So this make AI research on (cooperative) inverse reinforcement learning and more generally value learning risky, by making it more likely our worst case outcomes (worse than extinction) will be represented and optimized for, right?
Yep! Though of course the situation is complicated and there are many factors etc. etc.
So this make AI research on (cooperative) inverse reinforcement learning and more generally value learning risky, by making it more likely our worst case outcomes (worse than extinction) will be represented and optimized for, right?
Yep! Though of course the situation is complicated and there are many factors etc. etc.