[ad_1]
Google DeepMind Artificial Intelligence could learn to collaborate to play the Quake III Arena, but the results are not just about the game itself
The team behind the DeepMind project has released new progress results made by artificial intelligence. Within the project, the latest progress is related to the Quake III Arena, a multiplayer FPS. Within it, several artificial intelligence systems have learned to play Capture the Flag. One of the key results is that such an AI is now better than most human players.
The Google AI, called For Victory (FTW), has played about 450,000 Quake III Arena games to achieve the skills it currently has. The system has learned not only how to dominate the opponents, but also how to work effectively with other artificial intelligentsia or even humans.
In a championship with 40 human players, teams created from artificial intelligence have managed to stay undefeated in matches with human opponents. The odds of winning remain above 95% when people play with a computer.
Artificial intelligence did not receive the rules of the game, but the FTW learned in time strategies such as basic defense, tracking partners or camping in the opposing camp. The system has learned to do tagging, which involves touching an opponent to send him back to his own base.
The Quake III Arena was played with random card settings so the system learns how to proceed. win, regardless of the environment. In addition, the agents played both fast and slowly, but they also managed to create their own internal reward systems
[ad_2]
Source link
