A trapped prior is a belief that has gotten strong enough to consume its own evidence. New observations should weaken it; instead they feed it. The phobic person flooded with friendly dogs grows more terrified, not less. The partisan handed a clean fact about the other side reads it as further proof of malice. The mechanism is not willful denial. It is that the prior reaches into the raw experience itself and rewrites what arrived before the conscious mind gets a chance to evaluate anything.
Simple Picture
ELI5: every perception is a fight between what your brain expected and what your senses report. Most of the time the fight is fair and you experience something close to what is actually there. When one side wins by a wide enough margin, it stops being a fight at all — the expectation just becomes the experience. You can no longer see a friendly dog as friendly, because by the time the image of the dog reaches conscious awareness it has already been repainted in the colors of every previous fear. The dog you experience is not the dog in front of you. It is your prior wearing a dog costume.
And once that happens, every new encounter becomes more evidence that dogs are terrifying. The system has eaten its own escape hatch.
The Mechanism
Trapped priors are a degenerate case of predictive processing. The brain combines a top-down prior with a bottom-up sensory channel through a weighting algorithm. The output is what you experience as reality. Update happens when bottom-up surprise is large enough to force the prior to revise.
A trapped prior is what happens when the weighting algorithm assigns the bottom-up channel a coefficient near zero. The prior wins every match. The “experience” you have on encountering counter-evidence is not a clean perception of the counter-evidence — it is the prior re-rendered with the new stimulus as a prop. Then that experience — already shaped by the prior — gets fed back as the new datum. So the loop is:
prior → distorts perception → distorted perception → reinforces prior
Each cycle tightens. There is no leakage point at which the truth could enter, because every input channel is being filtered by the very thing it would need to update.
This is the neural account of why more information makes a trapped person wronger, not righter. Studies that find more scientifically literate people hold more polarized views are not paradoxes. They are exactly what the model predicts: information is processed top-down, so a stronger prior turns each new fact into a sharper instance of itself. The reader is not failing to update. The reader is updating in the wrong direction at the moment the prior takes the data.
Why Strong Emotion Traps Priors
The weighting algorithm is not fixed. It is modulated, and the modulation tracks something like emotional intensity. The brain narrows bottom-up bandwidth in proportion to threat. This is presumably adaptive — a prey animal that reprocesses every rustle as raw data dies the first time the rustle is real. Trusting the prior is metabolically and evolutionarily cheaper than trusting a noisy world.
But the same shortcut creates the trap. Trauma victims describing the moment of a violent event in oddly short, undetailed sentences are not being terse — the perceptual bandwidth was throttled at the time, and the memory has the resolution it had on intake. The window of tolerance names this clinically: outside the window, the prefrontal mediation that would let bottom-up signal compete with the prior goes offline. You are inside the prior. You cannot get out by thinking, because the thinking apparatus has been suspended.
This is why you cannot reason a phobic person out of a phobia, an abuse survivor out of a flinch, or a tribal partisan out of contempt by handing them facts. The facts arrive through a channel that has been narrowed to almost nothing. The argument lands inside an already-decided perception.
The Bitch-Eating-Crackers Layer
There is a stage of escalating dislike where the disliked person can do nothing without it counting against them. Eating crackers becomes evidence. Saying nothing becomes evidence. Saying something neutral becomes evidence with a hidden meaning. This is not exaggeration — it is exactly what trapped priors predict. The prior on malice has gotten strong enough that any input renders as further malice. Innocuous behavior is not perceived as innocuous and then explained away. It is perceived as sinister, with the same immediacy that a friendly dog is perceived as a threat.
The dog-whistle problem inherits this structure. Once a group is encoded as adversarial, every utterance from them becomes a coded version of the worst utterance you can imagine. The structure is unfalsifiable not because the believer is acting in bad faith but because falsification requires bottom-up signal, and the channel is closed. Paradigm lock-in is the same dynamic at the scale of a scientific community: the framework is the prior, anomalies are smoothed, and the machinery for revision is the very machinery the framework has captured.
Trapped Priors as the Common Form
Treating trapped priors as a single mechanism dissolves a lot of apparently-separate problems into one diagram. Many of the garden’s repeating motifs are this:
- Phobia and PTSD — an emotional prior on threat, weighting bottom-up to zero, recruiting every encounter as confirmation. The body keeps the score because the prior was set in the body and the body’s perceptual filters never got the memo that the threat is over.
- Polarization — a prior on the outgroup’s villainy, eating each piece of news as fresh proof, sharpening with information rather than with ignorance.
- Depressive realism — a prior of “nothing I do matters” rigid enough to be accurate when the world happens to comply and rigid enough to miss when the world doesn’t. The accuracy in zero-control conditions is a stuck dial coinciding with a true reading, not a measurement.
- The expert’s loneliness — a prior built from years of accurate prediction, now strong enough that the bottom-up signal capable of revising it cannot generate enough surprise. Expertise is a trapped prior that happened to be built on real ground.
