Using theoretical calculations, computer scientists show it would be fundamentally impossible for humans to control a super-intelligent AI.