You believe the café closes at 8 p.m.
You arrive at 7:30.
The lights are off.
A sign says:
New hours: 7 a.m.–7 p.m.
What should happen inside your head?
You should not conclude that cafés no longer exist.
You should not erase every memory of this place.
You should revise one part of the model:
closing time changed.
This small act contains one of cognition’s most important abilities.
We must change enough to fit the evidence without destroying everything that still works.
Quick Read
Revision is the process of changing a belief, representation, rule or model in response to new evidence, error, context or contradiction.
Good revision has two responsibilities:
- change what no longer fits;
- preserve what remains supported.
This article is about cognitive and model revision, not school “revision” in the sense of reviewing material for examinations. Those are related only indirectly.
One-sentence answer: Revision is the controlled updating of a representation so it fits new evidence better while preserving as much valid structure as possible.
A Mind That Cannot Revise Cannot Learn
Learning is not only adding facts.
Sometimes old knowledge remains correct and new information simply extends it.
Sometimes new evidence changes the model.
A child thinks all flying animals are birds.
Then encounters bats.
Something must change.
The concept “bird” becomes narrower.
The category “flying animal” becomes broader.
The old representation is not simply deleted.
It is reorganised.
Revision Is Not Forgetting
Forget a password and the representation disappears or becomes inaccessible.
Revise a belief and the representation changes because new information makes another form more appropriate.
Forgetting reduces access.
Revision changes structure.
Revision Is Not Contradiction
Contradictory information arrives.
That does not guarantee the mind will revise.
A learner may:
- ignore the contradiction,
- reinterpret it,
- treat it as an exception,
- reject the source,
- memorise the new answer without changing the old concept,
- build a genuinely new model.
A major review of how science learners respond to contradicting information found that anomaly-induced cognitive conflict is often inefficient when presented in isolation. Contradiction helps more when learners can process it meaningfully and have a viable alternative conception available.
Revision needs somewhere to go.
The Replacement Problem
Tell a student:
Your idea is wrong.
Now what?
If the old model is removed without a better replacement, the learner may return to it later because it remains more usable.
A revised model must be not only more accurate.
It must become cognitively available enough to use.
Belief Updating Is Only One Part of Revision
Joseph Sommer, Julien Musolino and Pernille Hemmer’s 2024 Psychological Review paper Updating, Evidence Evaluation, and Operator Availability makes an important distinction.
Before belief updating proper can happen, several other processes matter:
- What evidence was searched for?
- Was it accepted?
- Which hypothesis was specified?
- What information was integrated?
- How was the information reasoned about?
This means apparent “failure to update” may arise earlier in the pipeline.
The evidence never entered.
Or entered under the wrong representation.
You Cannot Revise From Evidence You Never Admit
A person sees contradictory evidence and immediately labels the source untrustworthy.
Sometimes that is correct.
Sources can be poor.
But if every contradiction is rejected before evaluation, revision becomes impossible.
The mind protects the model by controlling the evidence gate.
Local Revision
The café now closes at 7.
Change one field.
The rest of the model remains.
This is local revision.
It is cognitively efficient because most knowledge is stable most of the time.
A good mind does not rebuild civilisation after every typo.
Global Revision
Sometimes a mismatch is not local.
The entire explanatory frame is wrong.
A business believes customers are leaving because prices are too high.
Months of price cuts fail.
New evidence shows the actual problem is product reliability.
The model needs global revision because the governing causal structure changed.
The Minimum-Change Principle
When evidence contradicts a model, do not automatically rewrite everything.
Ask:
What is the smallest revision that explains the evidence without breaking what still works?
This is a useful default.
But it has a boundary.
If local patches multiply, the larger model may be wrong.
Patch Debt
Exception.
Another exception.
Another special case.
Soon the “simple” theory requires twenty caveats.
Patch debt is a reader-facing way to describe the point at which local repairs become evidence for global restructuring.
A robust model can handle exceptions.
A dying model survives only by exceptions.
Revision Should Preserve Invariants
Change the café closing time.
Keep its address.
Change a scientific parameter.
Keep the parts of the model still supported by evidence.
Revision without invariance becomes unnecessary destruction.
