Introducing AI that automatically plays Minecraft and crafts diamond items at explosive speed

OpenAI , which has developed many AIs with amazing abilities such as AI that can write natural sentences that are indistinguishable from humans and AI that automatically describes the continuation of source code, 'Minecraft diamond items within 10 minutes AI that can be crafted 'was announced. The announced AI uses a method called ' Video PreTraining (VPT) ' to streamline machine learning using movies.

Learning to Play Minecraft with Video PreTraining (VPT)

Minecraft can collect materials such as stones and iron to craft items. When creating diamond items, it is necessary to perform operations such as 'craft items for mining diamond ore' and 'dig deep underground in search of diamond ore', and even experienced players can create diamond items from the initial state. It takes tens of minutes to craft. OpenAI has developed an AI that allows AI to learn a huge number of Minecraft play movies that exist on the net and craft diamond items in about 10 minutes.

A technique called VPT was used to learn AI. In general, AI learning uses a large amount of data manually labeled by humans. However, this time, we prepared a relatively short-time Minecraft play movie with labels such as 'Which key was pressed' and 'How did you move the mouse' and let AI learn 'Correspondence between keys and mouse and play movie'. After learning, I learned a huge number of unlabeled play movies.

After the learning was completed, the AI was able to play 'attack the tree and get the log → process the log into wood → craft the workbench from the wood' at the time of unadjustment. Furthermore, it was possible to perform actions such as 'swimming in the water,' 'hunting animals,' 'eating food,' and 'stacking materials while jumping and moving upward.'

After that, the AI operation is optimized by adjusting the labeled data to be learned first, and rewards are set for tasks such as 'creating an iron pickaxe' and 'discovering diamond ore' necessary for creating diamond items. I advanced the learning of AI.

As a result, we have completed an AI that can create diamond helmets and pickaxes in about 10 minutes from the start of the game. The process of creating diamond items in Minecraft consists of keyboard and mouse operations. For this reason, OpenAI states that the know-how used for this AI learning can also be applied to AI learning, which is responsible for general computer operations.

You can see how AI creates diamond items in the movie placed at the bottom of the link below.

Learning to Play Minecraft with Video PreTraining (VPT)

in Software,   Game, Posted by log1o_hf