This has surprising similarities with data science theory: multi-armed bandit with a decreasing-epsilon strategy (children) or an epsilon-first strategy (adults).