Well researched and explained! Thank you for doing this. When I talk about how human brain-like AI could be good in that we have a lot of research to help us understand human-like agents, but alternately could be very bad if we get a near miss and end up with psycopathic AI.
I think it would be quite valuable for alignment to have people working on the idea of how to test for psychopathy in a way which could work for both humans and ML models. Reaction time stuff probably doesn’t translate. Elaborate narrative simulation scenarios work only if you have some way to check if the subject is fooled by the simulation. Tricky.
Well researched and explained! Thank you for doing this. When I talk about how human brain-like AI could be good in that we have a lot of research to help us understand human-like agents, but alternately could be very bad if we get a near miss and end up with psycopathic AI. I think it would be quite valuable for alignment to have people working on the idea of how to test for psychopathy in a way which could work for both humans and ML models. Reaction time stuff probably doesn’t translate. Elaborate narrative simulation scenarios work only if you have some way to check if the subject is fooled by the simulation. Tricky.