The Sleeping Beauty Problem
You're put to sleep, a coin is flipped, and you might be woken once or twice. On waking, what probability should you assign to heads?
Arnold Zuboff sketched a version of this problem in 1987, and Adam Elga formalized it in 2000. It divides probability theorists sharply into two camps with no consensus in sight, despite the scenario being simple enough to explain in a paragraph.
Elga, A. (2000). Self-Locating Belief and the Sleeping Beauty Problem. Analysis, 60(2), 143–147.
The setup
On Sunday, Sleeping Beauty is put to sleep. A fair coin is flipped. If it lands heads, she is woken once on Monday and the experiment ends. If it lands tails, she is woken on Monday, given a drug that erases her memory of waking, put back to sleep, and woken again on Tuesday. She knows all of this in advance. The question: on waking, what probability should she assign to the coin having landed heads?
She knows a fair coin was flipped. She has no memory of previous wakings. She has no way of telling what day it is.
Thirders and halfers
Thirders argue she should assign probability 1/3 to heads. Here is why: there are three possible waking situations, Monday-Heads, Monday-Tails, and Tuesday-Tails. If the experiment were run many times, she would find herself in each of these situations roughly equally often. She should therefore assign equal credence to each, making the probability of heads 1/3.
Halfers argue she should assign probability 1/2 to heads. When she fell asleep Sunday, she knew the coin was fair: probability 1/2 each way. On waking, she learns nothing she didn't already know. She knew she would be woken at least once regardless of the outcome. No new information means no update. The probability stays at 1/2.
Both positions are defended by serious philosophers and probability theorists. Neither has a knockdown argument. The disagreement is not about the facts of the case but about what probability means when the question involves your own location in time.
Self-locating belief
The Sleeping Beauty problem belongs to a class of puzzles about self-locating belief: uncertainty not about what happened in the world but about where and when you are within it. Standard probability theory was designed to handle uncertainty about states of affairs. It is less well-equipped for uncertainty about which observer you are, or which moment of your life you are currently in.
These puzzles arise in cosmology too. If the universe is very large, or infinite, there may be many copies of you in different situations. What probability should you assign to being in any particular one? The Sleeping Beauty problem is a clean, small version of that much larger question. The disagreement between thirders and halfers is not merely technical. It reflects a deeper split about whether your perspective, the fact that you are the one asking, is itself a piece of evidence about the world.
Discussion questions
- When Sleeping Beauty wakes up, is the probability of heads 1/2 or 1/3?
- Does it feel like a meaningful question to ask what the probability is from inside an experiment designed this way?
- Does your intuition match the mathematical answer, or does one of them feel wrong?
Take it to the dinner table.
Get 3 thought experiments for memorable conversations, designed for dinner, with friends, at events, or anywhere small talk has gone on too long.