The thing is, in many cases you don't need to know if the diagnosis is correct, so long as the results are directionally correct. If the outcome is an improvement in health, that meets your threshold for value delivered, hence payment rendered. I think there are many cases when you don't need to be smarter than the other person to benefit from their expertise. You can take it a step further and make the distinction between absolute value versus perceived/relative value -- if we have a patient with cancer, absolute value might mean destroying all cancer cells (which is an absolute metric but extremely hard to model). But perceived value might be "feeling better". It's important to make this distinction particularly in healthcare because absolute benefits and relative benefits are
extremely important. Hospice/end-of-life care is a good example of this (as is the cancer example above). Most people would prefer relative or perceived comfort as opposed to absolute results that only end up prolonging a painful process.
Outside of healthcare, for another example example, you don't need to understand engineering and tension dynamics in order to appreciate that the second floor of your home support you and your roof doesn't cave in on itself. I don't have examples on hand, but in medicine we've had cases where people do some logical variation of "the right thing for the wrong reasons." I.e., rituals that correlate with healthy outcomes because there is some not-yet-understood principle at play (i.e, you don't need to understand germ theory to benefit from cleanliness rituals).
I think this is one of those logical conundrums which falls into the trap of "in theory, in practice". The artificial constraints around the problem space result in artificial logical conundrums.