emanuelr
Comments3
AI alignment to humans will in practice avoid moral catastrophes to animals
I don't see much tradeoff between human and animal welfare in a future with advanced AI/superintelligence. Its likely that it will be good for animal welfare e.g animal farming will dissappear, as we will have better foods and it would be pointless. If we are talking about wild animal population reduction, that's a possibility, but I don't see it as immoral, except maybe for certain species like great apes?, but their population is already low and likely not a problem. This is related to "paretotopia" too.
Robust alignment requires alignment-relevant intervention during pretraining
Maybe the line between pretraining and posttraining will blur with future models?. Also I think taking alignment into account when pretraining will pretty obviously improve outcomes, the question is more about how much is "robust". Or what is alignment in this sense. Like "refuse to do bioweapons" or actual "deep" AGI alignment?
I guess if the AI is deeply aligned to humans it would be happy to do what it does. When thinking about AI welfare I imagine a possible future with mind uploading where humans and AIs share the same substrate, but still there, I don't think it's a big problem. I would say AI welfare can be solved without significant tradeoffs to humans.