DeepMind's Latest AI Gamers Defeat Human Players in Quake III

After defeating human players in different games like Alpha Go, chess, Hanabi, Dota II, and StarCraft II, Google-owned AI company DeepMind continues to develop AI agents that can dominate the gaming world. This time, the company created a system that uses AI gamers to play Capture the Flag in Quake III Arena.

The AI agents were reportedly trained with 450,000 rounds of Capture the Flag. The data was equivalent to four years of gameplay that DeepMind’s AI system was able to learn in just a few weeks.

In a paper published by the DeepMind researchers in the journal Science, they reported:

“We used a two-tier optimization process in which a population of independent RL agents are trained concurrently from thousands of parallel matches on randomly generated environments. Each agent learns its own internal reward signal and rich representation of the world. These results indicate the great potential of multiagent reinforcement learning for artificial intelligence research.”

AI Gamers Defeat Human Players in Capture the Flag

The AI gamers learned how to play the Capture the Flag mode of Quake III through reinforcement learning (RL). The AI agents played the game over and over in randomly generated environments.

With every game it played, the system picked pieces of information about how it works, including strategies and techniques. Eventually, the AI agents were pitted against human players and defeated them.

According to the researchers’ paper, the AI agents were able to exceed the win-rate of human players in new maps or environments neither of the two parties had seen before. In fact, human players were only able to win when partnered with an AI agent. The team wrote:

“Our work combines techniques to train agents that can achieve human-level performance at previously insurmountable tasks. When trained in a sufficiently rich multiagent world, complex and surprising high-level intelligent artificial behavior emerged.”

The DeepMind researchers believe that their study can help train other similar AI systems to perform real-world tasks beneficial to humans.