AI systems already decide how ambulances are routed, how supply chains operate and how autonomous drones plan their missions. Yet when those systems make a risky or counterintuitive choice, humans are often expected to accept it without challenge, warns a new study from the University of Surrey.