Revision and Identity
How much can a model change before it becomes a different model?
How much can a political belief change before it becomes another ideology?
How much can a story be edited before it becomes a new story?
Revision is an invariance problem as well as a change problem.
Conceptual Change Is Often Slower Than One Lesson
A student learns that heavier objects do not necessarily fall faster in the simple way intuition suggests.
They can reproduce the scientific answer in class.
Then months later, an intuitive answer returns in another context.
Why?
Old conceptions can remain cognitively available even after new knowledge is learned.
Conceptual revision can involve competition and inhibition, not clean deletion.
The Old Model Can Remain Useful Somewhere
Children learn Newtonian mechanics.
Later physics introduces relativity and quantum theory.
Does Newton become useless?
No.
The operating range changes.
Mature revision often nests an old model inside a larger one rather than deleting it.
Revision Can Narrow a Rule
Initial rule:
All birds fly.
Counterexample:
penguin.
Possible revision:
Many birds fly, but flight is not required for membership in the category.
The rule becomes more precise.
Revision Can Broaden a Category
Initial concept:
tool = object held in the hand.
Then consider software tools, machine tools and cognitive tools.
The category may broaden around function rather than physical form.
Revision Can Change the Dimension Entirely
A child groups whales with fish because both swim.
Later biology shifts the important dimensions:
- respiration,
- reproduction,
- anatomy,
- evolutionary relationship.
The revision is not merely moving one item between boxes.
The classification system itself changes.
Revision and Evidence Weight
One weak observation should not erase a well-supported model.
One decisive counterexample can destroy a universal mathematical claim.
Revision magnitude should match evidential force.
This is where evidence and logical form matter.
Revision and Confirmation Bias
Humans are not neutral updating machines.
We search selectively.
Interpret selectively.
Remember selectively.
Belief updating can therefore become asymmetric.
A 2024 review of optimism bias and belief updating discusses evidence that desirable and undesirable information can be weighted differently under some conditions, while also emphasising that mechanisms and boundary conditions remain active research topics.
Revision requires monitoring the evidence gate, not only the final belief.
Revision in Mathematics
A student conjectures:
Every prime number is odd.
Then meets 2.
The universal claim collapses.
Revision:
Every prime number greater than 2 is odd.
Mathematics trains unusually disciplined revision because counterexamples have precise logical force.
Revision in Science
A model predicts one pattern.
New data persistently contradict it.
Scientists may revise:
- a parameter,
- an auxiliary assumption,
- a measurement model,
- the mechanism,
- the entire theoretical framework.
The size of revision depends on where the contradiction survives scrutiny.
Revision in English
A first draft claims:
The character is selfish.
Later reading reveals repeated sacrifices for family.
The student can:
- ignore the evidence,
- force every scene into the old claim,
- revise toward a more nuanced interpretation.
Interpretation grows through revision.
Revision in Writing
Editing fixes errors inside the existing structure.
Revision can change the structure itself.
Move paragraph.
Replace thesis.
Remove evidence.
Change audience.
Good writers revise models of what the text is trying to do, not only sentences.
Revision in Organisations
A strategy worked for five years.
The environment changes.
Competitors enter.
Technology changes.
Customer expectations shift.
Past success can become a revision barrier because the old model carries status and sunk cost.
Robust organisations preserve capabilities while revising assumptions.
Revision Can Be Too Fast
One bad outcome.
Entire strategy abandoned.
One strange data point.
Entire theory rewritten.
This is noise panic.
Revision should be proportional to evidence and environmental volatility.
Revision Can Be Too Slow
Repeated failures.
Same explanation.
Repeated customer complaints.
Same process.
Repeated experimental anomalies.
Same theory protected by more patches.
This is state inertia.
Revision and Feedback
Feedback provides the return signal.
Revision changes the model.
If feedback arrives but no representation changes, the loop has not completed.
Revision and Error
Error says expected and actual differ.
Revision asks:
What part of the model deserves to change because of that difference?
Revision and Uncertainty
Revision should not always replace one certainty with another.
Sometimes the proper update is:
I am less certain than I was.
New evidence can widen the model before it narrows it.
