The newest artificial intelligence effort from OpenAI has apparently learned to play the video game Minecraft. The researchers employed a method known as "video pretraining" (VPT), which involves collecting sample data from movies of genuine human gameplay, including button presses and mouse movements, and developing an algorithm that labeled these activities.
The AI research and deployment business is sure that its model can learn to make diamond tools, which typically requires skilled humans over 20 minutes, with a little bit of fine-tuning (24,000 actions). Its paradigm, which is quite generic and takes a step toward general computer-using agents, employs the innate human interface of keypresses and mouse movements.
How was The AI Taught to play Minecraft?
Open AI proposes an innovative, yet straightforward, semi-supervised imitation learning technique called Video PreTraining (VPT) in order to make use of the abundance of unlabeled video data available on the internet (VPT). To start, the team collects a tiny dataset from contractors in which it records both their video and their behaviors, in this case, keystrokes and mouse movements. With the help of this information, the business can create an inverse dynamics model (IDM), which forecasts the actions that will be taken at each video step. By using "past and future information to guess the action at each step", the IDM enables the AI to navigate. The trained IDM then assigns mouse and keyboard motions to 70,000 hours of film.
According to Open AI, VPT paves the way for enabling agents to learn to act by watching the countless films available online.
Performance of The AI Project
According to OpenAI, its bot "performs tasks in 'Minecraft' that is almost impossible to achieve using reinforcement learning from scratch." This includes making a diamond pickaxe, which takes a competent human player 20 minutes on average to complete yet took the AI only 2.5 percent of the episodes it played that lasted 10 minutes. According to OpenAI, early training and improvement produced an AI that could "cut down trees to collect logs, craft those logs into planks, and then craft those planks into a crafting table." One of the game's simpler sequences of actions takes a player with a basic understanding of the game less than a minute to complete.
Although this bot isn't entirely autonomous, it does appear to be able to act in accordance with a goal it wants to accomplish. This development in thinking quickly could lead to some fairly significant opportunities, both good and bad.
More information you can read: https://openai.com/blog/vpt/