Coordinative AI development hypothetical — LessWrong