5 ms·
Concrete Problems in AI Safety
- jakozaur 10y agoHmm article seems to miss socialeconomics aspects. Can it put too much power or wealth in hands of few? Can hackers cheat AI to its advantage?is it more risky than current systems? Analogy, the article analyze as if main risk of nuclear power was accidents or uncontrolled reactions. However the main problem with nuclear is with weapons which could give destructive power to few individuals. Sometimes civilian use cases can be mixed with military one (e.g. Iran, North Korea).
- Smaug123 10y agoIt doesn't claim to give all the aspects; only some concrete and fairly well-specified ones.
- wrsh07 10y agoThis was posted earlier, with a link to the OpenAI blogpost: https://news.ycombinator.com/item?id=11950687 https://news.ycombinator.com/item?id=11950687