The Predictive Mind

Here is a fact that sounds like a riddle: the signals racing back down your visual system, from the higher brain toward the eyes, vastly outnumber the signals coming up from the eyes themselves. If perception were simply the world pouring in, the traffic would run the other way. It does not — because the brain is not receiving the world so much as predicting it, and using the eyes mainly to check its work. This inversion, once you see it, rewires how you understand almost everything about the mind: how it builds a stable reality, how that reality can go wrong, and what happens when a molecule reaches in and turns the predictions down.

The brain in a dark box

Start with the predicament your brain is actually in. It is sealed inside the skull, in total darkness and silence, with no direct contact with the world. Everything it will ever know arrives as electrical spikes traveling up the nerves — and those spikes are ambiguous. The same pattern on the retina could be cast by a small object nearby or a large one far away; the same sound could be a threat or nothing. Raw sensation, by itself, does not tell you what caused it.

So the brain does the only thing it can: it infers. It builds an internal model of the most likely causes of its sensations and runs that model forward as a prediction. What you experience as the seen, heard, felt world is that model — the brain’s best guess about what is out there, continuously tested against the incoming data. The philosopher’s old image of perception as a passive photograph is exactly backwards. Perception is an active hypothesis. You are, in a real sense, hallucinating a world and correcting it with your senses.

Predictions down, errors up

In 1999, Rajesh Rao and Dana Ballard gave this idea a concrete neural shape that still anchors the field. In their model of the visual cortex, the connections running down the hierarchy — from higher, more abstract areas to lower, more sensory ones — carry predictions: the higher level tells the lower level what it expects to see. The connections running up carry only the prediction error — the residue, the part of the signal the prediction failed to explain (Rao & Ballard, 1999).

The elegance is in what gets transmitted. If the world matches the prediction, there is nothing to report; the error is zero and the higher levels rest content. Only surprise travels upward. This is astonishingly efficient — the brain spends its energy encoding what it did not already expect — and it explained puzzling details of real neurons, like why a cell that fires to a short line can fall silent when the line is extended, as if a longer line were simply less surprising. The cortex, in this view, is a hierarchy of predictions, each layer trying to explain away the layer below.

The free-energy principle

Karl Friston took this local idea and pushed it to its most ambitious limit. His free-energy principle proposes that a single imperative governs the entire brain — indeed any self-organizing system that resists dissolving into its environment: minimize surprise. Over the long run, a creature must keep its sensory inputs within the narrow range it expects — a fish that finds itself out of water is in a state of catastrophic prediction error — and the brain does this by constantly reducing the gap between what it predicts and what it senses (Friston, 2010).

There are only two ways to reduce that gap. You can update the model to fit the world — that is perception and learning. Or you can act on the world to make it fit the model — that is action; you predict your hand is on the cup, and the prediction error drives your arm until it is true. Perception and action become two faces of one process: the endless minimization of prediction error. It is a strange, powerful lens, and whatever its ultimate fate as physics, it reframed the brain as fundamentally an inference engine.

Precision: the volume knob on belief

One piece makes the whole machine work — and it is the piece psychedelics act on. Not all predictions and not all errors are treated as equally trustworthy. The brain weights them by their expected reliability, a quantity called precision. A confident, high-precision prior can override noisy sensory data; a high-precision error can force even a cherished belief to yield. Precision is the brain’s way of deciding, moment to moment, whether to trust the top-down story or the bottom-up evidence.

Think of a prior as a belief and precision as how loudly it is asserted. When high-level priors are held with great precision, they dominate: the brain sees what it expects, discounts what contradicts it, and the world snaps neatly into the model. This is usually good — it is why you read messy handwriting effortlessly and recognize a friend in bad light. But turned up too far, precise priors become a prison. The model stops listening to the evidence. And that, the REBUS theorists argue, is a surprisingly good description of what goes wrong in a suffering mind.

REBUS: relaxed beliefs under psychedelics

In 2019, Robin Carhart-Harris and Karl Friston brought these threads together into the most influential mechanistic account of psychedelics to date: REBUS, for “RElaxed Beliefs Under pSychedelics,” laid out in a paper titled REBUS and the Anarchic Brain (Carhart-Harris & Friston, 2019). Its central claim is a single sentence: psychedelics relax the precision of high-level priors.

Follow the consequences. If the brain’s confident top-down predictions suddenly lose their grip — their precision turned down — then the bottom-up prediction errors they normally suppress are liberated. Information that was being explained away, discounted, filtered out, now flows up through the hierarchy unchecked. The high-level model, no longer able to force the data into its expectations, becomes fluid and revisable. This is why the theory calls it the “anarchic” brain: the usual top-down authority is relaxed, and the system runs hotter, freer, and less constrained — the same state the entropic brain hypothesis describes as higher-entropy, and which REBUS explicitly unifies with the free-energy framework.

