He says designing AI with protective drives like human caregivers offers the most effective safeguard against runaway superintelligence.
