An AI Takeover Thought Experiment — LessWrong