What Is This?
Predictive coding is a brain theory built around one clean idea:
the brain does not passively record the world; it constantly predicts it, then updates from prediction error
The brain sends top-down predictions about what it expects. Incoming sensory evidence pushes back. Perception, action, attention, and learning are partly the process of reducing the mismatch.
That is powerful. It is also dangerous.
A powerful framework can become a universal re-description machine. It can make old claims sound deeper without adding testable mechanism.
The useful question is:
when is predictive coding a mechanism, and when is it only a metaphor?
The 2025 Frontiers paper, "Predictive coding in psychopathology: mechanistic model or metaphorical re-description?", is valuable because it asks that question directly.
Why Does It Matter?
Predictive coding is now used to explain perception, attention, autism, psychosis, anxiety, depression, pain, placebo, interoception, and decision-making.
That breadth is either a strength or a warning.
A good framework should generalize. But if it explains everything after the fact, it may explain too little before the fact.
The reusable model for Jamie is:
framework quality = compression + mechanism + prediction + falsifiability
Predictive coding has compression. It can describe many mental phenomena with one vocabulary: priors, prediction errors, precision, updating, inference.
The test is whether it also supplies mechanism and prediction.
The Core Model
Predictive coding has three important pieces.
1. Priors
A prior is the system's expectation before new evidence arrives.
In human terms:
what the brain expects to be true before the world corrects it
Priors are not only beliefs in the explicit sense. They can be perceptual, bodily, emotional, and action-oriented.
2. Prediction error
Prediction error is the mismatch between expectation and incoming evidence.
prediction error = what happened - what was expected
The system can reduce prediction error by updating its model, changing attention, or acting on the world.
3. Precision weighting
Precision is confidence or reliability weighting.
Not all errors should update the model equally. Some signals are noise. Some are trustworthy. Precision weighting decides which errors matter.
This is the part that makes predictive coding useful for psychopathology. Many proposed accounts of mental disorder can be framed as problems of prior strength, error weighting, or model updating.
The Psychopathology Promise
Predictive-coding accounts can make mental disorders feel less mysterious.
Examples:
- Psychosis: overly strong or wrongly weighted prediction errors may make irrelevant stimuli feel meaningful.
- Anxiety: threat priors may dominate ambiguous evidence.
- Depression: negative priors about self, future, or agency may become sticky.
- Autism: differences in precision weighting may alter how sensory prediction errors are handled.
- Chronic pain: bodily predictions and interoceptive signals may reinforce persistent pain experience.
This is not useless. It can generate better questions.
Instead of asking:
what symptom does this person have?
it asks:
which expectation, error signal, or precision weight is failing to update?
That can be a real mechanism.
The Metaphor Trap
The trap is that the same vocabulary can be attached to almost anything.
A weak predictive-coding explanation sounds like this:
this disorder involves bad predictions or abnormal precision
That may be true, but it is not yet explanatory. It has not said:
- which prior;
- which level of the hierarchy;
- which error signal;
- how precision is measured;
- what changes under treatment;
- what prediction differs from competing theories.
The Frontiers paper's useful warning is that predictive coding can become a metaphorical re-description: translating symptoms into predictive-processing language without adding empirical constraint.
Mechanism Test
Use this checklist.
A predictive-coding account is doing real work when it can answer:
- What exactly is being predicted?
- Where is the prediction error measured?
- What carries precision?
- What evidence would change the model?
- What does this account predict that another account does not?
- What intervention should work if the account is right?
- What would falsify it?
If those questions cannot be answered, the framework may still be useful as a lens, but not as a mechanism.
Why Precision Weighting Matters
Precision weighting is the bridge between elegant theory and messy psychology.
A person may receive evidence that contradicts a fear, a belief, or a bodily expectation. But if the system assigns low precision to the corrective evidence and high precision to the old prior, the model will not update much.
That gives a clean way to think about sticky beliefs and symptoms:
change requires not only new evidence, but evidence that the system treats as reliable
This has obvious links to therapy, training, coaching, and learning. Telling someone the truth is not the same as changing the precision-weighted update their system performs.
How To Use This Outside Psychology
Predictive coding is also a model for learning and decision-making.
Jamie can use it as a question pattern:
1. What did I expect?
2. What happened?
3. What error did that create?
4. Did I update, ignore, or overreact to the error?
5. Was the error actually reliable?
This is useful for training feedback, business experiments, AI-agent evaluations, market signals, novelty-versus-distraction checks, and deciding when a failure is noise versus a model update.
The key is precision. Not every surprise deserves a strategy change.
Why Smart People Get This Wrong
They love grand unifying theories
Predictive coding is elegant. Elegance can make weak evidence feel stronger.
They confuse translation with explanation
Putting symptoms into new vocabulary does not automatically explain them.
They skip measurement
If priors, prediction errors, and precision are not measured or constrained, the model floats.
They forget rival explanations
A framework is stronger when it beats alternatives, not when it merely accommodates the same facts.
They overgeneralize from brain theory to life advice
Predictive coding can inform decision-making, but that does not make every intuition or habit a formally modelled prediction error.
What This Does Not Prove
This article does not prove that predictive coding is false or that it is only metaphor.
It proves a narrower discipline:
predictive coding is valuable when it creates constrained, testable explanations; it is weak when it only renames phenomena
Limits:
- psychopathology is heterogeneous;
- many predictive-coding accounts remain high-level;
- measurement of precision and hierarchical prediction is difficult;
- different disorders may share vocabulary but not mechanisms;
- clinical usefulness depends on whether the model improves intervention, prediction, or patient outcomes.
Practical Takeaways For Jamie
- Use predictive coding as a mechanism test. Ask what prior, error, and precision weight are actually involved.
- Do not reward pretty re-description. Better vocabulary is not better evidence.
- Precision is the useful primitive. Learning depends on whether the system treats evidence as reliable.
- Apply it to experiments. A surprise should update the model only if the signal is precise enough.
- Research-library fit: this is a durable epistemology model: mechanism beats metaphor.
Key Terms
- Predictive coding: theory in which the brain predicts sensory input and updates through prediction error.
- Prior: expectation before new evidence.
- Prediction error: mismatch between expected and observed input.
- Precision: confidence or reliability weighting assigned to a signal.
- Active inference: related framework where perception and action reduce uncertainty / prediction error.
- Psychopathology: study of mental disorders and abnormal experience or behaviour.
- Metaphorical re-description: explaining something by renaming it in a framework without adding testable mechanism.
Recall Questions
- What are the three core pieces of predictive coding?
- Why is precision weighting so important?
- What makes a predictive-coding account mechanistic rather than metaphorical?
- How can predictive coding explain sticky beliefs or symptoms without overclaiming?
- How can Jamie use the model when interpreting business, training, or AI-agent feedback?
Best Resources To Learn More
- Start with the Frontiers paper for the mechanism-versus-metaphor critique.
- Read Andy Clark for the broader predictive-processing frame.
- Read Karl Friston for the free-energy / active-inference lineage.
- Use review papers on autism, psychosis, pain, and anxiety only after checking whether they make testable predictions.
Sources
- "Predictive coding in psychopathology: mechanistic model or metaphorical re-description?" Frontiers in Human Neuroscience (2025). DOI: 10.3389/fnhum.2025.1743028. https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2025.1743028/full
- Andy Clark, Surfing Uncertainty: Prediction, Action, and the Embodied Mind (Oxford University Press, 2016).
- Karl Friston, "The free-energy principle: a unified brain theory?" Nature Reviews Neuroscience 11, 127-138 (2010). DOI: 10.1038/nrn2787.