Bayesian Updating

How it works

The core practice has three steps. First, start with a prior — your current best estimate of how likely something is, based on everything you know so far. Second, when new evidence arrives, ask: “How likely would I be to see this evidence if my belief were true? How likely if it were false?” Third, update your prior proportionally. Strong evidence should move your belief a lot. Weak evidence should move it a little. Irrelevant evidence shouldn’t move it at all.

The critical discipline is in how much you update. Most people either don’t update at all (clinging to their original view regardless of evidence) or update too much (completely flipping their view based on a single data point). Good Bayesian reasoning means proportional updating — neither stubborn nor reactive.

Case study: How Alan Turing cracked Enigma through systematic updating

During World War II, the German Enigma machine could produce 159 quintillion possible settings. Brute-force code-breaking was impossible. Alan Turing’s breakthrough at Bletchley Park was essentially applied Bayesian reasoning at industrial scale.

Turing’s team started with prior probabilities — educated guesses about which Enigma settings were being used, based on patterns in German military communication. Each intercepted message provided new evidence that either confirmed or contradicted these priors. Turing built a machine (the Bombe) that systematically tested hypotheses and updated probabilities, eliminating impossible configurations until only the correct one remained.

The critical insight wasn’t mathematical genius alone. It was the discipline of treating intelligence analysis as a probabilistic process — starting with priors, updating with evidence, and letting the accumulated updates converge on the truth. Each individual piece of evidence was weak. The systematic updating was what cracked the code.

Real-world examples

Hiring. Your prior: this candidate is about 50% likely to be a strong hire based on their CV. Evidence 1: they give thoughtful answers in the interview → update to 65%. Evidence 2: their reference is lukewarm → update back to 50%. Evidence 3: their work sample is excellent → update to 75%. Each piece of evidence shifts the estimate, but doesn’t determine it alone.

Medical diagnosis. A test for a rare disease comes back positive. Without Bayesian thinking, you’d panic — “the test is 95% accurate, so I almost certainly have it.” But if the disease affects 1 in 10,000 people, even a 95% accurate test gives more false positives than true positives. The base rate (your prior) matters enormously.

When to use it

Use Bayesian updating whenever you’re forming beliefs about uncertain things — which is most of the time. It’s especially important when evaluating evidence that confirms what you already believe (are you updating too little because it’s expected, or is this genuinely strong evidence?), when evaluating surprising evidence (is this strong enough to override your prior?), and when making predictions and tracking whether you’re well-calibrated.

Common mistakes

The biggest mistake is anchoring too strongly to your prior and barely updating regardless of evidence — the hallmark of a closed mind. The second biggest is the opposite: updating wildly based on weak evidence, the hallmark of an anxious one. The third is ignoring base rates. Before asking “does this evidence fit my theory?”, ask “how common is my theory in the first place?”

Try it now

Pick a belief you hold with moderate confidence — about a work project, a relationship, or a prediction about the future. Assign it a probability (e.g., “70% likely”). Now think of the last piece of relevant evidence you received. Did you update your probability? By how much? Was that proportional to the strength of the evidence?

Related models

These models complement Bayesian Updating — they address similar situations from different angles.