The alternate hypothesis for AIs in virtual worlds — LessWrong