The brain is a combination of top-down prediction reconciling with bottom-up sensory information. It surfs uncertainty in a controlled hallucination to generate reality. Minor perturbations are smoothed for coherence in the middle layers. Only when the error is large enough to trigger the surprise alarm does it cause a re-prediction.
Simple Picture
ELI5: you are not seeing the world. You are dreaming the world and checking the dream against what your senses report. Most of the time the dream is close enough and you never notice. When the dream and the senses disagree violently, you get surprised — and that surprise is the only moment you are actually perceiving rather than hallucinating.
The Core Model
The brain maintains a hierarchical model of the world. Each level generates predictions about what the level below will report. When the prediction matches the input, nothing happens — the prediction is the perception. When there is a mismatch, a prediction error propagates upward, and the model updates. The dual-aspect architecture this rests on is older than neuroscience: Spinoza’s parallelism already insisted that mind and body are not two substances passing messages but two readouts of one underlying process — the prediction model and the brain-state are the same event under different attributes, which is why the framework slots so cleanly into the modern computational vocabulary without anyone having to argue for the metaphysics.
This means most of conscious experience is generated top-down. You see what you expect to see. You hear what you expect to hear. The consciousness-as-ground frame pushes this further: the colors, sounds, and textures you experience are not properties of the external world but qualities created in consciousness — the prediction is not just a model of reality but the only reality you ever access. The bottom-up sensory stream is not “raw data” but a correction signal — it only carries the difference between what the model predicted and what actually arrived.
This single framework reframes most of the garden:
paradigm-lock-in is the prediction machine refusing to update. The paradigm is the top-down model, and evidence that contradicts it generates prediction errors that the system smooths away as noise — unless the errors accumulate past the surprise threshold. Semmelweis’s data was prediction error that the prevailing model suppressed. Hyper-distilled symbols are hyperpriors of this kind operating at civilizational scale — the cross, 道, ॐ have been compressed through so many minds over so many centuries that they generate perception rather than being updated by it, and bottom-up reality (historical context, textual criticism) gets smoothed away as noise. The Manufactured Normalcy Field is this same mechanism scaled to civilization: culture generates top-down predictions about “the present” and smooths away the future as noise, translating every novelty into a metaphor for something familiar.
Cached thoughts are predictions that have never been checked against bottom-up reality. The system treats them as confirmed because no prediction error has propagated — not because they are true, but because the environment never forced a confrontation.
focusing is the deliberate act of attending to the bottom-up signal — the body’s sensory report — rather than the top-down narrative. The felt sense carries prediction errors that the conscious model has been suppressing. Sitting with the felt sense is allowing those errors to propagate.
Buddhist enlightenment is the prediction machine recognizing that its most expensive prediction — the existence of a predictor — is unnecessary. Suffering is the metabolic cost of maintaining the self-simulation against the flow of reality. The three puzzles (impermanence, dependent origination, no-self) are three angles on crashing the simulator. The mechanism is precise: just as our priors snap “PARIS IN THE THE SPRINGTIME” to a coherent picture with only one “the,” the brain snaps the vibrating, arising-and-passing sensory flux into a permanent stable image of the world. Samatha meditation is concentrating on that image hard enough, without preconceptions, to catch the extra “the” — to perceive the gaps that priors are smoothing away. The subjective experience of an observer may itself be just a very strong hyperprior on thought-and-emotion-related sense data.
Dopamine as Precision Signal
The standard model — dopamine as reward, serotonin as calmness — is oversimplified. Dopamine mostly carries something related to precision, confidence intervals, and surprisal levels. It does not signal “this is good” but “pay attention to this — the prediction error here matters.”
In Parkinson’s, low dopamine creates high uncertainty in motor movement — hesitant, shaky motion. The system cannot distinguish signal from noise in its own motor predictions. The existing dopamine note in the garden focuses on anticipation; predictive processing adds the deeper function: dopamine modulates how seriously the brain takes its own prediction errors.
Autism and Schizophrenia as Prediction Failures
Autism: high reliance on bottom-up information creating constant surprisal. Tags on shirts do not coalesce into a stable non-changing model — they continue to generate new information and a sense of annoyance. Weak priors mean the world never settles into a comfortable hallucination. Everything remains raw.
