A Decision Theory Problem
The Blackmailer's
Empty Threat
When a decision theory tells you to ignore blackmail because refusing is "evidence of innocence," something has gone wrong.
Imagine you receive an anonymous message:
"I know your secret. Pay me $1,000 or I reveal everything."
-- Anonymous
Here's the thing: You have no idea if you have a secret worth hiding.
Maybe the blackmailer has real dirt on you. Maybe they're bluffing. You assign some probability to each possibility.
Standard decision theory (CDT) says: weigh the costs and benefits, consider your actual probability of having a damaging secret, and decide accordingly.
But Evidential Decision Theory says something bizarre:
"If I pay, that's evidence I have something to hide. If I refuse, that's evidence I'm innocent. Therefore, I should refuse!"
EDT confuses evidence with causation. Let's see why this matters.
The Blackmail Scenario
Play through the scenario multiple times. How often should you pay?
The blackmailer contacts you. Do you have a secret?
(Based on your prior, there's a 10% chance)
EDT vs CDT: The Difference
Adjust the parameters and see how each decision theory evaluates your options.
EEvidential Decision Theory
If I pay:
"Paying is evidence I'm guilty"
EU = -$1,000
If I refuse:
"Refusing is evidence I'm innocent"
EU = -$2,000
EDT recommends:
PAY
CCausal Decision Theory
If I pay:
Certain loss of $1,000
EU = -$1,000
If I refuse:
30% x 70% x -$10,000
EU = -$2,100
CDT recommends:
PAY
Both theories agree here. Try adjusting parameters to find where they disagree.
Evidence vs. Causation
The key insight: Your action provides evidence about your state, but it does not cause your state.
EDT's View
EDT treats your action as informative about your hidden state. Refusing "reveals" innocence.
CDT's View
CDT recognizes: Your action cannot change whether you have a secret. Both affect outcome independently.
The Fatal Flaw in EDT
EDT's reasoning: "Refusing to pay is correlated with being innocent. Therefore, if I refuse, I make it more likely that I'm innocent."
But correlation is not causation. Refusing does not make you innocent any more than carrying an umbrella makes it rain. Your guilt or innocence was determined before the blackmailer ever contacted you.
The Payoff Matrix
Hover over each cell to see the outcome in each scenario.
-$1,000
Secret hidden
-$1,000
Wasted money
-$10,000
Secret revealed
$0
Nothing to reveal
Hover over a cell to see the analysis
The key question: What is your actual probability of having a secret? CDT uses this directly. EDT ignores it and instead asks: "What would refusing reveal about me?"
Walk Through the Reasoning
Step through EDT's flawed reasoning process to see where it goes wrong.
Step 1 of 8
Start
A blackmailer contacts you demanding $1,000.
Why This Matters
Evidential Blackmail reveals a deep problem with EDT: it treats your own actions as news about the world, rather than as interventions on the world.
EDT says:
"My action is correlated with my state. Therefore, choosing the action correlated with the good state makes the good state more likely."
CDT says:
"My action cannot change my state. I should use my actual beliefs about my state and calculate expected utility accordingly."
The blackmail case makes EDT's error obvious, but the same flawed reasoning appears in many contexts:
Medical Testing
EDT might say: "Getting tested is evidence I'm sick. So I'll avoid testing to be healthier."
Insurance
EDT might say: "Buying insurance is evidence bad things happen to me. So I'll skip insurance."
Preparation
EDT might say: "Preparing for failure is evidence I'll fail. So I won't prepare."
In each case, EDT confuses the evidence your action provides with the causal effect of your action.
How EDT Defenders Respond
Some EDT defenders argue that in real decision-making, you already have "direct access" to whether you're guilty through introspection - a "tickle" or feeling.
On this view, once you condition on your introspective state, your action provides no additional evidence. EDT and CDT then agree.
Counterargument:
This response fails when there genuinely is uncertainty about your state. The blackmail scenario is specifically constructed so that you do not have certainty about whether you have a damaging secret.
Some defenders bite the bullet: "Yes, EDT says refuse. And that's the right answer! Being a refuser correlates with good outcomes."
Counterargument:
This seems to confuse "what kind of person to be" with "what action to take." The question isn't about character or disposition - it's about this specific decision. And refusing when you have 50% chance of having a real secret seems obviously wrong.
Some argue that in realistic blackmail scenarios, there are always additional factors that resolve the apparent paradox.
Counterargument:
While real cases are complex, the point of the thought experiment is to isolate the theoretical disagreement. If EDT gives the wrong answer in the idealized case, that reveals a problem with the theory itself.
Some philosophers have developed alternatives like Functional Decision Theory (FDT) that try to preserve EDT's good intuitions (in Newcomb's Problem) while avoiding its bad ones (in Evidential Blackmail).
This is a live area of research:
The relationship between evidence, causation, and rational action remains philosophically contested. Evidential Blackmail is one of several thought experiments that probe these deep questions.
The Bottom Line
Evidential Blackmail shows that EDT can give absurd recommendations when your action is correlated with (but does not cause) good or bad states of the world.
The Core Lesson
Your actions can be evidence about the world without being causes of the world. A rational decision theory must distinguish between these. EDT fails to do so.
Whether this means we should adopt CDT, or some newer theory like FDT, remains an open question. But Evidential Blackmail has convinced many philosophers that pure EDT cannot be the right theory of rational choice.
Your actions reveal who you are. They do not create who you are.
Explore More Decision Theory
This is part of our series on decision theory paradoxes and their implications for rational choice.
Reference: Egan (2007), Ahmed (2014)