A study pitting human children against large language models in basic reasoning problems revealed AIs to have one big blind spot: innovation.