DeepMind has developed a way for AI programs to learn how to team up effectively in a simple video game. Most strikingly, the AI agents can also collaborate with human players—and those players say the programs are better teammates than most people.
Team players: Researchers at DeepMind trained teams of AI agents to play a game of capture the flag in a modified version of the first-person shooter Quake III Arena. Teamwork is extremely difficult to develop effectively in AI programs, because it involves dealing with a complex and ever-changing situation.
Winning formula: The researchers used an algorithm, dubbed “For the Win,” that trains a host of agents in parallel using reinforcement learning, a machine-learning technique modeled on the way animals learn. A few tricks helped optimize and tune the process.
Game changing: It’s a promising advance, because AI programs will need to work nicely with each other. But it’s important to note that the world in which these programs operate is incredibly simple. Demonstrating teamwork in the real world will be a lot more challenging—and achieving that is likely to be a long way off.