[ad_1]
DeepMind, the company behind Alphabet's artificial intelligence systems, the company best known for its search engine Google seeks to perfect its developments in the hands video games with virtual participants ] that improve their skills by playing the title leather games in the first person Quake.
In this way, the Artificial Intelligence Company of Alphabet was able to learn to participate in a game with multiple human players ] in Quake 3 Arena's Capture Flag mode. Without knowing the rules, participants in DeepMind's artificial intelligence system team, called agents, worked independently while learning to interact with each other.
The use of games as a training tool for Google's artificial intelligence systems This is not new Two years ago, DeepMind IA beat the champion of Europe of Go, a millennial game of Chinese strategy. Later, Alphabet's subsidiary set out to develop a system capable of learning for five years the strategies of the StarCraft video game in order to defeat a human
— A View of the Agents' Operation in Quake 3 – –
In this case, the rules of this modality of Quake 3 Arena consist of a scenario with two teams with opposing goals: one side must defend a flag that another group of players must tear off . With this premise, the team of DeepMind managed to have a performance similar to that of humans .
DeepMind used a modified version of Quake to train and test its agents. In this way, the artificial intelligence system has managed to evaluate their skills in strategy, tactics and team play, points where this type of development seeks to improve
" Acting independently and working as a team is a huge challenge: agents must have the ability to adapt in an environment of constant change, "said the company in its official blog
After several games the agents managed to understand the dynamics of the game independently to the point of exceeding the number of victories achieved by teams made up of human players.
Source : The Nación of Argentina-GDA
Source link
