a.i. Players master Quake III Arena and Beat Humans

The challenge for DeepMind was not to see if its AI agents could defeat human players in battle, but rather they could work together on procedurally generated levels to accomplish a goal – in this case, capture the flag.

Because the structure of the levels changes every time they play, the agents simply memorize places in order to return them to the flag. This forced them to learn the strategies needed to win in the same way that human players could improve the game. The agents even used an emulated game controller to control their characters in Quake III Arena

"Agents are never aware of the rules of the game, but learn basic game concepts and effectively develop an intuition to capture the flag," DeepMind said in a blog, "In fact, we can find neurons." individuals that directly code for some of the most important game states, such as a neuron that will activate when the flag of the agent is activated or a neuron that will activate when the team is activated. An agent holds the flag 19659003] As AI agents have won matches, the behavior they used to do it has been strengthened and these agents have been paired with human players ns to impressive results. Finally, DeepMind held a tournament Quake III Arena and not only did these A.I. agents have a higher success rate than humans, they were also judged to be "more collaborative" in a post-tournament poll. Unfortunately, just as with so many kids on Xbox Live, the A.I. agents also learned to "camp". may be a legitimate strategy.