Shravan writes,
In his book, Fooled by Randomness, Taleb essentially rejects the notion that past results [knowledge] can increase information incrementally. For example, he says [approvingly] that Popper “refused to blindly accept the notion that knowledge can always increase with incremental information–which is the foundation of statistical inference” (p. 127). Isn’t this the same thing as saying that informative priors are not really informative? Informative priors represent our previous knowledge–Taleb rejects that as a basis for predicting the future. I see his point regarding trading practices, but I wonder if his position would extend to any statistically driven inference in, say, experimental psychology. I would think not.
My reply: I think the point is that the model itself continually needs to be reassessed, and in good work it ends up getting revised at irregular intervals; see here.