In this era of intelligence creating intelligence, safety is paramount. I look to solipsism. AI systems in a simulated world. Cause and effect occurs there. Simulation is indistinguishable from reality for that AI model. Reality fed to an AI model doesn’t need to be base reality we humans cohabitate. No matter how bad AI gets, a switch that toggles between base reality and simulated reality for a robot is a failsafe that cannot fail.
AI model can plan and plan and plan in its head like a daydreamer.