Happy Nice Mall

During practice, players first encounter simple single-player games such as finding a purple cube or placing a yellow ball on a red ground. They move on to more complex multiplayer games, such as hide and seek or capture the flag, where teams compete to find and capture their opponent’s flag first. The playground manager does not have a specific purpose, but aims to improve the overall capacity of the players over time.

Why is this cool? AIs like DeepMind’s AlphaZero have beaten the world’s best human players in chess and Go. However, they can only learn one game at a time. As DeepMind co-founder Shane Legg said when I spoke to him last year, Having to replace your chess brain with your Go brain every time you want to change games.

Researchers are now trying to create AIs that can learn multiple tasks at once, meaning teaching them general skills that make it easier to adapt.

CONTEMPLATION

An exciting trend in this direction is open-ended learning, where AIs are trained on many different tasks without a specific goal. In many ways, this is how humans and other animals seem to learn through aimless play. However, this requires a huge amount of data. XLand automatically generates this data in an endless stream of challenges. Similar POETis an AI training dojo where bipedal bots learn to navigate obstacles in a 2D landscape. But XLand’s world is much more complex and detailed.

XLand is also an example AI learns to make itselfor Jeff Clune, who helped develop POET and led a team work on this topic In OpenAI, it calls AI generating algorithms (AI-GAs). “This work pushes the boundaries of AI-GAs,” says Clune. “It’s so exciting to see.”