Google DeepMind published "Human-level control through deep reinforcement learning" in Nature. Given only screen pixels and the game score as input, and using the same algorithm, network architecture, and hyperparameters throughout, a deep Q-network learned to play 49 different Atari 2600 games, reaching a level comparable to a professional human tester on many of them. The paper showed that combining deep learning with reinforcement learning worked in practice, and it demonstrated publicly why Google had bought DeepMind. The same deep reinforcement learning framework was used in AlphaGo the following year.