OpenAI successfully trained a Minecraft bot using 70,000 hours of gameplay videos

Why it matters: Minecraft may not seem like a significant tool that supports advanced AI research. After all, what could be so important in teaching a machine to play a sandbox game released over a decade ago? Based on recent OpenAI efforts, a well-trained Minecraft bot is more relevant to the advancement of AI than most people realize.

OpenAI has always focused on advancements in artificial intelligence (AI) and machine learning that benefit humanity. Recently, the company successfully trained a bot to play Minecraft using over 70,000 hours of gameplay videos. Realization is more than just a bot playing a game. It marks a giant leap forward in advanced machine learning using observation and imitation.

OpenAI’s bot is a great example of imitation learning (also called “supervised learning”) in action. Unlike reinforcement learning, where a learning agent is rewarded after achieving a goal through trial and error, imitation learning trains neural networks to perform specific tasks by watching humans perform them. In this case, OpenAI leveraged available gameplay videos and tutorials to teach its bot to perform complex in-game sequences that would require around 24,000 individual actions for the typical player.

Imitation learning requires video inputs to be labeled to provide the context of the action and the observed outcome. Unfortunately, this approach can be very laborious, resulting in limited available data sets. This dearth of available datasets ultimately limits the agent’s ability to learn by observation.

Rather than embark on an extensive manual data tagging exercise, the OpenAI research team used a specific approach, known as Video Pre-Training (VPT), ​​to dramatically increase the number of Tagged videos available. The researchers initially captured 2,000 hours of annotated Minecraft gameplay and used it to train an agent to associate specific actions with specific on-screen outcomes. The resulting model was then used to automatically generate labels for 70,000 hours of previously unlabeled Minecraft content readily available online, giving the Minecraft bot a much larger data set to examine and mimic.

The entire exercise proves the potential value of available video repositories, such as YouTube, as an AI training resource. Machine learning scientists could use available and appropriately labeled videos to train AI to perform specific tasks, ranging from simple web browsing to assisting users with real physical needs.

We wish to thank the author of this post for this outstanding material

OpenAI successfully trained a Minecraft bot using 70,000 hours of gameplay videos

Discover our social media profiles as well as other pages related to it.