The idea in one sentence
Bayes’ theorem is a rule for changing your mind sensibly. You start with how likely you thought something was, you see new evidence, and the rule tells you how likely it is now. The key point is that the answer depends on two things: how well the evidence fits the idea, and how plausible the idea was before you saw anything.
You do not need to memorise a formula to use this. The same reasoning can be done by counting cases, which is how the examples below are worked.
A two-deck example
Picture two small decks of 10 cards each. Deck A has 8 red cards and 2 black cards. Deck B has 2 red and 8 black. These are made-up decks, chosen to keep the arithmetic clean.
Someone picks one deck by a fair coin toss, so before any card is drawn each deck is equally likely: 50 percent each. They draw one card from the chosen deck without telling you which deck it is, and show you that it is red.
How likely is it that the deck is A?
Count the ways. Imagine running this setup 100 times. About 50 times the deck is A, and in those, a red card appears 80 percent of the time, which is 40 cases. About 50 times the deck is B, and red appears 20 percent of the time, which is 10 cases. So there are 50 cases that produce a red card, and 40 of them came from deck A.
The chance the deck is A, given a red card, is 40 out of 50, or 80 percent.
That is Bayes’ theorem in action. Written out, it is the chance of the evidence under the idea times the starting chance of the idea, divided by the overall chance of the evidence: (0.5 × 0.8) ÷ (0.5 × 0.8 + 0.5 × 0.2) = 0.4 ÷ 0.5 = 0.8.
Updating a second time
Now suppose the card is put back, the deck is shuffled, and another card is drawn from the same deck. It is red again. Your starting point is now 80 percent for deck A and 20 percent for deck B, because that is what you learned last time.
- Deck A path: 0.8 × 0.8 = 0.64
- Deck B path: 0.2 × 0.2 = 0.04
The chance of deck A is now 0.64 ÷ (0.64 + 0.04) = 0.64 ÷ 0.68, which is about 94 percent.
Each new piece of evidence nudges the belief, and the nudges build on each other. Notice that the belief moved a lot with the first card and less with the second. Evidence has diminishing effects once you are already fairly sure.
Why the starting chance matters
The most common mistake with Bayes’ theorem is ignoring how likely the idea was to begin with. Here is another illustration with invented numbers.
Say 1 in 100 dice in a bag is loaded so that it shows a six half the time, while the other 99 are fair and show a six one time in six. You pick a die at random, roll it once and get a six. Is it likely to be loaded?
Take 10,000 dice. About 100 are loaded, and about half of those, 50, show a six. About 9,900 are fair, and about one sixth of those, 1,650, show a six. Among the 1,700 dice that showed a six, only 50 are loaded. That is about 2.9 percent.
A six is more likely from a loaded die than from a fair one, so it does count as evidence. But loaded dice were rare to begin with, so one six still leaves the die very probably fair. This is why a single odd result is weak grounds for accusing a game of being rigged. Repeated odd results are different, and a proper test of dice fairness uses many rolls.
Where this shows up in games
Bayesian thinking appears whenever you hold a belief that new information can revise:
- In card games, each card you see changes the likelihood of what remains, which connects to conditional probability and to drawing without replacement from a standard deck.
- In games with hidden information, observed behaviour is evidence about what an opponent might hold. The quality of that inference depends on how common each possibility was beforehand, and on how reliably the behaviour separates one possibility from another. Poker players use informal versions of this when they reason about opponent patterns.
- When judging whether a strange streak points to something real, the starting belief matters. Most streaks are ordinary random chance.
Limits and cautions
Bayes’ theorem does not produce certainty. It converts your assumptions and your evidence into a revised probability, and the output is only as good as the inputs. If your starting belief or your estimate of how well the evidence fits is wrong, the result will be too.
It also does not make hidden information readable in games where the odds are fixed against you. The theorem helps you reason clearly about what you observe; it does not change what the game pays. Games with a negative expectation still lose on average over time, however carefully you update your beliefs.
If gambling is affecting your wellbeing, a local support service can offer help.