Google DeepMind's Joel Z Leibo warns that "solipsistic" AI training paradigms make superintelligent systems inherently uncooperative upon deployment

The training treats environments as static sources of feedback.