The phenomenon of insight (also called “Aha!” or “Eureka!”moments) is considered a core component of creative cogni-tion. It is also a puzzle and a challenge for statistics-basedapproaches to behavior such as associative learning and rein-forcement learning. We simulate a classic experiment on in-sight in pigeons using deep Reinforcement Learning. We showthat prior experience may produce large and rapid performanceimprovements reminiscent of insights, and we suggest theo-retical connections between concepts from machine learning(such as the value function or overfitting) and concepts frompsychology (such as feelings-of-warmth and the einstellung ef-fect). However, the simulated pigeons were slower than thereal pigeons at solving the test problem, requiring a greateramount of trial and error: their “insightful” behavior was sud-den by comparison with learning from scratch, but slow bycomparison with real pigeons. This leaves open the questionof whether incremental improvements to reinforcement learn-ing algorithms will be sufficient to produce insightful behavior.