DeepMind's AI system, Dreamer, has achieved a remarkable milestone by learning to collect diamonds in Minecraft without any prior instructions, showcasing a significant advancement in AI capabilities.

Minecraft, which boasts over 300 million copies sold and more than 200 million active players, features a procedurally generated world that presents unique challenges for AI.

The success of Dreamer is attributed to its ability to create a 'world model' that aids in decision-making by predicting the rewards of actions with reduced computational demand.

This approach allowed Dreamer to develop a mental model of its surroundings in Minecraft, enabling it to simulate scenarios before executing actions.

Danijar Hafner, a computer scientist at Google DeepMind, believes that Dreamer's ability to 'imagine the future' could lead to more intelligent robot interactions in real-world applications.

Hafner suggests that the techniques developed for Dreamer could be applied to real-world robotics, where the costs of trial and error are significantly higher than in a video game.

The findings of this research were published in the journal Nature on April 2, 2025, highlighting the potential of model-based reinforcement learning for various applications.

Georg Martius from the Max Planck Institute praised the study as groundbreaking, emphasizing the wide-ranging applications of model-based reinforcement learning from video games to robotic control.

The process of finding diamonds in Minecraft requires multiple steps, including crafting tools and navigating hazards like lava, which adds complexity to the task.

After nine days of training, Dreamer learned to mine diamonds in under 30 minutes, matching the performance of a skilled human player who typically takes 20 to 30 minutes.

Dreamer received rewards for each step completed towards finding diamonds, which accelerated its learning process through reinforcement learning.

Unlike previous AI attempts that relied on human gameplay videos or guided training, Dreamer uses reinforcement learning to explore and learn independently from its environment.