I was mostly thinking in the case of overlapping research as mentioned in the parent comment, not independent research years after the publication (where indeed it sounds impossible to prove you didn't read it).

Basically, if you could show that you had already made significant progress before the publication date, then you might be granted the patent as well.