People won't invest in a broken system. If you're worried about AI alignment, what you're actually concerned with is accountability. You want to discourage people from building bad systems - so do it. Make people liable for implementing AI in situations where it can cost human lives. Vague premonitions of "alignment" don't really bolster an otherwise baseless argument. You'd be better off appealing to my fear of humans than my fear of machines.
When someone gets run over by a steamroller, we don't call it a 'steamroller alignment issue'. It is a matter of human liability, either conscious or subconscious. Wielding superhuman power is not new, nor is dealing with it's liabilities.
2 comments
[ 4.8 ms ] story [ 22.1 ms ] threadWhen someone gets run over by a steamroller, we don't call it a 'steamroller alignment issue'. It is a matter of human liability, either conscious or subconscious. Wielding superhuman power is not new, nor is dealing with it's liabilities.