I remember reading a great post somewhere about using the visual shape of usenet discussion threads (as rendered by common news readers) to spot patterns like trolling, political tangents, etc. This was only proposed as an NI (Natural Intelligence) method, not AI, but I wonder...