Protect the Means of Error Correction
Judge any system by whether it can still find and fix its own mistakes.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 78%
Deutsch treats fallibility as the baseline: mistakes are the normal condition of humans, and all we can do is try to find them. That reframes what matters morally and practically. It is not whether a decision is right, because you cannot guarantee that. It is whether, once the decision turns out wrong, the error can still be found and fixed. Not destroying the means of correcting errors is therefore the heart of morality — because if there is no way of correcting errors, sooner or later one of them will get you. Applied to institutions, the test is concrete: can leadership be replaced, can a policy be reversed, can dissent be voiced. North Korea is the extreme case — political error correction has been destroyed, so the society is trapped in a local minimum it is very hard to climb out of.
Origin
Deutsch's fallibilist reading of Popper, stated in this Naval conversation as the basis of morality rather than as a political theory.
Core principles
- 01Mistakes are the normal condition, not the exception — assume you are making them now.
- 02It does not take malevolence to cause harm; well-intentioned errors nobody could see are enough.
- 03The survivable failure is a mistake you can still correct.
- 04Any move that removes the correction mechanism eventually kills you, whatever its immediate merit.
- 05Anti-rational memes that block criticism are the real constraint, not physical limits.
How to run it
- 1
Assume you are currently wrong about something
Start from the position that mistakes are normal and that some of your current errors are invisible to you. This removes the incentive to design for being right and shifts it to designing for being corrected.
Watch out Malevolence is not required — most damaging errors are well-intentioned and unseen at the time.
- 2
Name the correction mechanism
For the system in question, identify explicitly how an error would actually get found and reversed: elections, code review, a reversible migration, a customer who can leave, a person allowed to object.
Pro tip If you cannot name the mechanism in one sentence, it probably does not exist.
- 3
Score every proposal against that mechanism
Ask of each change whether it makes errors easier or harder to detect and undo. Efficiency gains that route around criticism are the ones to scrutinise hardest.
Pro tip Speed bought by removing a check is a loan against a future you cannot see.
- 4
Refuse the trades that remove correction
Treat 'this eliminates the ability to reverse or object' as a veto condition rather than a cost to be weighed. The asymmetry is that one uncorrectable error eventually gets you.
Watch out Local minima are easy to enter and very hard to climb out of once the mechanism is gone.
- 5
Watch for anti-rational memes
Look for norms that punish criticism rather than answering it — ideas that survive by making themselves uncriticisable. These destroy correction without any formal rule change.
Pro tip A claim that critics are not worth listening to is the tell.
In the wild
Deutsch and Naval use North Korea as the limit case. There are no elections, and revolution is very hard because the gang in charge is armed to the teeth. The political means of error correction have been destroyed for a long time, so whatever errors the regime is making cannot be found and reversed from inside. The result is not merely bad policy but a trap: humanity there is stuck in a local minimum, and it is very hard to climb out of that hole. Deutsch's wider worry is that if too much of the world adopts that mindset, the species stagnates, having lost the ability to make new discoveries.
→ The damage is structural rather than tied to any single policy — no mechanism means every future error also stands.
An illustrative application: a company frustrated by slow releases removes the staging environment, the rollback path, and the reviewer who kept objecting, and ships faster for a quarter. Each removal was defensible on its own and each was a correction mechanism. When a genuinely bad change lands, nobody detects it early and nothing can be reversed cleanly, so the failure runs to its full length. Under Deutsch's rule the decisive question was never whether any individual release was correct, but whether a wrong release could still be caught and undone.
→ Speed gained by deleting correction machinery is repaid in full by the first error that has nowhere to be caught.
Common mistakes
Optimising for being right
Effort spent guaranteeing a decision is correct is largely wasted, because fallibility is the baseline. The durable investment is in the machinery that catches the errors you will make anyway.
Treating dissent as friction
Objection is the cheapest form of error detection available. Removing it feels like removing overhead and is actually removing the alarm.
Assuming harm requires bad intent
Waiting for a malevolent actor before defending correction mechanisms misses the normal case, which is well-intentioned errors nobody could see were errors.
Is it for you?
Best for
Anyone designing institutions, teams, governance, or long-lived systems where the cost of an uncorrected error compounds.
Not ideal for
Short, fully reversible, one-off decisions where no correction machinery is at stake.
From the transcript
“mistakes are the normal condition of humans all we can do is try to find them”
“correcting errors is the heart of morality because if there is no way of correcting errors then sooner or later one of those will get…”
“they've destroyed the means of political error correction”
From the episode
David Deutsch: Knowledge Creation and The Human Race, Part 1