- Local optima — the configuration that minimizes prediction error in the current basin. Any move out of the basin spikes error before it lowers it, so the system rejects the move as bad data rather than as the cost of relocation.
- Wound-driven relationships — a prior on chaos, abandonment, or insufficiency, expecting it from every partner and shaping every interaction until the partner’s behavior matches. The misery is the confirmation the system was after.
What looks like six different pathologies is one architecture under different load conditions.
The Difference Between a Strong Prior and a Trapped One
Not every confident belief is trapped. The diagnostic is what happens at the boundary. A strong prior is one with a high weight that will still update under sufficient surprise. A trapped prior is one whose weight on the bottom-up channel has collapsed to ~0, so no amount of surprise reaches the update step. The difference is structural, not a matter of degree of belief.
A useful test: if you cannot describe the observation that would change your mind in concrete detail, the prior is probably trapped, not just confident. Confidence has a falsification condition it can articulate. Trappedness has rationalizations for every imaginable input.
Escape Routes
If the trap is the perceptual channel, the escape has to be at the level of the channel. Argument, evidence, and exhortation all assume an open channel and so cannot help. What can:
Graduated exposure with regulation. Standard exposure therapy fails alone because exposure to the feared stimulus, with the prior intact, produces another distorted experience that strengthens the prior. The protocol that works pairs exposure with bottom-up regulation — breath, grounding, the felt sense, a calm therapist’s nervous system co-regulating yours — wide enough that some bottom-up signal reaches awareness intact. The dog has to be experienced as a dog, not as a prior-rendered dog, even briefly. One clean perception leaks the loop.
Mechanistic reasoning. Asking someone to explain how their belief actually produces the consequences they expect (rather than asserting the belief) is one of the few interventions shown to reduce political extremism. The mechanism is interesting: trying to articulate a chain of cause-and-effect requires temporarily holding the prior aside and looking at the world’s actual joints. The bottom-up channel widens by force.
Pharmacological prior loosening. Psychedelics — especially classical serotonergic agonists — appear to lower the precision weighting on top-down models, which lets accumulated bottom-up evidence update the prior all at once. The clinical data on PTSD, treatment-resistant depression, and end-of-life anxiety is consistent with the mechanism: a single high-bandwidth window in which the prior is too soft to filter, and twenty years of un-integrated counter-evidence gets to land at once. Annealing is the same family of move at lower temperature — energy in the system high enough to let the model recrystallize.
Practices that widen the bottom-up channel. Meditation, focusing, sensory work, somatic therapy, and contact with the felt sense all do the same thing under different brand names: privilege the bottom-up signal long enough that the prior cannot eat it on arrival. None of them work by argument. All of them work by giving the perceptual channel back.
The shape these have in common is not “convince the brain” but “force a window in which the prior loses its grip on perception.” The prior cannot be dislodged from inside its own filter.
Common Misread
The dimwit take is “they’re just being stubborn — they can see the evidence.”
The midwit take is “humans are biased; we should all just be more rational and consider the other side.”
The better take is that trapped priors are not a failure of rationality applied to perception — they are a malfunction of perception itself. The conscious “rational mind” never sees the data the prior already ate. Telling a phobic person to “just look at the dog” is asking them to use a tool that the prior controls. The fix is not better thinking. It is widening the channel through which thinking gets its raw material.
This also reframes what it means to disagree with someone. If you believe their position is held inside a trapped prior, no argument you make will land — and the move is not to argue louder but to ask whether there is a way to widen the perceptual channel itself, usually by lowering the emotional stakes enough that bottom-up signal can compete. Most fights fail this test before they begin: the fight itself raises the threat signal, narrows the channel, and traps both priors more firmly. You cannot reason someone out of a position you helped trap them into.
Main Payoff
A unified diagram for a class of problems that usually get filed apart. Mental illness, political polarization, expert blindness, persistent grievance, stuck relationships, and the way arguments fail to change minds are all the same shape: a prior with the bottom-up channel pinched to nothing, eating evidence as confirmation, getting stronger with every encounter that should have weakened it.
The diagram suggests the right level of intervention. Not better arguments. Not more information. An open channel is a prerequisite for any belief revision, and trapped priors are exactly the configuration in which the channel is not open. Everything else follows.
The author of the original framework noted, only half-jokingly, that his own prior on the theory had become trapped — every new domain he applied it to confirmed it. That is the right note to end on. The model is powerful enough to explain its own adoption. The fact that it would predict its own runaway success is not a knock on it. It is the model performing the move it describes, on the person describing it, in real time.
References:
- Scott Alexander, Trapped Priors As A Basic Problem Of Rationality, Astral Codex Ten
- Andy Clark, Surfing Uncertainty
- Sloman & Fernbach, The Knowledge Illusion — mechanistic reasoning as a depolarizing intervention