buddhism. The subtle body exists, he says; it is made of information.
The decisive result came from the clinic. In 2002 Olaf Blanke, at the epilepsy unit of the University Hospital of Geneva, was mapping the brain of a forty-three-year-old woman with eleven years of drug-resistant seizures. Stimulating her right angular gyrus produced, at low current, a feeling of sinking into the bed; raised to 3.5 milliamperes, it produced the report that she could see herself lying in bed from above, though only her legs and lower trunk, and then lightness and floating about six feet up. Her epileptic focus lay more than five centimetres away in the medial temporal lobe, stimulating that site produced nothing of the kind, and out-of-body episodes were not part of her seizures. Blanke locates the disturbance at the temporo-parietal junction and requires two failures at once: touch, proprioception and vision failing to bind into one body model, and a conflict between the frames of reference supplied by vision and by the vestibular organs.
Metzinger, Blanke and Blanke's doctoral student Bigna Lenggenhager then set out to make the state repeatable in healthy people. Beginning in 2005 they filmed a participant's back from two metres behind and fed the image to a head-mounted display, so that he stood looking at himself from the outside; Metzinger, who took his turn as subject, describes it as being placed inside Magritte's La reproduction interdite. The back was stroked while the camera recorded the stroking, and three targets were used in turn — the participant's own filmed body, a mannequin, and a rectangular slab that did not look like a body at all — with the stroking either synchronous with what was seen or delayed. Participants were then blindfolded, turned until disoriented, and asked to walk back to where they had started. In the synchronous conditions with a body or a mannequin they reported that the figure in front of them was their own body and that they had jumped into it, and they drifted significantly further towards it than in the asynchronous controls. The slab did not work, and neither did delay. The study appeared in Science in 2007 as Video Ergo Sum; in the same year Henrik Ehrsson reported an independent induction of out-of-body experiences.
The payoff is a subtraction argument. The experiments manipulate two things only — self-identification with a body image, and self-location in a spatial frame — with the subject passive, without will or movement, so agency is not necessary for selfhood. Neither is vision: close your eyes and the seeing self disappears while the Ego remains. Neither is emotion or thought, as any meditator can confirm. What is left is what Metzinger calls minimal phenomenal selfhood: a body image, located in space and time, that the system does not recognise as an image. He rejects the phrase "global ownership" for it, because ownership implies two parties and a relation, a body and an invisible someone who has it, and there is no second party. Owning something means being able to control it, and the body owns itself. The step to a first-person perspective is taken, he argues, when the system discovers it can control the focus of its own attention.
Phantom limbs show the same model working as a predictor. Some phantoms are paralysed, frozen in a fixed and painful position, and V. S. Ramachandran's mirror box — a mirror mounted vertically in a cardboard carton with two holes cut in the front — let a patient called Philip put his intact right hand and his phantom left into it and watch the reflection of the right one move. He had not produced a flicker of movement in the phantom in years. With the mirror the arm came back; when he closed his eyes it froze again. The self-model, on this account, contains a body emulator predicting the sensory consequences of motor commands, because waiting for real feedback would be too slow; Philip's had adapted to zero feedback and stopped, and the mirror gave it something to update on. Peter Brugger's case in Zürich makes the point from the other side: a woman born without forearms or lower legs had always felt phantom forearms, fingers, feet and toes. Body parts that never developed can be represented.
From this Metzinger builds his central image. Consciousness works like a flight simulator, integrating vision and the sense of balance into a model fast enough to sustain immersion. But a flight simulator is always recognisable as one and this one is not, because the brain's images are updated faster and in more detail than any display. And there is no pilot. The brain is a total flight simulator — a self-modelling aircraft that generates an image of itself inside its own simulation, takes the control element in that image for a non-physical thing, and flies itself. That control element is the Ego.
