Want to develop a risk-management framework for AI? Treat it like a human.