Conjecture and Refutation
Treat every belief as a guess and hunt for the test that could kill it.
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 82%
The default theory of knowledge is justified true belief: gather evidence, justify a claim, and call it known. The Popperian correction is that nothing can be shown true — all we ever have are conjectures about reality. What makes a conjecture knowledge is not the evidence stacked behind it but the attacks it has survived. So the working loop is: state the guess explicitly, put a genuine rival next to it, design an experiment whose outcome would rule one of them out, run it, and discard the loser. The survivor is held only for as long as no problem arises. Because every theory stays open to refutation, the method never terminates — which is precisely what makes unbounded progress possible. Newton's gravity was never true; it contained truth, and general relativity corrected its errors.
Origin
Karl Popper's philosophy of science, extended by David Deutsch in The Beginning of Infinity, and discussed at length in this episode of Naval alongside the crucial-experiment idea and the parallel with evolution by natural selection.
Core principles
- 01No method exists that can show any piece of knowledge to be true.
- 02All knowledge is conjectural — a best current guess, held provisionally.
- 03Progress comes from removing errors, not from accumulating confidence.
- 04A guess that survived serious attack outranks one that was never tested.
- 05Rising confidence from repeated confirmations is instinct, not method.
- 06A good explanation explains the seen in terms of the unseen.
How to run it
- 1
State the guess in a form that could be wrong
Write the theory you actually hold as a claim about the world, specific enough that some observable outcome would contradict it. If nothing could contradict it, it is not yet a conjecture.
Pro tip Write the sentence that starts 'I would be wrong if…' before you write the sentence that starts 'I believe…'.
- 2
Put a genuine rival on the table
Find the strongest competing explanation that is actually on offer, not a version you can knock over. Refutation only creates knowledge when there is somewhere better to go.
Pro tip If you cannot state the rival in a form its holder would recognise, you do not understand it yet.
Watch out Comparing your theory against 'no theory' teaches you nothing — you will always win.
- 3
Design the crucial experiment
Find the observation that splits the two theories: if it goes one way, one is ruled out; if it goes the other, the second is ruled out. That asymmetry is what makes a test worth running.
Pro tip Write down both predicted outcomes and what each would force you to abandon, before you look at the data.
Watch out A test both theories predict identically is a confirmation ritual, not an experiment.
- 4
Run it and kill the loser
Execute the test and discard the refuted theory rather than adding a patch to save it. The point of the exercise is the deletion, not the survival.
Pro tip Say out loud what you have just stopped believing — it makes the correction stick.
Watch out Endless caveats and special cases added to rescue a refuted theory are how a live idea becomes a dogma.
- 5
Hold the survivor loosely and keep hunting
Keep the surviving theory only for as long as no problem arises with it. Treat it as the current best approximation and stay on the lookout for the next flaw.
Pro tip Schedule a recurring review that asks one question: what has shown up since that this theory cannot explain?
Watch out Declaring a theory final ends the error-correction loop and stops progress in that domain.
In the wild
For centuries every experiment on gravity came back consistent with Newton's theory. On a confidence-accumulating view, the day before it was overturned was the day you should have believed it most. Then two rival theories were on the table at once — Newtonian gravity and general relativity — and a crucial observation around 1919 ruled one of them out. Newton's theory was never true; it contained truth, approximating the inverse square law closely but not exactly, and general relativity corrected its errors. General relativity is now the only theory of gravity on offer, and it too is held provisionally: there can be no final theory of gravity, only better approximations.
→ One decisive test replaced centuries of confirmations and moved the field forward, without ever claiming final truth.
A founder believes churn is a pricing problem; her head of product believes it is an onboarding problem. Instead of gathering more evidence for the pricing story, she writes both as conjectures and asks what observation would rule one out: if churn is pricing, a cohort moved to a lower tier should retain; if it is onboarding, retention should track completion of setup rather than plan. Both cohorts run for one quarter. Retention moves with setup completion and is flat across tiers.
→ The pricing theory is discarded in a quarter rather than defended for a year, and the team moves on to attack the surviving onboarding theory.
Common mistakes
Collecting confirmations instead of designing tests
Every additional data point that fits your theory feels like progress but cannot distinguish it from a rival that fits equally well. Confidence rises right up to the moment the theory is refuted.
Treating the surviving theory as final
A theory that has survived every test so far is still only the best current approximation. Declaring it settled removes the mechanism that would have caught its next error.
Dressing up an untestable hunch as a conjecture
A claim that no possible observation could contradict cannot be refuted and therefore cannot generate knowledge, however sophisticated it sounds.
Is it for you?
Best for
Anyone who has to hold beliefs about a moving reality — researchers, founders, investors, and operators making bets under deep uncertainty.
Not ideal for
Purely statistical estimation tasks where you only need a calibrated probability, not an explanation of the underlying mechanism.
From the transcript
“The problem with this is that there is no method of showing as true any piece of knowledge.”
“No, it's a guess that has stood up against trials, against attempts to show that it's false.”
“You've got these two competing theories, and you have a particular experiment that, if it goes one way, one theory is ruled out, but the…”
From the episode
The Beginning of Infinity, Part 2