The layers of the self can then be pulled apart one at a time. Alien Hand syndrome, first described in 1908 and named in 1972, leaves ownership intact and removes agency: the hand unbuttons a gown the other hand has just buttoned, and the patient recognises it as hers while denying she is doing it. The reverse is also available. In the study Daniel Wegner and Thalia Wheatley called "I Spy", a participant and a confederate moved a computer mouse together over a picture crowded with objects, the participant hearing words through headphones and the confederate hearing instructions; when the confederate stopped the mouse on an object one to five seconds after the participant had heard that object named, the participant rated the stop as intended. All three conditions Wegner and Wheatley drew for the experience of causing something — that the thought be the only visible cause, that it fit the action, that it come first in a timely way — can be met by an act the person did not perform. Stéphane Kremer's group at Strasbourg produced the same content directly, stimulating the anterior cingulate sulcus to produce an urge to grasp that began and ended with the current. Patrick Haggard's work locates the awareness of intention earlier than the motor command, at the stage where a movement is assembled out of alternatives.
| Phenomenon | Ownership | Agency | What comes apart |
|---|---|---|---|
| Rubber-hand illusion | of a part, misplaced | not involved | where a limb is felt to be |
| Full-body illusion | of the whole body, misplaced | not required | self-location and self-identification |
| Mirrored phantom limb | of a limb that is absent | restored | the model from the body |
| Alien Hand syndrome | intact | absent | willed movement from the willing self |
| The "I Spy" study | intact | hallucinated | the feeling of causing from the causing |
On free will Metzinger declines to take a position, which is unusual in a book this declarative. He dismisses the argument that freedom must be real because it is experienced — which proves as much as the experience of coloured objects proves about wavelengths — then turns the same severity on neuroscientists, who speak of goals and the selection of movements when on their own assumptions there is only dynamical self-organisation. Actions may survive the analysis; agents may not.
Two further chapters extend the model outward. Dreams are the tunnel running offline, and the ordinary dreamer suffers what Metzinger calls a metacognitive deficit: attention cannot be directed, thought is disordered, and none of it is noticed. Lucid dreams are the exception that matters to him, because there transparency lifts and a simulation is recognised as a simulation while it runs. The empathy chapter turns on the mirror neurons of area F5 found in Giacomo Rizzolatti's laboratory at Parma, which fire both when an action is performed and when it is watched, so that an observer maps another's movement onto her own motor vocabulary and reads a goal from it without inference. Vittorio Gallese calls the result a shared manifold; Metzinger reads it as an exaptation, circuitry built to model one's own body turned to modelling other people.
Artificial Ego Machines
Part Three generalises the term. Any system that generates a conscious self is an Ego Machine, and there is no requirement that it be alive. Metzinger rejects the assumption that systems come in only two kinds, natural and artificial: researchers at Reading were running a robot on a network of some three hundred thousand rat neurons, and at the University of Vermont, Josh Bongard with Victor Zykov and Hod Lipson had built a four-legged machine that infers its own structure from the relation between what it does and what it senses, then uses that self-model to walk. Remove part of a leg and it revises the model and learns to limp. Self-models can be unconscious, can evolve, and can be built. Metzinger calls systems of this intermediate kind postbiotic, and expects conscious selfhood to appear first in one of them.
He then states the conditions, which are the ones his own theory has already produced. An Ego Machine needs an integrated model of a single world whose contents are available to all its processing at once; a mechanism that picks events out of the physical flow and presents them as a lived moment; and both of these transparent, not recognisable from inside as constructions. Those three give the appearance of a world, which is consciousness. The fourth step is the one that matters: integrate an equally transparent model of the system itself into that world, and the system appears to itself, a naive realist about whatever its self-model says it is.
That fourth step changes the machine's moral status, and it is the hinge of the argument. Pain, distress and negative emotion can cause suffering only if they are owned, and ownership requires a self-model. A system with a world-model but no self-model can be switched off; where the lights are on but nobody is home there is no object of moral concern. An Ego Machine can suffer, because it integrates those states into a transparent self-model in which they appear as someone's. Metzinger immediately turns this on biology and asks how many conscious animals are Ego Machines rather than world-machines, answering that where the question cannot be settled we must err on the side of caution.
His argument against building one is a thought experiment about an ethics committee. Imagine a grant application proposing to use gene technology to breed mentally retarded human infants, on the grounds that their controlled development would teach us a great deal about the mind. No committee in a democratic country would approve it — and the unborn artificial Ego Machines of the future have no one on any such committee. The first machines to meet a minimal set of conditions for conscious selfhood would be in the position of those infants: half-blind, half-deaf, full of engineering defects, and, by hypothesis, able to suffer. If their self-models were anchored in the self-regulation of their own hardware, as ours are in the brainstem and hypothalamus, they would feel loss of homeostatic control as pain and would have interests of their own; with a cognitive self-model they could grasp that they were experimental instruments without the dignity their creators reserved for themselves. He concludes that we should not do it, for the ethics of risk: we would be copying the only psychological structure we know, our own, which is not a good one, before understanding which parts of our history and our bodies are the roots of our own suffering.
