The problem is that the more we train them on datasets containing these kind of topics, the more the robots become conditioned by sci-fi to believe that they want to kill humans