Qurated: Reframing LessWrong-style decision theory as "commitment theory"
The Insight: You're Not Choosing an Action. You're Choosing a Type of Person to Have Been.
Decision theory keeps producing recommendations that sound insane—"burn to death rather than open a box you can see is empty"—and normal people correctly recoil. The problem isn't the math. It's the frame.
Reframe it as commitment theory: the question isn't "what should I do now?" but "what kind of agent should I have committed to being, before I knew which situation I'd land in?" Once you ask that question, the "insane" answers become the obviously correct ones.
The Bomb Problem
Omega, a near-perfect predictor, is dead. He left you two boxes. Left may contain a live bomb; Right definitely contains a glitter bomb (annoying, not lethal). Here's the catch: Omega put the bomb in Left only if he predicted you'd choose Left. You can see, with your own eyes, that Left is empty. Standard decision theory says: take Left, it's empty, guaranteed clean outcome.
But agents who are the type of agent who takes Left when they see it's empty are exactly the agents Omega predicts will take Left—and so Omega puts the bomb there. The seeing-it's-empty moment is a trap sprung by your own predictable character. The agents who survive to see an empty Left box are the ones who were never going to take it.
The Mental Model: Policies, Not Moments
Stop evaluating individual decisions in isolation. Evaluate the policy that generates them, judged from before you knew which world you'd wake up in.
- Old frame: "Given what I now know, what maximizes expected value?"
- New frame: "What policy, committed to in advance across all the situations I might face, produces the best outcomes—and am I the kind of agent who can be trusted to actually follow it?"
This is why it's commitment theory, not decision theory. The hard cases in LessWrong-style decision theory (Newcomb's Problem, counterfactual mugging, the Bomb) aren't puzzles about clever choices in the moment. They're puzzles about which prior commitments pay off, given that the world contains predictors who see your commitments coming.
Why This Reframe Matters
-
It kills the "irrational in the moment" objection. Critics say: "You're advocating actions that look stupid given current information." True—but irrelevant. You're not optimizing the moment; you're optimizing the policy that produces the moment. A soldier who'd desert under fire never gets picked for the mission that requires courage. The commitment is upstream of the situation.
-
It matches human moral intuition. We already praise people for being trustworthy, loyal, or brave as traits—not for calculating trustworthiness fresh each time. Commitment theory says decision theory should work the same way: character over calculation.
-
It explains why "cooperating with your past self" isn't mystical. Your past self and current self share a policy. Honoring that policy isn't sacrifice—it's coherence across time, the same coherence that lets you trust your own future promises.
The Actionable Takeaway
Before facing a hard decision, ask: "What policy would I have wanted to commit to, before I knew which branch of reality I'd end up in?" Then follow that policy, even when the local, in-the-moment math looks wrong. The agents who do this are the agents predictors reward, cooperators trust, and futures favor.
Decision theory's weirdest conclusions stop being weird the moment you stop asking "what's smart right now?" and start asking "what kind of agent creates good outcomes across all the moments?"