In this blog post, I want to argue that EDT gets it right: one-boxing is the correct action in medical Newcomb problems.
This is a remarkable conclusion. Did you think through what you are actually arguing here? That people should make medical decisions on the basis of confounded correlations rather than on the basis of causal effects?
The structure of the argument seems to be as follows:
(1) Evidential decision theory allegedly reaches the right conclusion in certain thought experiments involving a perfect predictor, Omega.
(2) Therefore, EDT is the correct decision theory in general, and people should use it in medical applications which do not involve a perfect predictor
(3) Patients should therefore choose treatment on the basis of confounded correlations, rather than causal effects
However, if we are in the business of declaring certain decisions to be correct, and then evaluating decision theories on the basis of whether they output our preferred answer, we could just as well declare “two-boxing” as the correct answer for medical Newcomb’s problems, and exclude EDT on the basis of getting it wrong. The structure of this argument would be identical to yours and reach the opposite conclusion.
If you want to challenge what you call “two-boxing” in Medical Newcomb’s problem, then please demonstrate that EDT will maximize patient utility in situations that do not involve Omega.
On a sidenote: Can we please agree, as a community, to stop using the terms “one-box” and “two-box” when discussing problems which are not Newcomb’s problem, such as the misnamed “Medical Newcomb’s problem”?
This is a remarkable conclusion. Did you think through what you are actually arguing here? That people should make medical decisions on the basis of confounded correlations rather than on the basis of causal effects?
The structure of the argument seems to be as follows:
(1) Evidential decision theory allegedly reaches the right conclusion in certain thought experiments involving a perfect predictor, Omega.
(2) Therefore, EDT is the correct decision theory in general, and people should use it in medical applications which do not involve a perfect predictor
(3) Patients should therefore choose treatment on the basis of confounded correlations, rather than causal effects
However, if we are in the business of declaring certain decisions to be correct, and then evaluating decision theories on the basis of whether they output our preferred answer, we could just as well declare “two-boxing” as the correct answer for medical Newcomb’s problems, and exclude EDT on the basis of getting it wrong. The structure of this argument would be identical to yours and reach the opposite conclusion.
If you want to challenge what you call “two-boxing” in Medical Newcomb’s problem, then please demonstrate that EDT will maximize patient utility in situations that do not involve Omega.
On a sidenote: Can we please agree, as a community, to stop using the terms “one-box” and “two-box” when discussing problems which are not Newcomb’s problem, such as the misnamed “Medical Newcomb’s problem”?