Well, they can if all you do is expect them to actually fly the plane. And even then, these autopilots make trivial mistakes if the get the wrong input data.

Human pilots make tons of non-trivial decisions that don't have anything to do with the actual maneuvering of the airplane, including where to land the airplane in an emergency. So again, the fallacy is believing that the job is a specific, well-defined task. When we have sentient computers that can process 3D vision, hearing, smell etc as well as a human and exceed humans in decision making, we can replace human pilots and expect the same performance.