Why do 23 people give a 50% chance of a shared birthday?
Because you are not comparing 23 birthdays against 365 days — you are comparing 253 pairs of people against each other, and 253 chances at a 1-in-365 event is a coin flip.
Simple intuition
The plain reason, in everyday words
Most people answer this question by imagining themselves in the room and asking how many others they would have to check before finding their own birthday. With 22 other people and 365 days, that feels hopeless, and it is — the chance is only about 6%. But that is not the question. The question lets any two people in the room match, not just you and someone else. So it is not 22 comparisons, it is every possible pairing: you with each of the 22, then the next person with the 21 they have not been compared to yet, and so on. That comes to 253 separate pairs, each with its own small chance of matching. Give a 1-in-365 event 253 attempts and you are at roughly even odds. The room is much better connected than it looks.
You need about 183 people, because that is half of 365.
That would be roughly right for guaranteeing a match by the pigeonhole principle, where 366 people force one. Even odds arrive far sooner because the number of pairs, not the number of people, is what grows.
It means there is a 50% chance someone shares your birthday.
With 22 others that chance is about 6%. Reaching 50% for one specific date takes around 253 people. The two questions differ by an order of magnitude and swapping them is the source of the surprise.
The result depends on assuming birthdays are evenly spread, so it is unrealistic.
Real birthdays are uneven, and every departure from evenness makes matches more likely, not less. The uniform assumption gives the most pessimistic answer, so 23 is if anything slightly too many.
It is a paradox — the mathematics contradicts itself.
Nothing contradicts anything. It is called a paradox only in the older sense of a result that offends intuition. The arithmetic is elementary and checkable in a few lines.
This is the cheapest available cure for a specific and expensive intuition failure: assuming that the chance of a coincidence scales with the number of things rather than the number of pairs. That mistake shows up when people find a striking pattern in data, when engineers size an identifier space, and when a hash function is chosen for a signature scheme. The birthday case is memorable precisely because you can check it against a real room of people.
Who worked it out
The problem is usually credited to Richard von Mises in 1939, though Harold Davenport is said to have known it earlier without publishing. It circulated as a lecture-room demonstration long before it appeared in textbooks.
What problem forced it
Its early appeal was pedagogical: it is the shortest convincing demonstration that untrained probabilistic intuition is not merely imprecise but systematically biased in a predictable direction.
How it changed since
It stopped being a curiosity when cryptographers realised the same square-root scaling bounds the security of every hash function. The birthday attack became a standard design constraint, and 'you only get half your bit length against collisions' is now a rule every protocol designer works around.
The pigeonhole principle
The companion result that tells you when a match becomes certain rather than merely likely, and why the two numbers are so far apart.
Hash collisions and the birthday attack
The same square-root rule, applied where it determines whether a signature can be forged.
Written for Curio rather than collected from a forum — it is part of the curated corpus that ships with the platform. The references it draws on are listed under Sources.