Q-learning is a reinforcement learning algorithm created by Christopher Watkins in his 1989 doctoral thesis, Learning from Delayed Rewards. It learns by trial and error, keeping a running estimate of how good it is to take each action in each situation, and updating those estimates from the rewards it receives plus the best expected future reward. Because it learns from experience without needing a model of the world, it became the theoretical backbone of reinforcement learning, the branch of AI that learns by doing rather than by being told. Its ideas run beneath the modern era of AI. Deep Q-Networks, DeepMind's 2015 program that learned to play Atari games from raw pixels, was Q-learning upgraded with neural networks, and the same family of techniques powered the reinforcement learning systems that followed, including the lineage that led to AlphaGo. Q-learning is the foundation of reinforcement learning, the ground AlphaGo and everything that followed stand on. The ecosystem grew around it too. Q-learning gave reinforcement learning its core idea, and OpenAI Gym in 2016 gave it a standardized training ground, the ecosystem behind the modern RL boom.