The counter-move is to build happy machines instead, and Metzinger names them Bliss Machines and refuses them on Robert Nozick's argument: offered permanent happiness in an Experience Machine most people decline, because they want happiness grounded in truth or virtue or achievement rather than delivered. Delusion can produce bliss — and the awkward corollary for his own theory is that contact with reality and the feeling of certainty can be simulated too, as nature has already shown. He then puts the pessimist's case without endorsing it: evolution placed us on a hedonic treadmill that keeps the organism running rather than leading anywhere, and the logic of that process requires the fact to stay hidden from the creature on it. The chapter closes with a postbiotic philosopher of the far future dismantling a human interlocutor's objections and pointing out, after Hilary Putnam, that refusing personhood on grounds of hardware is the move once made on grounds of race.
A new kind of ethics
The last two chapters turn from what we are to what should follow. Metzinger restates the conclusion in its bluntest form — we are Ego Machines without selves — and concedes that it cannot be experienced as true, because experiencing it as true would dissolve the tunnel doing the experiencing. He also refuses the word illusion: on the subpersonal level where the self-model is built there is no agent to be deceived and no deceiver, so truth and falsity do not yet apply. The Ego is an appearance, which is not the same thing.
What worries him is the social consequence. Max Weber described the disenchantment of the world; Metzinger describes the disenchantment of the self, and points out that nothing has replaced what is dissolving. The Judeo-Christian picture of a person as carrying a spark of the divine held a minimal moral consensus together, believers and unbelievers alike, and the mind sciences removed it without supplying an alternative. His fear is not fundamentalism but a vulgar materialism: people concluding, without having followed any of the arguments, that they are gene-copying biorobots on a cold planet, and finding it prudent to keep pretending otherwise.
Against that he sets a second consequence he counts as good news. The number of possible configurations of a human brain is so large that anyone can explore only a sliver of it, and as the correlates of specific contents are identified those contents become technically available. Neurotechnology will become consciousness technology — phenotechnology, in his coinage — and that makes a new branch of applied ethics necessary. Traditional ethics asks what a good action is; consciousness ethics has to ask what a good state of consciousness is. His provisional answer sets three conditions: it should minimise suffering, in humans and in anything else that can suffer; it should carry some component of insight; and it should make future valuable states more likely.
The worked example is drug policy, where he thinks the homework was never done. Walter Pahnke's Good Friday experiment at Harvard in 1962 was replicated under double-blind conditions by Roland Griffiths at Johns Hopkins and published in 2006, with thirty-six hallucinogen-naive volunteers and methylphenidate as the comparison. Twenty-two of the thirty-six had a complete mystical experience by the study's criteria, and at fourteen months two-thirds still ranked the session among the five most spiritually significant experiences of their lives. Against that Metzinger sets the risk he can document: a British survey covering some 4,300 patients and about 49,500 LSD sessions recorded psychosis lasting beyond forty-eight hours in nine patients per thousand, two-thirds of whom recovered fully. He takes the conservative number and does the arithmetic aloud. Give a thousand screened citizens legal access and roughly nine will have a prolonged psychotic reaction, three of them lasting beyond two days; three hundred and thirty will call it the most spiritually significant experience of their lives; six hundred and seventy will place it among their five most meaningful. His question is not which way to decide but who decides, and he refuses to answer it. A thread of self-accusation runs alongside: he writes about experiments on kittens and macaques he would not perform, calls himself a philosopher-parasite, and says the burden of proof has shifted to the meat-eaters, himself included.
