Man vs. Machine: AI bots domination

May 31, 2019 | Media
The artificial intelligence of DeepMind not only beats people in chess, but now dominates for the first time in team games.

After AI systems have beaten the best human player in chess, they reach even a higher success rate in games with multiple team members.

Bot versus human: the framework

The researchers used the “Capture the flag” mode of the multiplayer shooter “Quake 3 Arena” released in 1999 and listed in this country. In this computer game, two teams have to grab the flag of each opponent, bring them into their base – while protecting their own flag. The symmetrically designed rooms and corridors should give both teams the same chances, they are generated randomly. The players move through the virtual terrain in the first person perspective. They need to work with their group members and keep their opponents at bay. A particular challenge for AI lies in the three-dimensional world – this is much more complex than that of board games like Go and Chess. But FTW coped well with it – as the results show.

Man against machine: the evaluation

After about 200,000 games, the AI ​​bots (also called agents) were on average better than the best people. With each game they learned and developed strategies that are similar to tactics of human players. After some 450,000 games, they finally won clearly: When two human players competed against two AI agents, the latter conquered an average of 16 more flags in a time frame of about five minutes. AI agents responded to the emergence of an adversary on average after 258 milliseconds, people after 559 milliseconds. But even if the researchers slowed down the response time of the AI ​​agents, the artificial ones remained superior to the human players.