I feel like it won’t be AI until we figure out how to point it back at itself, have it review its own answers and then be ‘happy’ when it’s answers are right. Not necessarily like if the user gives it a good score, but if it recognizes an answer it had given was actually used, or a prediction it makes if proved true (if I answer this way, the user is likely to ask this as its next question, etc) and it starts changing its behaviour, and asking itself questions to get better at that.