Revision and Confidence
A high-confidence error should trigger deeper revision than a low-confidence guess, all else equal.
Why?
The model was not merely wrong.
It was strongly trusted.
Confidence identifies which wrong representations are deeply embedded.
Revision and Counterexample
A counterexample can force revision by proving that a universal rule is too broad.
But the cognitive response is not automatic.
People can treat the counterexample as an exception and preserve the old rule.
The next article asks how one case can carry enough logical force to break a beautiful generalisation.
The Revision Audit
- What exactly does the new evidence contradict?
- Is the evidence reliable and discriminating?
- Could the mismatch be noise?
- Which part of the model still works?
- What is the smallest adequate revision?
- How many patches already exist?
- Would a new representation explain more with fewer exceptions?
- What alternative model is available?
- What prediction changes after revision?
- What future evidence would test the revised model?
A Practical Exercise: Patch or Rebuild?
Take one belief that received contradictory evidence.
Write:
- original model,
- new evidence,
- local patch,
- possible global alternative,
- what each predicts next.
Then wait for future evidence.
Revision becomes testable rather than emotional.
A Practical Exercise: Find the Protected Assumption
When a model accumulates exceptions, ask:
Which assumption am I refusing to revise?
That assumption may be carrying the whole failure.
A Primary-to-Adult Progression in Revision
Primary: change the answer when evidence changes
Children learn that mistakes can be corrected and categories can gain exceptions.
Lower secondary: revise rules and explanations
Students learn to distinguish a local exception from evidence that a whole rule needs narrowing.
Upper secondary: manage competing models
Learners compare hypotheses, revise confidence, inspect assumptions and understand why contradictory data do not automatically identify the correct replacement.
Adulthood: preserve useful structure while replacing obsolete assumptions
Professional revision requires change without amnesia: keep reliable capabilities, update failing models and test the revised system against new reality.
Five Revision Failures
1. Revision Refusal
No evidence can change the model.
2. Revision Panic
One noisy observation causes total rebuilding.
3. Patch Addiction
The model survives through endless exceptions instead of structural repair.
4. Replacement Vacuum
The old model is attacked without a usable alternative, so the learner returns to it.
5. Revision Without Test
A new explanation feels better but produces no new prediction that can be checked.
Frequently Asked Questions
Is cognitive revision the same as studying for an exam?
No. Exam revision commonly means reviewing learned material. Cognitive revision here means changing a belief, concept or representation in response to evidence.
Does contradictory evidence automatically change beliefs?
No. Learners may reject, ignore or reinterpret anomalous information. Conceptual change research suggests that contradiction works better when learners can engage with it and possess a viable alternative explanation.
Should we always make the smallest possible change?
Usually it is sensible to preserve supported structure, but repeated local patches can become evidence that the deeper model is wrong.
Can old and new models coexist?
Yes. Old intuitions can remain cognitively available, and older scientific models can remain useful inside limited operating ranges even after broader models are learned.
Why do people resist revision?
Reasons can include weak alternatives, selective evidence search, identity and motivation, prior investment, habitual accessibility and genuine doubts about source quality.
Research Notes and Further Reading
For a modern theoretical framework separating belief updating from evidence search, acceptance, hypothesis specification and reasoning, see Sommer, Musolino and Hemmer, Updating, Evidence Evaluation, and Operator Availability: A Theoretical Framework for Understanding Belief.
For how learners respond to contradictory information, see Patrice Potvin’s review Response of Science Learners to Contradicting Information. It cautions against assuming that anomaly or conflict alone is sufficient for conceptual change.
For contemporary work on asymmetric belief updating, see From Perception to Projection: Exploring Neuroaffective Advances in Understanding Optimism Bias and Belief Updating. The article reviews evidence and competing frameworks rather than establishing one universal mechanism.
Final Thought: A Good Model Knows How to Survive Being Wrong
You arrive at the café.
The door is locked.
Your old belief fails.
But almost everything else survives.
The café still exists.
The address remains.
The menu remains.
One thing changed.
Closing time.
This is the elegance of revision.
Do not worship the old model.
Do not burn the whole library.
Change exactly what the evidence deserves.
Then return the revised model to the world and see whether it works better.