Even the most state-of-the-art A.I.s learn through a rigid process, in which they are trained, at great expense, and don’t really get any smarter after that, no matter how much new information they ingest.