On the other side too, many of the best chat bots are being built as essentially Expert Systems, rather than Deep Learning, and there's a lot of old research and ideas to dust off from that field. Right now a lot of the chat bot Expert Systems are ad hoc, from what I gather, and it will be interesting if there's a resurgence in exploring and modernizing a lot of the old ideas and classic systems.

As with the brief flurry of the Semantic Web era, I think the lesson with chat bot VI/AI will again be as much in striking the balance between things like Deep Learning and things like Expert Systems, assuming Deep Learning will never be perfect and Expert Systems will never be easy/cheap (building a good one still requires a domain expert in that field to train it in whatever rules engine you are using to express the domain).