Notes

I Learned to Play Minecraft and it's actually Pretty Good

Chess Go, Chess, and now-Minecraft. Artificial intelligence models have added a brand new victory to their gaming kill list.



- Off - English


Utilizing an unlabeled mix of Minecraft videos as well as a tiny dataset of ones which were labeled by contractors intelligence company OpenAI was able to train a neural network to competently play Minecraft. This is a major milestone for the technology, which was previously struggling to master the game's simple, but tense gameplay. Open AI engineers released their experiment in an article and blog post this week.



The model of OpenAI could perform more than basic survival and crafting. It could perform many of the same complex tasks as a human Minecraft player. In its blog post, OpenAI shows a video of its model swimming, hunting, and cooking animals. It even successfully figured out the game's "pillar jumping" technique. Deepmind was also in a position to train its MuZero AI to play Atari Games.



To beat traditional games such as Chess and Go, previous AI models relied on reinforcement learning. Minecraft however is a game that can be understood by children of all ages but poses a challenge for AI systems due to its open-world design and open-ended structure.



There are plenty of videos on the internet that discuss Minecraft gameplay. But those videos only tell a small portion of the story of how an AI learns how to play the game. OpenAI claims that the video data, which is not labeled, does a good job demonstrating "what" but doesn't give precise key presses or mouse moments necessary for an AI understanding "how" to play.
LOG DESIGN


The engineers tackled this "how" problem by creating an imitation learning technique that is semi-supervised they call "Video PreTraining," or VPT. OpenAI essentially gathered a smaller and more compact dataset of contractors that contained not only Minecraft gameplay, but also videos of key presses and other actions documented by contractors. OpenAI created an additional model that utilizes contractors' videos to predict the next step in every step of the Minecraft film. Armed with the fundamental knowledge the AI was capable of understanding large amounts of Minecraft videos on the internet. Instead of simply dumping the data into their AI, the engineers took the time to teach it the fundamentals of inputs basics first.



OpenAI declared that their models are able to perform many tasks at human-level performance. They were also the first to report computer agents capable of crafting diamond tools. This could take skilled humans up to 20 minutes (24,000 environment actions) to complete.



All that training and support for contractors was reportedly a cost that was around $160,000. Most of that cash according to ZDNet was spent on paying out the contractors who collectively completed around 4,500 hours gaming. The contractors were paid $20 per hour.



Below you will see the AI cutting wood, directing its inventory, and exploring caves.



If watching an AI worth the annual salary of surgeons playing an 11-year-old indie game isn't exactly impressive, it's worth taking a look at how far technology has come. Three years ago it was the MineRL competition gave teams of technologists with the task of developing an AI that could mine a diamond in Minecraft. There were 660 contestants who tried to meet this challenge and every last one of them failed. OpenAI's model can now craft diamond tools.



OpenAI isn't the sole tech company that is turning to Minecraft for its AI experiments. In October, during its Build conference, Microsoft revealed a new AI Minecraft "agent" that is part of the game. When communicating with Microsoft Minecraft agents, users can type commands into the game, which are automatically generated by the game's API. Wired informs us that players can type phrases like "come here" into the Minecraft bot and it will translate it into Minecraft code. The bot then walks forward. Microsoft's Minecraft agent is able to do more than simply walk. It also can retrieve items from the game world and use them to create something. And , look at it, it is likely to do it better and quicker than this writer, who's many years removed from his last Minecraft session.


My Website: https://log-design.com/