Different goals may bring AI into conflict with us — LessWrong