This extends autism-and-dimensionality with a mechanism: the autistic brain processes more dimensions not because it chooses to but because its prediction model does not compress. What neurotypical brains smooth away as “already predicted,” the autistic brain continues to register as novel. High IQ correlating with autism makes sense in this frame: increased mental precision means increased prediction error detection.
Schizophrenia: extremely strong top-down models that override all bottom-up information. The predictions become so delusional that sensory reality is ignored entirely. The person attributes personal agency to external forces (aliens, government) because their model says those forces exist and the bottom-up evidence cannot break through. They can tickle themselves — because their prediction model does not attenuate their own motor output the way a healthy brain does.
Anti-psychotics antagonize dopamine, which causes the brain to treat information as noise rather than signal — damping down the precision weighting on prediction errors.
Practical Implications
You cannot tickle yourself because your prediction model attenuates the signal from your own touch. By the same principle, your own exerted force is always perceived as less than the force you receive. This explains escalating violence: each person perceives the other as hitting harder, because their prediction model discounts their own blow.
The inner game maps directly: Self 1 is the top-down prediction model. Self 2 is the bottom-up motor system. When Self 1 over-predicts (“bend knees more,” “follow through”), it interferes with Self 2’s natural error-correction. The zone is what happens when the top-down model loosens its grip and lets bottom-up signals drive motor behavior directly.
The brain would rather be right and miserable than wrong and happy. A nervous system trained on chaos expects chaos. A calm environment generates massive prediction error — the world is not matching the model — and the brain’s cheapest move is to reject the data rather than update the model. It sabotages the peace, manufactures a crisis, or selects an environment that confirms the existing predictions. Being miserable but correct is neurologically cheaper than being happy but confused. This is why wound-driven relationships persist: the anxiety is not a feeling the brain is trying to escape but a prediction it is trying to confirm. It is also why local optima are so sticky — the current configuration keeps prediction error low, and any move toward a better state temporarily spikes it. Depressive realism is the same principle in clinical dress: the depressed brain runs a “nothing I do matters” prior so rigidly that it under-updates even when control is genuinely available — accuracy in zero-control conditions is a coincidence with a stuck dial, not a measurement.
Trauma is a prediction model stuck on “danger.” The system predicts threat in every ambiguous situation, and ambiguous situations are processed top-down as confirmations of the threat model. The bottom-up signal (this room is safe, this person is kind) cannot override the prediction because the precision weighting on the threat model is maxed out. The body’s alarm fires constantly — not because the world is dangerous but because the model says it is and the model generates reality.
Motor visualization works because the brain controls movement by strongly predicting that an action has happened and letting the lower levels’ desire to minimize prediction error do the work. Rumination produces choking by the same mechanism: the top-down prediction of failure generates motor behavior consistent with failing.
Common Misread
The dimwit take is “reality is fake — everything is just in your head.”
The midwit take is “this is interesting neuroscience but has no practical implications.”
The better take is that predictive processing explains why smart people get stuck (strong priors resist update), why trauma persists despite safety (the model generates the danger), why expertise can become a trap (the prediction model compresses away the details that matter), and why beginners sometimes see what experts miss (weak priors let more bottom-up signal through). The controlled hallucination is not a defect. It is how an organism survives in a world with too much information. The failure mode is when the hallucination becomes more real than reality — and that failure mode has a name in every domain this garden covers.
Main Payoff
The context vortex is a prediction model that has become self-confirming: new information enters but gets smoothed into the existing predictions. Exploration is the process of deliberately weakening priors so that bottom-up signal can propagate. The cultural immune system is a collective prediction model that rejects novelty to preserve coherence. The prig is a person whose precision weighting on moral predictions is so high that no amount of bottom-up reality can generate a correction signal.
The brain does not show you the world. It shows you a dream of the world that is just accurate enough to keep you alive. The question is always: how much prediction error are you willing to tolerate before you update?
References:
- Andy Clark, Surfing Uncertainty: Prediction, Action, and the Embodied Mind
- Scott Alexander, Book Review: Surfing Uncertainty, Slate Star Codex