> They seem to be good at predicting things with known variables, but not with unknown unknowns.

Exactly. 96% consensus on a guess is just as good as 99% or a 1%. It doesn't matter how many experts agree if they base their decision on unreliable information.