Predicate logic does indeed say that it a fallacy to conclude B=>A from A=>B. However, Bayes theorem does allow a weaker version of this, and lots of the machine learning that I do (did?) at work relies on this. Also, there's the issue that no matter how strong your grasp on predicate logic is, you don't know how strong everyone else's is.