Google’s DeepMind team has trained an AI called Dreamer to master Minecraft completely independently in only nine days. This is a significant feat because, unlike previous AI models which were trained using human gameplay footage, Dreamer managed to learn Minecraft without previous exposure to the game.

AI Model Masters Minecraft

Minecraft Bedrock Edition. Minecraft official website
Minecraft Bedrock Edition; Photo: Minecraft official website

The University of Toronto partnered with researchers at Google DeepMind to develop a unique reinforcement learning system for Dreamer. The system was designed to reward Dreamer for collecting diamonds – one of the game’s most valuable resources.

Rather than step-by-step instructions, the AI was simply encouraged to explore and improve its strategy through trial and error. During training, researchers reset the Minecraft world every 30 minutes to force Dreamer to adapt to a new, randomly generated environment.

Dreamer proved that it was able to adapt well and, by the end of the experiment, it was able to mine diamonds in under 30 minutes. This means the AI model was able to match human players after only nine days of gameplay.

Google DeepMind researcher Danijar Hafner explained, “Dreamer marks a significant step towards general AI systems.”

Part of Dreamer’s success stems from its ability to create a mental model of its environment, allowing it to simulate different actions prior to trying them. This allowed Dreamer to more quickly find the most efficient path to its goal.

Interestingly, this AI model’s learning process mimics how humans learn. Rather than relying on massive datasets of human input like traditional AI models, Dreamer learned by gathering information from its surroundings, distinguishing between successful and unsuccessful actions, and moving forward with the most effective strategies.

There are also real-world applications of this research as an AI system that is capable of learning and improving without human guidance could be used in automation, robotics, and problem-solving tasks in various industries.

“Dreamer is, to our knowledge, the first algorithm to collect diamonds in Minecraft from scratch without human data or curricula,” the Google researchers stated in the study.