OpenAI, which attracts attention with Dall-E, which transforms written sentences into visuals, has many artificial intelligence and machine learning-based projects. In one of these projects, the AI learned to play Minecraft by watching gameplay videos online.
Although it may not seem like a big deal at first glance that artificial intelligence can build something in a sandbox game that has been released almost 10 years ago, it actually presents us with some important facts.
Can AI play Minecraft?
Minecraft is a game that can trigger people’s creativity in very interesting ways. There are people who can build a computer within the game and play Minecraft on that computer. This shows how the game offers us an unlimited world. OpenAI AI also uses the videos of the players to train its AI to take advantage of this freedom and the freedom that the game brings. The result is an AI that learns to play Minecraft and succeeds. Trained with 70,000 hours of gameplay videos, artificial intelligence actually shows how well machine learning has come in terms of observing and imitating.
Under normal conditions, being able to train artificial intelligence under supervision requires a very serious workforce. This workforce is used to tag structures in videos and images and help AI understand what’s what. Video Pre-Training (VPT), the method used by the OpenAI team to overcome this problem, helps to significantly reduce this workload. Tagging 2000 hours of video, the OpenAI team makes the process faster and easier by having another artificial intelligence tag 30,000 hours of video.
This study shows how effective training artificial intelligence with videos can be. Videos on platforms such as YouTube pave the way for customizing artificial intelligence for thousands of different studies.
17715
