← Back to Paradoxes

The Sleeping Beauty Problem

A probability puzzle that divides experts: Is the answer 1/2 or 1/3?

PHILOSOPHERS STILL CAN'T AGREE!

You're put to sleep. A fair coin is flipped. If Heads: wake once (Monday). If Tails: wake twice (Monday & Tuesday), with memory erased between. When awakened: What's the probability the coin landed Heads?

The Experiment

🌙
Sunday Night
Sleeping Beauty goes to sleep. A fair coin is flipped.
🪙
Coin Flip
Heads: Wake only Monday. Tails: Wake Monday AND Tuesday.
💊
Memory Wipe
After Monday, memory drug makes her forget being awakened.
❓
The Question
Each awakening: "What's your credence the coin landed Heads?"
👸
Sleeping...
?
Flip the coin
Sunday
Setup
Monday
?
Tuesday
?

Sleeping Beauty is Asked:

"What is your degree of belief that the coin landed HEADS?"

Your Position

1/2 Halfer

The coin is fair. She knew she'd wake up. No new information → still 50%!

1/3 Thirder

There are 3 possible awakening situations. Only 1 has Heads → 33%!

Run Experiment

Statistics

0
Heads Flips
0
Tails Flips
0
Total Awakenings
0
Heads Awakenings
Fraction of awakenings that are Heads:
--
--
Thirders predict: ~33% | Halfers argue this is the wrong question

The Two Arguments

Halfer (P = 1/2)

Premise: Sleeping Beauty knew she would be awakened. Being awake provides no new information about the coin.

Argument: Before the experiment, she knew the coin was fair (50/50). When she wakes, she still knows the coin was fair. Nothing has changed → P(Heads) = 1/2.

Advocate: David Lewis

Thirder (P = 1/3)

Premise: There are three equally likely "centered" situations: (Heads, Monday), (Tails, Monday), (Tails, Tuesday).

Argument: Only one of these three involves Heads. By self-locating uncertainty, P(Heads) = 1/3. Long-run frequency confirms this!

Advocate: Adam Elga (majority view)

Why It Matters

Anthropic Reasoning: This puzzle is central to debates about the Doomsday Argument, the Simulation Argument, and fine-tuning in cosmology.

Self-Locating Belief: How should rational agents reason about "where" they are in possibility space? Standard probability theory doesn't directly answer this.

AI Safety: If we build AI systems that need to reason about their own copies or simulations, they'll face versions of this problem.

The Key Insight

The paradox reveals that there might be two different legitimate questions:

Question A: "What is the probability of the coin flip outcome?" → 1/2 (the coin is fair)

Question B: "Given that I find myself awake, what's the probability this awakening is a Heads-awakening?" → 1/3 (frequentist)

Halfers answer Question A. Thirders answer Question B. Perhaps the paradox is simply about which question Sleeping Beauty is asking!