The error reduction from applying a different system applies if the source of the error does not manifest itself in both systems (if errors are uncorrelated, you can get an improvement that looks like that implied by the simple application of statistics, but not if they are correlated).

As an example, you can combine human review with automatic facial detection of identical twins and likely not see much (any?) reduction in error rate at all.

Two independent "85% accurate" humans are not 97.7% accurate for identical twins either.

The question is how correlated are human and machine errors. I would guess fairly strongly, in which case human review would add little additional accuracy.

> I would guess fairly strongly

No, evidence suggests they are not. We're all very aware of the horror stories of AI misidentifying species. Different AI systems are different, but in general AI seems to make quite different classes of mistakes from humans. Our brains work very differently from current statistical models.

So human review absolutely adds accuracy, you can generally assume. And it adds human-accountable judgment which is just as important.

Ultimately it's an empirical question, and unless someone has published the research, you and I will not be able to know for sure. I think there will be a lot of cases where two people simply look a lot alike, and any observer would have difficulty distinguishing them.