Why this feels like something

Predictive coding is not just an abstraction on a whiteboard — it maps onto the texture of the psychedelic experience with uncanny precision. When high-level priors relax and raw error floods upward, ordinary things stop being predictable and become charged. A familiar face, a patch of carpet, a piece of music: each carries more information than the model can quietly absorb, and unabsorbed prediction error is, subjectively, the feeling that something is deeply significant. This is one reason psychedelic experiences so often feel meaningful — the brain is generating surprise it cannot explain away.

The same logic runs down to raw perception. The brain normally imposes strong priors on visual noise, smoothing it into blank walls and steady surfaces. Relax those priors and the noise is no longer suppressed — it self-organizes into the lattices, spirals, and tunnels of geometric hallucination. And it runs all the way up to the self. The sense of being a bounded, continuous “I” is itself a very high-level prior — perhaps the highest. Relax that one, and you get the loosening, dissolving, and boundary-melting of ego dissolution. One mechanism — relaxed precision — threads through the whole strange landscape.

The clinical promise: unsticking a stuck model

Here is where REBUS becomes more than a beautiful theory. Carhart-Harris and Friston propose that many forms of psychological suffering are, at bottom, pathologically overweighted priors — beliefs held with such crushing precision that no evidence can dislodge them. The depressed brain predicts worthlessness and futility and then filters the world to confirm it. The anxious brain predicts threat and reads every ambiguous signal as danger. The addicted brain predicts relief from the substance with a certainty that drowns out every contrary fact. In each case a high-level model has become too confident, too rigid, deaf to the bottom-up evidence that might update it.

If that is the disease, then relaxing the precision of those priors is a plausible route to a cure. A psychedelic, in this frame, does not implant a new belief — it temporarily loosens the old one, making the whole model briefly plastic and revisable at exactly the moment a skilled therapist, a new perspective, or a powerful experience can help it re-form in a healthier shape. It reframes the afterglow not as a lingering drug effect but as a window of revisable belief — which is precisely why psychedelic therapy places such weight on set, setting, and integration. The molecule opens the model; the work reshapes it. This dovetails with everything we know about the plastic, growth-primed brain the days after a session.

REBUS and its rivals

Science advances by argument, and REBUS has a serious challenger worth naming. A competing model — sometimes called ALBUS, or the “altered” rather than simply “relaxed” beliefs view — argues that psychedelics do not only relax high-level priors. In some circumstances they may strengthen certain predictions, which could explain why psychedelic insights can feel so unshakably true, occasionally to the point of implanting convictions that are not warranted. The relationship between top-down and bottom-up under psychedelics may be more layered than a single dial being turned down. This is a live scientific debate, not a settled matter, and honest neuroscience holds REBUS as a leading framework rather than a proven law.

It also connects to the rest of the psychedelic brain rather than replacing it. A relaxed high-level prior is the computational description; a loosened thalamic gate and a disintegrated default mode network are among the physical events that may implement it. These are different levels of the same story — algorithm and hardware — not competing claims.

The honest cautions

Predictive coding and the free-energy principle are powerful frameworks, but they are frameworks — sweeping, still-debated, and hard to falsify in their grandest forms. REBUS inherits both their reach and their uncertainty. The human evidence for “relaxed priors” is real but indirect, drawn largely from imaging studies and clever perceptual experiments rather than a direct readout of precision in the brain. Thoughtful researchers disagree about whether relaxation is the whole story, and the ALBUS critique should be taken seriously.

The clinical picture demands the same restraint. That a psychedelic can render beliefs plastic is not automatically therapeutic — a mind whose model has come unstuck is suggestible and fragile as well as open, which is exactly why context and support are not optional extras but part of the mechanism. An opened model can be reshaped for better or for worse. The theory that explains the healing also explains the risk.

The guess you live inside

Return, at the end, to the ordinary miracle you began with. Right now your brain is predicting the world — the page, the light, the weight of your body — and showing you the prediction as if it were simply reality. Most of the time the guess is so good, so seamless, that you never suspect it is a guess at all. Predictive coding is the claim that this is the deepest thing the brain does; the free-energy principle is the claim that it is all the brain does; and REBUS is the claim that a psychedelic works by loosening the guess — letting the world back in, unexplained and unedited, so that a model grown too sure of itself can, for a few hours, be written again.

OOTW Journal is educational and does not provide medical advice. Predictive coding, the free-energy principle, and REBUS are scientific models under active debate, not settled facts, and the human evidence is largely indirect. Psychedelics are controlled substances, are not safe for everyone — including people with a personal or family history of psychosis or bipolar disorder, or those on serotonergic medications — and nothing here is a recommendation to use them.