Google DeepMind artificial intelligence can master the game without learning the rules

Google’s DeepMind artificial intelligence has evolved to the fourth generation, named MuZero. The latest version can master the game without learning the game rules such as chess. This technology can be applied to robotics, industrial systems, and the chaotic real world.

As early as 2016, AlphaGo defeated Go world champions Li Shishi and Ke Jie. This is only the first version of DeepMind’s artificial intelligence. It requires a large amount of human game data and rules data for training before it can be used for actual combat. AlphaGo is the first artificial intelligence that can fully master Go using neural networks and tree search.

Since then, the second generation of AlphaGo Zero, launched in 2017, can train itself by entering rules without relying on human game data, and finally master Go.

Google’s third-generation artificial intelligence AlphaZero can not only learn to Go independently but also master chess and Japanese shogi only by knowing the rules in advance.

According to an article published by Google in “Nature” magazine, the fourth-generation artificial intelligence MuZero can learn the above four chess types and plan winning strategies without knowing the rules. In addition, you can also master the handheld game Atari.

DeepMind said that researchers have been looking for a way for many years to learn to build a model to explain the current environment and use this model to make the best decisions. Until today, most methods are difficult to plan effectively in games like Atari.

MuZero was first launched in 2019 to learn to build a model and solve problems by focusing only on the most important aspect of the environment. By combining this method with AlphaZero’s powerful search tree technology, MuZero’s capabilities have achieved a major leap. In addition, MuZero also uses forward search and model-based planning to solve problems. Specifically, MuZero plans for three critical issues in the environment:

• Value: How good is your current position?
• Strategy: Which action is the best?
• Reward: How good is the result of the final step?

Atari launched the Atari 2600 game console in the United States in 1976. This was the first real home game console in history. Its entire life cycle lasted until 1992 and sold 30 million units.

(Source)