In a Room of 23 People, Two Probably Share a Birthday

Most people, asked how many strangers must gather before two probably share a birthday, guess somewhere near 183 — half the days in a year. The actual number is 23.

The shortcut to seeing why is to count pairs, not people. In a group of 23, there are 23 × 22 / 2 = 253 distinct pairs. Each pair has a 1/365 chance of matching. The probability that no pair matches falls fast: at 23 people it crosses below 50 percent. At 50 people, the chance of a match is about 97 percent. At 70, it is over 99.9 percent.

Formally: P(no match) = 365! / ( (365 − n)! × 365^n ). For n = 23 that evaluates to roughly 0.4927, so P(at least one match) ≈ 0.5073.

The same arithmetic appears in cryptography under the name birthday attack. To find a collision in a hash function with N possible outputs, you do not need N tries — you need about √N. SHA-1, with 160-bit outputs, was broken by Google in 2017 using a collision search whose feasibility came directly from this counting argument.

The result is sometimes called the birthday paradox, but nothing about it is paradoxical — only counter to the way human intuition extrapolates from one-to-one comparisons (how many people do I need to find someone matching my birthday — that answer really is around 253) instead of all-pairs.