AI systems assume humans are always certain. Teaching them to anticipate and understand uncertainty may be key to limiting human error.