AI Learns to Play Tag (and breaks the game)

Two AI agents named Albert and Kai learn to play tag through reinforcement learning in a simulated environment. Starting with no knowledge of the game rules, they develop increasingly sophisticated strategies over weeks of training. Albert learns to dodge and eventually discovers ways to exploit the environment by jumping on displays and escaping through walls, while Kai becomes faster at tagging. The experiment progresses through multiple environments with added complexity like blocks and walls, leading to emergent behaviors where Albert learns to throw objects and cause chaos to win. After over 2 months of training, both agents become skilled players, with Albert ultimately winning a final 5v5 battle.