Provably safe systems: the only path to controllable AGI

Max Tegmark, Steve Omohundro

We describe a path to humanity safely thriving with powerful Artificial General Intelligences (AGIs) by building them to provably satisfy human-specified requirements. We argue tha…