Meanwhile, in "The Economist".

[rain_gryphon]

This is fascinating. Someone has written a utility that "reads" a substantial portion of the internet, then, using what it's harvested, assembles answers that seem most likely to result from questions asked. The answers aren't necessarily meaningful in and of themselves, although they pass the Turing test. Most of them sound like a politician's or CEO's Rogerian non-answer. What you're getting here, IMHO, is the concensus opinion of a part of the internet on a subject, although I'm sure there's some sort of filtering to keep the answers coherent and the viewpoint within the scope of the answer consistent.

It would be interesting to train the AI on smaller, selected portions such as chat boards, etc, and see what concensus opinions emerge from those groups. I'm sure the military has had that ability for some years now.

It would be interesting to try differently-worded variations on logically-identical questions, to see whether different answers result. It has the potential to be a tool to help keep politicians and advertisers honest, although like everything else, I'm sure it will be used for evil and stupidity.