Towards Safe Robots: Approaching Asimov's 1st Law — LessWrong