– ArticleSet.com Google DeepMind – KI Wins Catching Flag Against People in Quake 3 Arena



[ad_1]

  The Google AI DeepMind has managed to master Quake 3 Arena's catch-the-flag mode so well that it has been able to beat a human team. The Google AI DeepMind By training with themselves, they mastered Quake 3 Arena's catch-the-flag mode so well that they could beat a human team.

Following the great success of the IA AI AI League of Legends Project against the human players are followed by Google with its own DeepMind AI: The search team managed to train the AI ​​to beat a human team in the capture-the-flag mode of the Quake Arena 3.

Like TheVerge reported, AI developers rely on the so-called reinforcement of learning . Here, AI trains on some sort of trial and error principle and plays against itself until it has understood the game as much as as possible. DeepMind has passed nearly half a million games, each lasting five minutes, or about 41,667 hours (or 1736 days) of pure workout.

Game Mode under Windows 10 – Best Performance in Game Mode?

During training, AI learns all elements of Capture the Flag at Quake 3, including capture and flag guarding and encampment at the base of the enemy. During the training, however, the AI ​​did not rely on digital data – as in the case of Open AI and LoL – but received the game information exclusively on the Quake-3 user interface.

After completing the training, AI developers had DeepMind competing against two-player teams, including teams of robots, humans and a combination of both. As a result, the pure bot team scored the best with a 74 percent win rate – the (strong) human team only reached 52 percent.

AI Apprentice for Civilization – AI Revolution from Germany

However, the victory rate of AI has decreased the most players in the AI ​​team. For four artificial intelligence players, the win rate was only 65% ​​- so the DeepMind team work still has to work a little bit. However, the sighting factor, which is also very important in a shooter like Quake in addition to the tactics, is not discussed in The Verge's report.

Programming a bot to hit the target is likely to be much easier than to teach him significant tactical behavior – who has already fought the last opponent bot Xaero in the single-player mode of Quake 3 Arena on the level of difficulty 'Nightmare', can sing a song.



  20 Years GameStar: Retro PC - We build a high-end gaming PC from 1997.






16:58




20 years GameStar: Retro PC – We build a high-end gaming PC from 1997.

[ad_2]
Source link