Similar Items: The Archimedean trap: Why traditional reinforcement learning will probably not yield AGI