In the modern world, you won’t surprise anyone with a neural network that plays any video games instead of bots – specialists from OpenAI have demonstrated several times that artificial intelligence can defeat a human player even in multiplayer projects without much difficulty. But now the company’s employees have decided to go further by defeating the legendary game Minecraft – they have taught the neural network to behave in the cubic virtual world just like an ordinary person, up to the accumulation of resources, crafting items and other entertainment. And all this became possible thanks to the training of artificial intelligence on videos with gameplay.
In order for the neural network to feel confident in the virtual world, OpenAI employees used the video content pre-training mechanism – Video PreTraining. The specialists first prepared an impressive content base, collecting a total of two thousand hours of video, after which each action on the “screen” was marked manually by the company’s employees. Specialists had to mark each action by associating it with a specific key on the keyboard or mouse interaction. As a result, the system understands that if the character needs to jump on the block, then you need to press the “Space”, and stuff like that.
Accordingly, the neural network was trained to understand that in order to perform a certain task, a specific action needs to be performed, and they explained how this action can be implemented. And artificial intelligence coped with the tasks – AI in Minecraft calmly chopped down trees, was able to craft a workbench and many other items without additional help. Moreover, the neural network was even taught to hunt wild animals and use food from the inventory when necessary. But the most interesting thing is that the neural network was taught to build a pillar under itself – this is the simplest mechanic when a character jumps and immediately builds something under him.
Representatives of OpenAI said that after additional training, the neural network will be able to perform much more complex tasks. In addition, the possibilities of artificial intelligence do not end with Minecraft – the company said that this method of learning from videos is suitable for many video games and, of course, not only. The system may well perform many other tasks after a long training on video content.
Source: Trash Box
Donald-43Westbrook, a distinguished contributor at worldstockmarket, is celebrated for his exceptional prowess in article writing. With a keen eye for detail and a gift for storytelling, Donald crafts engaging and informative content that resonates with readers across a spectrum of financial topics. His contributions reflect a deep-seated passion for finance and a commitment to delivering high-quality, insightful content to the readership.
