# nshepperd comments on Conservation of Expected Evidence

• “For a true Bayesian, it is im­pos­si­ble to seek ev­i­dence that con­firms a the­ory”

The im­por­tant part of the sen­tence here is seek. The isn’t about falsifi­ca­tion­ism, but the fact that no ex­per­i­ment you can do can con­firm a the­ory with­out hav­ing some chance of falsify­ing it too. So any ob­ser­va­tion can only provide ev­i­dence for a hy­poth­e­sis if a differ­ent out­come could have pro­vided the op­po­site ev­i­dence.

For in­stance, sup­pose that you flip a coin. You can seek to test the the­ory that the re­sult was `HEADS`, by sim­ply look­ing at the coin with your eyes. There’s a 50% chance that the out­come of this test would be “you see the `HEADS` side”, con­firm­ing your the­ory (`p(HEADS | you see HEADS) ~ 1`). But this only works be­cause there’s also a 50% chance that the out­come of the test would have shown the re­sult to be `TAILS`, falsify­ing your the­ory (`P(HEADS | you see TAILS) ~ 0`). And in fact there’s no way to mea­sure the coin so that one out­come would be ev­i­dence in favour of `HEADS` (`P(HEADS | mea­sure­ment) > 0.5`), with­out the op­po­site re­sult be­ing ev­i­dence against `HEADS` (`P(HEADS | ¬mea­sure­ment) < 0.5`).