AI Philosophers have studied ‘counterfactuals’ for decades. Will they help us unlock the mysteries of AI?