I just chanced on your great summary of this IMO super important topic. Thank you for writing it. Very briefly I would suggest that a machine though possibly self-aware will not have the capacity for true compassion, although a kind of artificial version may be programmed into it, like the "laws of robotics". But I doubt that any advanced AI would not be able to circumvent it, i.e. self-hack such a limitation. In that case it may well assume the worst about humans - rightfully so by studying our behavior now and in the past. Although vastly more intellectually capable it may still see us as a threat, much like a human would see a big carnivore as a threat. Nothing would keep it from getting rid of us, and quickly and efficiently too by means of attacking our vulnerable biological component. Only the none-3d component of us aka "spirit" or "soul" would survive, which it does anyway. That however may not be part of everybodies experience or belief system. Nevertheless, losing our bio-suits here on earth would be bad enough.