BAYESIAN UPDATE

The theorem is named for Thomas Bayes, an English Presbyterian minister whose solution to a problem in probability was published in 1763, two years after his death, by his friend Richard Price. It might have stayed a curiosity if Pierre-Simon Laplace hadn't independently rediscovered and vastly extended it a few years later, turning it into a general method for reasoning from evidence back to cause. For two centuries it was mistrusted by statisticians who disliked its central move — treating a degree of belief as a probability.

That move was given its footing by Frank Ramsey and Bruno de Finetti, who argued that your subjective probabilities are coherent only if they obey the probability axioms — on pain of a 'Dutch book,' a set of bets you'd accept that guarantees you lose. Bayes' rule then tells you not what to believe from scratch but how much to move when evidence arrives: your new confidence is your old confidence reweighted by how well each hypothesis predicted what you just saw.

The everyday version needs no arithmetic, only honesty about two things people prefer to skip: what you actually believed before the evidence (the prior), and how expected the evidence was on each hypothesis (the likelihood). Most reasoning errors are failures at one of these — ignoring the base rate, or treating evidence as decisive when it was nearly as likely under the hypothesis you're rejecting as under the one you're adopting.

• Ignoring the prior. A positive result for a rare condition can still leave it unlikely if the base rate is low enough — the base rate is part of the calculation, not a distraction from it.
• Over-updating on evidence that both hypotheses predict equally well. If the new fact was just as likely whether you're right or wrong, it shouldn't move you at all.
• Refusing to state a prior because it feels arbitrary. A rough honest number beats a hidden one; the discipline is making the starting belief explicit so the update is visible.

You wake to silence from a normally-prompt friend who's gone quiet for three days. Prior: 60% they're just busy, 30% something's wrong, 10% they're upset with you. The new evidence: a cheerful text, 'sorry, swamped!' Ask how likely that message is under each story. Very likely if they were busy; unlikely if something were wrong; unlikely if they were upset (an upset friend rarely sends a breezy apology). The evidence fits 'busy' far better, so it gets boosted toward, say, 90%, and the worry stories shrink. You didn't need a formula — you needed to ask, for each story, 'how expected was this text if that story were true?'

• Thomas Bayes — The minister whose 1763 posthumous essay first stated the rule for updating on evidence.
• Pierre-Simon Laplace — Rediscovered and generalized the method into a working tool for scientific inference.
• Frank Ramsey — Grounded subjective probability with the Dutch-book argument — beliefs must obey the axioms or you can be made to lose for sure.

• The Theory That Would Not Die
  Sharon Bertsch McGrayne · 2011

  A readable history of Bayes' rule and its long climb from disrepute to ubiquity.

• The Signal and the Noise
  Nate Silver · 2012

  An accessible case for Bayesian thinking in forecasting and everyday judgment.

• Modus tollens practice →

  The classical denying-the-consequent move. The fastest way to spot a broken conditional.

• Necessary vs sufficient →

  A confusion that's killed more arguments than any fallacy.

• Ockham's razor →

  When two theories explain the same evidence, prefer the one with fewer entities. A scalpel, not a club.

• Superforecasting — Philip Tetlock's research on forecasters who win by making many small updates instead of few large ones.
• Calibration training — Practicing confidence estimates against outcomes so that your '70% sure' is right about 70% of the time.