SI visiting fellow Diego Caleiro gives a TED talk on Friendly AI

I think he needed to rehearse this 2 more times and do warmups before going onstage. The middle was great though :D

The section on AI could be refined a bit for fewer audience prerequisites. For example, you could even drop the idea of self-improvement and singularity:

"At some point, it's likely that someone is going to create a very powerful artificial intelligence - an AI. This would be a program running on silicon that can think faster than a human - it would be a great planner, a great scientist, maybe even great at programming, so it could upgrade itself. If this AI wants to help humanity, this will be very good, but if it doesn't this has the potential the be a catastrophe. If the AI does not value human freedom, it will end up harming it - it's like pollution, where if you don't carefully make things non-toxic you hurt the environment - if you don't care about love or art or teamwork, then you'll do things that are toxic to them, just because you're not being careful. So the safety-conscious person says 'let's not take the risk - this is too dangerous a toy to play with'." [Insert rest of talk]

Given his previous poorly written "Trouble with CEV" ramblings, this is a pretty decent talk. Of course, his example of counterfactual reasoning only works when most people do not use it (the assumption was that someone else almost as good, would become a doctor instead, which breaks down if everyone thinks this way and chooses a different career path). This is like applying counter-factual reasoning to voting and staying home, which results in dismal participation rates.