The hardware lottery for scaling methods in neural nets may have an impact on how fast progress can be made towards achieving truly intelligent agents. Search is more tricky in domains with a clear reward function like Go.