AI safety: различия между версиями

'''AI safety''' is a field of research used to describe a sequence of increasingly specific problems when developing [[Artificial Intelligence]] and [[AGI]]. The goals center around reducing risks posed by AI, especially powerful AI and includes problems in misuse, robustness, reliability, security, privacy, etc. (Subsumes AI control.) AI control: ensuring that AI systems try to do the right thing, and in particular that they don’t competently pursue the wrong thing. [[Value alignment]]: understanding how to build AI systems that share human preferences/values, typically by learning them from humans.<ref>https://ai-alignment.com/ai-safety-vs-control-vs-alignment-2a4b42a863cc</ref>