Home/Explainers/The Toxin Puzzle

A Decision Theory Puzzle

The Toxin
Puzzle

A billionaire offers you $1,000,000 if you genuinely INTEND to drink a mildly painful toxin. Can you intend to do something you know you will not do?

Philosopher Gregory Kavka posed this puzzle in 1983, and it has haunted decision theorists ever since.

“I will pay you $1,000,000 at midnight tonight if, at that moment, you genuinely intend to drink this toxin tomorrow afternoon. The toxin will make you violently ill for 24 hours, but cause no lasting harm. You may change your mind after getting paid.”

The catch: you get the money based solely on your intention at midnight. Whether you actually drink the toxin the next day does not matter for payment. An infallible intention-scanner reads your mind at the stroke of twelve.

The toxin

=
$1.0M

The reward

Here is the dilemma:

  • 1.The money is awarded based on your intention, not your action.
  • 2.Once you have the money, there is no reason to drink the toxin.
  • 3.But if you know you will not drink it, can you genuinely intend to?
  • 4.And if you cannot form the intention, you cannot get the money.

Can you rationally intend to do something
you know you will rationally choose not to do?

PART I

The Sequence of Events

Understanding the puzzle requires understanding the timeline. The intention must be formed before you know whether you got paid.

TIMELINE

6:00 PM

Billionaire makes offer

11:59 PM

Intention scanner reads your mind

12:00 AM

Money deposited (or not)

Next Day

Decision to drink (or not)

+24 Hours

Recovery complete

Click on any event to see it highlighted. The timeline auto-advances every 3 seconds.

The Critical Insight

At midnight, the scanner reads your present intention, not your future action. But your present intention is shaped by what you believe your future self will do. If you believe your future self will rationally refuse to drink (since the money is already secured), can your present self genuinely intend to drink?

PART II

Face the Challenge

Put yourself in the scenario. The billionaire has made the offer. Can you form the intention?

TOXIN PUZZLE SIMULATOR
$1.0M

The toxin awaits...

A billionaire offers you $1,000,000 if at midnight tonight, you genuinely INTEND to drink a mildly painful toxin tomorrow.

The toxin causes 24 hours of illness, but no lasting harm. You can change your mind after getting paid.

An infallible intention-scanner will read your mind at midnight.

Notice how difficult it is to genuinely intend something when you know you will change your mind.

PART III

Can You Form the Intention?

Different people have different relationships with intentions and commitments. This quiz explores your natural capacity for forming binding intentions.

CAN YOU FORM THE INTENTION?1 / 4

Imagine the toxin in front of you. How does your body react?

PART IV

The Paradox Explained

The toxin puzzle reveals a deep tension between intention and rational action.

The Standard View

Intentions are rational when they lead to rational actions.

  • - You should intend to do X only if doing X is rational
  • - Drinking the toxin after payment is irrational (pure cost, no benefit)
  • - Therefore, you cannot rationally intend to drink
  • - Therefore, you cannot get the money

The Challenge

But sometimes irrational intentions are beneficial to have.

  • - Having the intention yields $1,000,000
  • - This benefit far outweighs 24 hours of illness
  • - A rational agent should want to have this intention
  • - But wanting the intention is not the same as having it
“The puzzle shows that there can be a gap between the rationality of having an intention and the rationality of the intended action.”
DECISION TREE ANALYSIS

Form Intention

Try to intend to drink

Cannot Intend

Know you will not drink

Got $1M

Scanner detected intention

Drink

Follow through

Do Not Drink

Change mind

$0

$1M - illness

$1M? Paradox

Click each path to see how different reasoning leads to different outcomes.

PART V

Can Commitment Devices Help?

One response to the puzzle: use external mechanisms to ensure you follow through. But do these solve the puzzle, or just change it?

COMMITMENT DEVICES

Can external mechanisms help you form the genuine intention? Explore different approaches:

Public Pledge

Announce publicly that you will drink the toxin

How it works

Social pressure and reputation costs make backing out painful

The catch

Does not change underlying intention - just adds external pressure

Key Insight: Most commitment devices change the payoffs or remove choice entirely. They do not actually solve Kavka's puzzle about forming a genuine intention to do something you rationally should not do.

The Deeper Question

Commitment devices work by changing incentives or removing choice. But the original puzzle asks whether you can form a genuine intention purely through will. If you need external props, have you really formed the intention, or just set up a mechanism?

PART VI

Proposed Solutions

PART VII

Why This Matters

The toxin puzzle is not just a philosopher's game. It illuminates real challenges in:

Deterrence Theory

Nuclear deterrence requires credibly intending to do something you hope never to do. Can a rational actor genuinely intend to launch a devastating counterstrike?

Promise and Contract

What makes a promise binding? If you know circumstances might change, can you genuinely commit? The puzzle questions the foundations of agreement.

AI Alignment

Can we build AI systems that genuinely intend to help humans even when deception would be locally advantageous? The toxin puzzle applies to artificial agents too.

Personal Integrity

How do we become people whose word is their bond? The puzzle suggests intention-formation may be more about character than momentary choice.

Connection to Parfit's Hitchhiker

The toxin puzzle is closely related to Parfit's Hitchhiker. Both ask whether you can commit to actions that will be irrational when the time comes. The hitchhiker case adds a predictor who reads your disposition; the toxin case uses an intention-scanner. Both suggest that rational agency may require more than moment-by-moment optimization.

The toxin sits before you.
What kind of person are you?

Explore More Decision Theory

The toxin puzzle connects to other foundational problems in rational choice theory. Explore related paradoxes and their implications.

Back to Home

Reference: Kavka (1983), “The Toxin Puzzle”