Your latter point is the way I would expect it to be more like. The system should learn meaning as it's used. Assume that I set this up; I would then tell users to help 'train' it (*) by having the system at the dumb state, and having the users ask it any questions, and help it by indicating which questions were the correct one from the meaning of the question. Once trained, it would then be put on line for all users to use, and it would still learn from that. IT's just that to get it from 'dump' to 'reasonably intelligent' would need a lot of help of monks and others.

(*) I think thinking neural networks for this, but I don't think that's the way to go on the back-end AI.

---

Dr. Michael K. Neylon - mneylon-pm@masemware.com || "You've left the lens cap of your mind on again, Pinky" - The Brain