LogIn
The model learned to swim, hunt animals, eat food, and craft diamond tools.
OpenAI, known for its text-to-image AI tool DALL-E, has taught a neural network to play Minecraft with Video PreTraining (VPT). The researchers used a massive unlabeled video dataset of human Minecraft play and a small amount of labeled contractor data.
The model can learn to craft diamond tools, which usually takes proficient humans over 20 minutes (24,000 actions). It uses the native human interface of keypresses and mouse movements and "represents a step towards general computer-using agents."
In particular, OpenAI used the data to train an inverse dynamics model (IDM), which predicts the action being taken at each step in the video. It can use past and future information to guess the action at each step. The team then used the trained IDM to label a much larger dataset of online videos and learn to act via behavioral cloning.
Trained on 70,000 hours of IDM-labeled online video, the behavioral cloning model learned to chop down trees to collect logs, craft those logs into planks, and then craft those planks into a crafting table; according to the paper, this sequence takes a human proficient in Minecraft approximately 50 seconds or 1,000 consecutive game actions.
"Additionally, the model performs other complex skills humans often do in the game, such as swimming, hunting animals for food, and eating that food. It also learned the skill of “pillar jumping”, a common behavior in Minecraft of elevating yourself by repeatedly jumping and placing a block underneath yourself."
Overall, the company thinks VPT "paves the path toward allowing agents to learn to act by watching the vast numbers of videos on the internet." The open-ended nature of Minecraft led the researchers to believe the method can be used in other similar domains, e.g. computer usage.
Learn more about the research here and don't forget to join our new Reddit page, our new Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.