Advancing AI Research and Applications by Learning from What Went Wrong and Why