AI has learned to play Minecraft like humans! Artificial intelligence or AI researchers have taught a neural network to play Minecraft. And it is now playing such a good game that its performance is compared with human gamers. Researchers at AI research firm OpenAI have shown the neural network how to play games by showing 70,000 hours of footage of Minecraft.
In addition, the Neural Network has trained contract gamers to perform specific complex tasks inside the game from a small database of videos. In addition, the gamer’s mouse and keyboard input information were also in the database.
At the end of the training, AIT can now perform all kinds of complex tasks inside the game, according to technology site TechRadar. Swimming inside the game, hunting, eating the flesh of prey, AIT can do it all.
That AI can even do ‘pillar jump’. Pillar jumps are known as ‘High Skill Maneuvers’ among Minecraft gamers. After jumping in it, the gamer must put a pillar or block under his feet in the middle. As a result, the gamer can jump higher.
However, AIT’s most significant achievement is its ability to make tools with diamonds inside the game. The gamer has to follow many complex steps inside the game. OpenAI calls this capability of neural networks an “unimaginable achievement” for a computer.
Also Like: Google AI is Fighting with People
The Breakthrough of AI Technology?
The success of the neural network, says TechReader, demonstrates the effectiveness of OpenAI’s new strategy, video pretraining (VPT), in training AI models. OpenAI says this training strategy will further accelerate the construction of general computer-centric AI technology.
The main obstacle in training AI by showing the video was – although it is easy to understand what is happening by watching the video, it is not easy to understand how the events are happening. As a result, even if it understood what was being asked, AI did not know how to do it.
In addition to displaying gameplay videos collected from the Internet at VPT, OpenAI also offers the opportunity to learn by watching AIT with keyboard commands and mouse movement data.
The research team then integrated the neural network with a smaller database to provide more sophisticated training. That database was organized to teach neural network-specific complex tasks. AIT has seen a video of gamers cutting down trees and making tables inside the game. OpenAI researchers claim that AIT made more reliable and significant progress by learning those small things.
In a blog post, OpenAI said, “Although we’ve only tested Minecraft, the game is open-ended and the use of the two main interfaces used by gamers – the mouse and the keyboard – is very common here. So we think it’s good news for the same kind of job as running a computer (AI).”
Read More: Sony AI is Losing the Best Gamers