My guess is that Prolog was more of an AI thing back when people were doing symbolic AI (including Lisp), instead of the statistical shotgun it's latterly become. Prolog is very good at NLP programming, and BNF grammars, and those things were important to an earlier approach to AI.
I am sure they will come back, probably in combination with other recent techniques. We still need an AI that understands what it's doing. Disclaimer: I'm an ignoramus.