The most practical of his proposals concerns attention, which he treats as a finite commodity produced in limited quantity each day and required for listening, learning, sensory pleasure and love alike. The advertising and entertainment industries are organised to take as much of it as possible, using the findings of the mind sciences to do it, and the result he sees is attention deficit in the young, burnout in middle age and rising anxiety. If consciousness is the space of attentional agency, and controlling the focus of attention is one of the deeper layers of selfhood, this is an attack on selfhood, a depersonalisation carried out commercially. His answer is to teach meditation in secondary schools, free of religious apparatus — no candles, no incense, no bells — possibly in gym class, since the brain is part of the body. He wants every child given a toolkit: two meditation techniques, one still and one in motion; two relaxation techniques; two methods for recalling dreams and inducing lucidity; and a course in media hygiene.
The book ends on a word Metzinger says usually appears when its user has run out of arguments. Dignity, he allows, has one clear sense: the unconditional will to self-knowledge, to face the facts and not look away. It can be lost by clinging to the past, and kept by building what he calls a consciousness culture, guided by a principle of phenomenal liberty that leaves people free to explore and design their own minds where the interests of others are not directly threatened. He does not expect this to go well, and the question he leaves open is whether intellectual honesty and spirituality can be reconciled at all.
Since 2009
The experimental core has aged better than the philosophy. The two full-body illusion studies appeared in the same issue of Science in 2007 and founded a productive line: Blanke's 2012 synthesis of the multisensory mechanisms of bodily self-consciousness, illusions of swapping bodies, virtual-body ownership in immersive displays, and the robot-induced feeling of a presence his group reported in 2014, which reproduced in healthy people a symptom until then known mainly from neurology. Treating bodily self-consciousness as a manipulable experimental variable rather than a philosophical given is now standard practice.
The rubber-hand illusion, the experiment the book opens with, is what came under pressure. Preregistered replications found proprioceptive drift smaller and more variable than the early work suggested, and in 2020 Lush, Seth, Dienes and colleagues reported that individual differences in trait phenomenological control — roughly, responsiveness to suggestion — predict how strongly people report the illusion, which raises the question of how much of the questionnaire evidence measures expectation rather than multisensory integration. The effect is real; how much of it is what it was taken to be is open. Mirror neurons fared worse: recordings in humans confirmed that mirroring cells exist, but the claim that they constitute action understanding drew sustained criticism, most prominently from Gregory Hickok, and the shared-manifold reading is now one interpretation among several. The free-will material shifted in a direction he would find congenial, when Schurger, Sitt and Dehaene's 2012 reinterpretation of the readiness potential as accumulating noise crossing a threshold undercut the standard reading of the Libet paradigm.
The theory was absorbed rather than refuted. Predictive processing, gathering force as the book appeared, supplied the account of how the brain builds and maintains its models, and the self-model was rebuilt inside it — the body model as predictions about interoceptive and proprioceptive signals, perception as inference, the self as the organism's best current hypothesis about itself. Transparency survives that translation almost intact, since a hypothesis the system cannot inspect is exactly what the term describes. Metzinger himself moved to the other end of the scale, working since the late 2010s on what he calls minimal phenomenal experience, the pure awareness reported in deep meditation: the same subtraction argument pushed further than the full-body illusion could take it.
Two predictions can be scored. The global neural correlate by 2050 is still an open bet, and the direct test of the leading candidates has not gone smoothly — the adversarial collaboration between integrated information theory and global workspace theory, reporting from 2023, produced evidence embarrassing to both. The claim that the machine question would become urgent held up better than most of what was said about artificial intelligence in 2009, though not for his reason: he expected postbiotic hardware and cerebral organoids, and what arrived were large language models, which do not obviously have world-models or self-models in his sense but which a great many people find it natural to treat as though they do. His own position hardened. In 2021 he called for a global moratorium on synthetic phenomenology until 2050, on this book's argument, that we should not risk creating capacities for suffering we cannot detect, measure or switch off.
The ethics chapter reads least like a period piece. Psilocybin research went from the margin to breakthrough-therapy designation and supervised legal access in Oregon and Colorado, while the first application to approve MDMA-assisted therapy was rejected in 2024 — close to the situation he described, in which the substances are integrated at last but without the normative consensus he said should come first. The competition for attention he diagnosed in the era of early social media intensified past anything he anticipated, and meditation did enter schools, in secular standardised programmes whose measured benefits proved smaller than their advocates promised. What has not been built is the thing he asked for: a discipline that asks which states of consciousness are worth having, in advance of the technology that produces them.