Probabilistic Thinking

The future isn’t yes or no

Our brains like certainty. Will it rain? Will the launch work? Will I get the job? We want a yes or a no. But the world rarely hands us either. It hands us odds.

Probabilistic thinking means holding several possible futures at once, giving each a rough likelihood, and updating those numbers as you learn more.

A forecast of “30% chance of rain” doesn’t mean it will rain 30% of the day. It means that on days like this one, it rains about 3 times in 10.

Run the 100-day simulator at two different chances

A hundred days like today

Chance of rain30%

Press the button to see what “30%” looks like.

Three habits of probabilistic thinkers

Think in odds

Say “maybe 70%” instead of “yes”. A number forces you to admit how sure you really are.

Update as evidence arrives

Start from how often things like this usually happen, then nudge your estimate with each new clue.

Weigh the payoff, not just the odds

A likely small win can be worth less than an unlikely big one. Multiply chance by size.

Pick a bet

Three offers. Choose the one you’d take, then play it a hundred times and see how it turns out.

Play a bet a hundred times

How well do you know what you know?

Being calibrated means that when you say “80% sure”, you’re right about 80% of the time. Most people are overconfident. Answer each claim, then say how sure you are.

Finish the calibration test

Finish the quiz

Calibration test

Traps your intuition falls into

Update your beliefs, step by step

A school uses an AI detector on student essays. It flags Arjun’s essay. The vendor says the detector is “90% accurate”. Should the school punish him? Work it out properly.

Reach a decision you can defend

First, your gut

The detector flagged the essay and is “90% accurate”. How likely is it that Arjun actually used AI?

My guess50%

Lock in a number. We will come back to it.

Start with how common it is

Before any test, how many students in this school actually use AI to write essays? This is the base rate. Imagine 1,000 essays.

Students who use AI5%

Used AI Honest

Now run the detector on all 1,000

“90% accurate” hides two separate numbers. Set both.

Catches real AI essays90%

Wrongly flags honest essays10%

AI, flagged Honest, flagged Not flagged

Arjun is one of the flagged essays

Throw away everything that wasn’t flagged. Of the flagged essays, what share were really written with AI?

chance he used AI

your gut guess

Gather more evidence

Today’s answer becomes tomorrow’s starting point. Suppose a second, independent check also points to AI, such as a teacher noticing the essay doesn’t match Arjun’s class work. Watch the estimate move.

What should the school do?

Treat beliefs as estimates with a confidence attached, and move them as evidence arrives.

What are the odds?

Reach for it when

You’re predicting something uncertain: a launch, a hire, a market, an exam result.
A test, a signal or a rumour seems to point one way.
You must choose between options with different risks and rewards.
People around you are speaking in certainties.
Something rare but very costly could happen.

Put it down when

The outcome is genuinely known or easy to look up.
The choice is trivial and quickly reversible. Just decide.
You’re using numbers to make a guess look more precise than it is.

The framework

Find the base rateHow often does this kind of thing happen in general? Start there, not with the story.
Name your estimatePut a number on it, even a rough one: 20%, 60%, 90%.
Weigh the evidenceHow much more likely is this clue if you’re right than if you’re wrong?
UpdateMove your estimate, more for strong evidence and less for weak. Never jump to 0% or 100%.
Weigh the outcomesMultiply each chance by what it would cost or gain. Protect against the ruinous ones.
Decide, then reviewAct on the best estimate, and later check whether your confidence was well placed.

Rules of thumb

Base rates first, stories second.
“90% accurate” means nothing until you know how rare the thing is.
Small risks multiply across long plans.
Judge a decision by its process and odds, not only by how it turned out.
Avoid bets that can wipe you out, even when the odds look good.

Questions to ask yourself

How often does this usually happen?
How sure am I, as a number?
What evidence would change my mind, and by how much?
What is the worst case, and can I survive it?

Common traps

Ignoring the base rate because the specific story is vivid.
Believing a streak means the next result is “due”.
Treating 70% as certain, then being shocked by the 30%.
Refusing to update because it means admitting you were wrong.

All mental models Next: Second-Order Thinking