OpenAI has developed a neural network that can play Minecraft like humans. The Artificial Intelligence (AI) model was trained over 70,000 hours of miscellaneous in-game footage, along with a small database of videos in which specific in-game tasks were performed. Keyboard and mouse inputs are also recorded.
![Microsoft Holds Its Xbox 2016 Briefing During Annual E3 Gaming Conference Microsoft Holds Its Xbox 2016 Briefing During Annual E3 Gaming Conference](https://1734811051.rsc.cdn77.org/data/images/full/407241/microsoft-holds-its-xbox-2016-briefing-during-annual-e3-gaming-conference.jpg?w=820)
(Photo : REDERIC J. BROWN/AFP via Getty Images)
Lydia Winters (R) and Sax Persson speak about Minecraft during Microsoft's E3 2016 Xbox press conference at the Galen Center in Los Angeles, California on June 13, 2016, where the latest titles for Xbox were introduced one day ahead of the start of the 2016 Electronic Entertainment Expo, or E3, at the LA Convention Center.
OpenAI fine-tuned the AI, and now, it is skillful as a human-it can swim, hunt for animals, and eat. The AI can also do the pillar jump, where a player places a block of material below themselves in mid-air to gain more elevation.
One of the most impressive things that it can do is craft diamond tools.
Also Read: Minecraft 2022: How to Check Where the Screenshots are Saved
Video PreTraining (VPT)
This Minecraft project shows the efficacy of the new OpenAI technique, the Video PreTraining (VPT). VPT can help boost the development of general computer-using agents.
With the VPT, OpenAI can pain a large video dataset drawn from public web sources with relevant keyboard and mouse movements to establish the foundational model.
Fine tuning the base model involves plugging in smaller datasets to teach specific tasks. OpenAI used players' footages that perform early-game actions, like cutting down trees and building crafting tables.
OpenAI also implemented rewarding the AI model for achieving each step on a sequence of tasks, which is also known as reinforcement learning. Through this, the neural network can collect all ingredients for a diamond pickaxe at a human level.
In a blog post, OpenAI said that the VPT allows agents to learn to act by watching a lot of videos on the internet. Therefore, it creates an exciting possibility of learning large-scale behavioral priors that go beyond just language.
The reason they experimented with Minecraft is because its open-ended and the human interface is very generic.
Still, OpenAI will further experiment and has partnered with the MineRL NeurIPS competition. They donated their contractor data and model code to contestants who will attempt to use AI to solve complex Minecraft tasks. The winner will take home $100,000.
AI in Gaming
Today, AI is being used in gaming in many ways, and it can perform a diverse set of actions in games.
For example, in MMORPG games like World of Warcraft, AI-controlled players engage in a war against each other. Currently, this is limited to simple actions like casting spells, using special abilities.
In these games, AI players can fully take on other players, and they can develop through experience, meaning they get better and better with time. This is a far cry from the AI employed by game bots, which are able to play the game, but at an extremely basic level.
Additionally, the AI in games can also be used to study machine learning algorithms and neural networks. Games are perfect for creating an environment where we can analyze an AI's behavior and how it reacts under certain conditions and in certain scenarios.
AI has been a hot topic this year and it will be an important part of the future of gaming. Developers are already implementing AI into their games and advancements will be made. However, AI will never replace players. Instead, it will make games more immersive, engaging, and fun to play.
Related Article: AI to Improve with Vision and Language-New OpenAI's DALL-E 2 is One Promising Technology
This article is owned by TechTimes
Written by April Fowell