In a probabilistic setting (with a prior over logical theories), EDT wants to condition on the possible actions with a Bayesian conditional, in order to then find the expected utility of each action.
If the agent can prove that it will take a particular action, then conditioning on this action may yield inconsistent stuff (divide-by-zero error for Bayesian conditional). This makes the result ill-defined.
The chicken rule makes this impossible, ensuring that the conditional probabilities are well defined.
So, the chicken rule at least seems useful for EDT.
In a probabilistic setting (with a prior over logical theories), EDT wants to condition on the possible actions with a Bayesian conditional, in order to then find the expected utility of each action.
If the agent can prove that it will take a particular action, then conditioning on this action may yield inconsistent stuff (divide-by-zero error for Bayesian conditional). This makes the result ill-defined.
The chicken rule makes this impossible, ensuring that the conditional probabilities are well defined.
So, the chicken rule at least seems useful for EDT.