Module I of VI
Beliefs as bets
How sure are you that you locked the front door this morning? You’d probably say “pretty sure.” Pressed, you might upgrade to “almost certain.” Somewhere under those words sits a number, and the whole art of Bayesian thinking begins with coaxing it out. “Pretty sure” might mean 80 percent. It might mean 98. Those two beliefs should lead to different behavior — one of them turns the car around.
A probability, on this view, is a measure of confidence. The claim “there’s a 70 percent chance it rains tomorrow” isn’t a statement about many repeated tomorrows; it says how strongly you believe in this one. And there’s an old, honest test for what your number really is: a bet. Imagine a ticket that pays 100 € if the belief turns out true and nothing otherwise. The most you’d pay for that ticket is your probability. Someone 70 percent sure who wouldn’t pay 40 € is fooling themselves about one of the two numbers.
Panel I·1 — The belief meter
probability · odds · priceThe panel shows the same belief in three dialects. Percent is the familiar one. The second — odds — will matter more here: 75 percent means three chances for, one against, written 3 : 1. Odds feel old-fashioned, like something shouted at a racetrack, and they have one property that makes the rest of this tutorial run: when evidence arrives, odds update by simple multiplication. That’s a promissory note; Module IV redeems it.
The third dialect, price, exposes something percentages hide. A belief priced at 99 € leaves only a euro of doubt, and going from 99 to 99.9 buys ten times more certainty than it sounds like. The distance between “very likely” and “certain” is where most expensive mistakes live.
One more idea completes the toolkit: calibration. Your numbers are calibrated when, of all the things you call “80 percent likely,” about 80 percent actually happen. Weather forecasters, checked against decades of their own predictions, turn out to be remarkably good at this — a classic finding of Allan Murphy and Robert Winkler’s; most of the rest of us run overconfident, treating “90 percent sure” as a rhetorical flourish rather than a promise to be wrong one time in ten. Calibration is a skill, and it improves with exactly the kind of practice this tutorial’s panels provide: state a number, see the outcome, adjust.
Next, these numbers go to work on a puzzle that famously humbles intuition — and we’ll solve it with nothing fancier than counting.