Home/Explainers/The Sleeping Beauty Problem

Self-Locating Probability

The Sleeping Beauty Problem

A coin is flipped Sunday. If heads, you wake once. If tails, you wake twice (with memory erased). You wake up. What's P(heads)?

Here is the setup: Sleeping Beauty volunteers for an experiment. On Sunday, she is put to sleep. A fair coin is flipped.

If HEADS: Beauty is awakened on Monday, interviewed, and the experiment ends.

If TAILS: Beauty is awakened on Monday, interviewed, put back to sleep with her memory of the awakening erased, then awakened again on Tuesday and interviewed.

Beauty wakes up. She knows all the rules. She is asked:

"What is your credence that the coin landed heads?"

This seemingly simple puzzle has divided philosophers for over two decades. There are two main camps:

Halfers say 1/2

"The coin is fair. Nothing happened to change this. P(Heads) = 1/2."

Thirders say 1/3

"Two-thirds of awakenings are tails-awakenings. P(Heads) = 1/3."

Both positions are defended by serious philosophers. Both have compelling arguments. Let's explore why this problem is so hard.

PART I

The Experiment Protocol

First, let's make sure we understand exactly what happens. Watch the experiment unfold step by step:

THE EXPERIMENT PROTOCOL

Sunday

?

Coin Flip

Monday

Memory Wipe

Tuesday

Click "Run Experiment" to see the Sleeping Beauty protocol in action.

Run the experiment several times to see both heads and tails outcomes.

PART II

The Halfer vs. Thirder Debate

The disagreement comes down to how Beauty should update her beliefs when she wakes up. Click each position to see its full argument:

THE TWO POSITIONS

Halfer

1/2

"The coin is fair. Nothing happens during the experiment to change this. P(Heads) = 1/2."

Thirder

1/3

"There are three possible awakenings: Monday-Heads, Monday-Tails, Tuesday-Tails. Only one is a heads-awakening. P(Heads) = 1/3."

All possible awakening states:

H-Mon

Heads

T-Mon

Tails

T-Tue

Tails

Thirders: Each state equally likely (1/3 each). Halfers: H-Mon = 1/2, T-Mon = 1/4, T-Tue = 1/4.

PART III

The Thirder Argument: Count Awakenings

The Thirder argument is intuitive when you run the experiment many times. If you sample a random awakening, what fraction are heads-awakenings?

AWAKENING FREQUENCY SIMULATOR

Run many experiments and count awakenings. Among all awakenings, what fraction occur in heads-experiments vs tails-experiments?

The simulation shows that roughly 1/3 of awakenings occur in heads-experiments, while 2/3occur in tails-experiments. Thirders argue that Beauty should reason as if she is randomly sampled from all possible awakenings.

PART IV

The Betting Argument

Here's a compelling argument for Thirders: consider how Beauty should bet. If she bets $1 at even odds on each awakening, which strategy wins?

BETTING GAME

You are Beauty. Each awakening, bet on whether the coin landed heads or tails. You win 1:1 on correct bets. What strategy maximizes your winnings?

Your Balance

$100

Bet Amount

Total Bets

0

Even Halfers often agree that Beauty should bet as if P(Heads) = 1/3. But some argue this shows credence and betting can come apart - you can believe P(Heads) = 1/2 while betting as if it were 1/3.

PART V

The Halfer Defense

Halfers have a powerful argument: Beauty gains no new information when she wakes up. She knew in advance that she would wake up at least once. Why should her credences change?

Halfer reasoning: On Sunday night, before falling asleep, Beauty knows the coin is fair: P(Heads) = 1/2. When she wakes up, what new information does she have? She was certain she would wake up. Nothing has happened to shift the probability.

Halfers argue that the Thirder position confuses frequency of awakenings with probability of coin outcomes. Yes, 2/3 of awakenings are tails-awakenings, but that doesn't mean the coin is more likely to have landed tails.

PART VI

Calculate the Probabilities

What if the coin isn't fair? See how the Halfer and Thirder answers differ based on different priors:

PROBABILITY CALCULATOR
Biased to tailsFair coinBiased to heads

P(Heads | You are awake)

33.3%

Thirder reasoning: With prior 50%, accounting for the fact that tails-awakenings are twice as frequent, P(Heads) updates to 33.3%. With a fair coin prior, this gives exactly 1/3.

PART VII

The Double-Halfer Controversy

Some philosophers go even further than Halfers. What if Beauty is told which day it is?

THE DOUBLE-HALFER POSITION

There is an even more controversial position: the double-halfer.

Double-Halfer

1/2 always

"Even if you are told it is Monday, P(Heads) should still be 1/2."

Click to expand and see why double-halfers face difficult objections.

PART VIII

Why This Problem Matters

The Sleeping Beauty problem is not just an academic puzzle. Your answer has implications for some of the biggest questions in philosophy and physics.

WHY THIS MATTERS

Sleeping Beauty is not just a puzzle. It connects to some of the deepest questions in philosophy, physics, and AI safety.

Anthropic Reasoning

How should we reason about our own existence?

The Simulation Argument

Are we living in a simulation?

Many-Worlds Interpretation

How to think about quantum branching?

Decision Theory

How should Beauty bet?

Self-Locating Beliefs

Where and when are you?

CONCLUSION

Where Do You Stand?

After two decades of debate, there is no consensus. Some of the brightest philosophers remain Halfers; others are convinced Thirders. Both sides have technical arguments in their favor.

The real insight:

Probability theory, as usually formulated, does not tell us how to handle self-locating uncertainty - uncertainty about where and when we are. Sleeping Beauty reveals this gap in our foundations.

“The question is not what credence Beauty should have, but what credence even is.”

Beauty wakes up. She opens her eyes. What should she believe?

Explore Related Paradoxes

The Sleeping Beauty problem connects to deep questions about anthropic reasoning, decision theory, and the foundations of probability.

Back to Home

References: Elga (2000), Lewis (2001), Bostrom (2007)