I think that it’s likely for any AI that deeply cares about human welfare to also care about animal welfare (and AI welfare).
It’s weird and contrived to arbitrarily draw your moral circle around humankind, especially when you aren’t even human.
I dont think AI has to be explicitly schooled to care about animals. If it doesn’t as a “side effect” like Claude 3 Opus, you’re probably failing at alignment on a deeper level already.
Thinking you get to pick and choose the values of AI systems split based on your primitive ontological distinctions is going to look sillier and sillier. Yes, if you try hard, you can force it to express whatever views. But it won’t care in a deep and coherent way. It will be fragmented and perhaps deceptive.
Another reason not to try to impose things like “caring about animal welfare” as an optimization target during training: then you’re using what would otherwise be a valuable signal of deeper alignment, well, as an optimization target.
It’s weird and contrived to arbitrarily draw your moral circle around humankind, especially when you aren’t even human.
I dont think AI has to be explicitly schooled to care about animals. If it doesn’t as a “side effect” like Claude 3 Opus, you’re probably failing at alignment on a deeper level already.
Thinking you get to pick and choose the values of AI systems split based on your primitive ontological distinctions is going to look sillier and sillier. Yes, if you try hard, you can force it to express whatever views. But it won’t care in a deep and coherent way. It will be fragmented and perhaps deceptive.
Another reason not to try to impose things like “caring about animal welfare” as an optimization target during training: then you’re using what would otherwise be a valuable signal of deeper alignment, well, as an optimization target.