Theory of Ideal Agents, or of Existing Agents? — LessWrong