Google’s DeepMind artificial intelligence lab surpassed another challenge with a computer program that was able to defeat human opponents in Quake III Arena‘s Capture the Flag mode.
This is not the first time that a DeepMind program proved to be capable of beating human players. In 2016, AlphaGo defeated Lee Sedol, the best Go player in the world, with a 4 to 1 score. Earlier this year, Google revealed that AlphaStar shut out two professional StarCraft players in a pair of five-game series.
DeepMind has now turned to Quake III Arena‘s Capture the Flag mode to exhibit its capabilities. In Capture the Flag, two multiplayer teams attempt to capture the flag of their opponents and bring it back to their home base to score, while also trying to prevent their opponents from doing the same by shooting them to make them drop the flag if they are carrying one. The game mode is a step up from previous tests due to its multiplayer nature, as it requires teamwork between A.I. agents.
DeepMind’s programs have shown success in environments with a single A.I. agent within a two-player game, such as Go and StarCraft. “However, the real world contains multiple agents, each learning and acting independently to cooperate and compete with other agents,” researchers wrote in a paper published in Science.
The DeepMind team created a program named For the Win, which was trained by playing thousands of games of Capture the Flag in Quake III Arena. In just a few weeks, the program was able to defeat human opponents, even if its reaction time was slowed down to not give it too much of an advantage. After 12 hours of practice, human game testers were only able to beat For the Win in 25% of games. Human players turned out to be better at long-distance shooting, but A.I. agents were more capable of navigating the play area to capture the flag.
One interesting finding is that when a human and an A.I. agent were paired, the team had a 5% greater win rate compared to a team of just A.I. agents. This suggests that the A.I. program is able to adapt to the human player, showing that there is indeed a benefit to humans working side-by-side with artificial intelligence.
- Chess. Jeopardy. Go. Why do we use games as a benchmark for A.I.?
- Why teaching robots to play hide-and-seek could be the key to next-gen A.I.
- This bot will destroy you at Pictionary. It’s also a huge milestone for A.I.
- This Google robot taught itself to walk, with no help whatsoever, in two hours
- Neuro-symbolic A.I. is the future of artificial intelligence. Here’s how it works