AI learns to play Minecraft by watching YouTube videos

Robert Collins

Looking for more info on AI, Bing Chat, Chat GPT, or Microsoft's Copilots? Check out our AI / Copilot page for the latest builds from all the channels, information on the program, links, and more!

OpenAI’s latest artificial intelligence project has apparently learned to play the video game Minecraft. This is according to a post on the company’s official blog, which details how a neural network was trained to accomplish the feat.

OpenAI is an organization dedicated to artificial intelligence research, and was founded by Elon Musk in 2015. Apparently the researchers used a technique called “video pretraining” (VPT) that involved gathering sample data from videos of actual human gameplay such as button presses and mouse movements, and created an algorithm that annotated these actions.

Researchers then trained an “inverse dynamics model” (IDM) to predict the sequence of actions being taken in the gameplay videos. Finally, the trained IDM was shown 70,000 hours of online video.

The model was able to copy actions from the videos. It was able to collect logs, craft logs into planks, and then craft planks into a crafting table. The model could also swim, hunt animals and eat food. It could, in fact, even accomplish the rather complex feat of “pillar jumping” – repeatedly jumping and placing blocks underneath the player.

The AI even managed to craft a diamond pickaxe, a complicated skill that takes a human player around 20 minutes to accomplish.

So what do you think of the artificial intelligence that can play Minecraft? Do you find it fascinating, disconcerting, or both? Let us know in the comments.