The Evolution of AI in Human-Computer Gaming

Category Machine Learning

Wednesday - January 10 2024, 12:39 UTC - 10 months ago

tldr #

Human-computer gaming has a long history of AI development, with systems like Chinook and Deep Blue defeating world champions in games like checkers and chess. Recent AIs like AlphaGo and OpenAI Five have shown great success, using modern techniques to defeat professional human players in complex games. A new paper in Machine Intelligence Research reviews recent breakthrough AIs, their corresponding games, and predicts future trends in this field.

content #

Human-computer gaming has been a crucial tool in testing and advancing key artificial intelligence technologies. Since the proposal of the Turing Test in 1950, researchers have been striving to develop AIs that can challenge and defeat professional human players. From defeating checkers world champion Marion Tinsley in 1994 to beating chess grandmaster Garry Kasparov in 1997, AIs have continuously pushed the boundaries of human-computer gaming .

The first human-computer game, the Turing Test, was proposed in 1950 to test machine intelligence.

In recent years, there has been a rapid development in human-computer gaming AIs, with AIs like DQN agent, AlphaGo, Libratus, and OpenAI Five successfully defeating professional human players in certain games. These AIs use a combination of modern techniques like deep learning, reinforcement learning, and self-playing to display decision-making intelligence.AlphaGo Zero, for instance, uses Monte Carlo tree search, self-play, and deep learning to defeat dozens of professional go players .

In 1994, Chinook, a checkers AI, became the first AI to defeat a world champion human player.

OpenAI Five, on the other hand, uses self-play, deep reinforcement learning, and continual transfer via surgery to become the first AI to beat the world champions in an eSports game. The success of these AIs in complex games like StarCraft and Dota2 has shown that current techniques have the potential to solve very complex games.With the recent breakthrough of AIs in games like Honor of Kings and Mahjong, researchers have started to question the challenges of current techniques in human-computer gaming and predict future trends .

Deep Blue, developed by IBM, became the first AI to beat a world champion human player in chess in 1997.

A new paper published in Machine Intelligence Research aims to review recent successful human-computer gaming AIs and provide insights into the challenges and future trends in the field.The paper covers four types of games - board games, card games, first person shooting (FPS) games, and real-time strategy (RTS) games - and their corresponding AIs. These include well-known AIs like AlphaGo, AlphaGo Zero, AlphaZero, Libratus, and DeepStack, as well as more recent breakthrough AIs like DouZero, Suphx, FTW, AlphaStar, OpenAI Five, JueWu, and Commander .

Recent AI systems like AlphaGo, OpenAI Five, and AlphaStar have shown remarkable success in defeating professional human players.

The remainder of the paper discusses each game and its corresponding AI in depth, exploring the techniques used and analyzing the challenges and potential future developments. With the rapid progress of human-computer gaming AIs, the possibilities for future advancements in this field seem endless.

hashtags #
worddensity #

Share