Google's DeepMind trains AI agents to play Quake III Arena like humans



[ad_1]

  DeepMind AI Agents May Pwn Some N00bs

DeepMind AI Agents May Pwn Some N00bs

GOOGLE & # 39; s UK division of artificial intelligence (AI) DeepMind showed that he can play the frenetic multiplayer game Quake III Arena as a human.

Use reinforcement training, whereby AI learn to play the game of the pursuit of some prisons based on the receipt of a reward signal, DeepMind has formed a team of agents AI to work together and learn general strategies to play a flag-catching game. The idea was that the AI ​​agents play the game like humans, which they managed to do by adopting strategies such as camping the entrances of enemy bases and following their teammates rather than to go alone – We know a lot of human players who can not even handle that in games.

"From a multi-agent perspective, the CTF requires that players cooperate successfully with their teammates and DeepMind's teams explained:" To make things even more interesting, we consider a variant of CTF in which the layout As a result, our agents are forced to acquire general strategies rather than memorizing the layout of the map. game, our learning agents experience the CTF world in the same way as humans: they observe a flow of pixel images and emit actions via an emulated game controller. "

[19659004] The result of this effort was a team of AI agents who learned from nothing how to see, act, cooperate and compete in invisible environments, all while pursuing a single signal of reinforcement. While their team won or was defeated

trained, the agents were put in mixed teams of humans and AI fighting together and against each other. The agents, dubbed FTW (for victory) had a better victory rate than humans and even better together.

Does this ultimately mean that intelligent robots will eventually work together to enslave us all? We can not be sure yet, but they will probably improve by playing computer games.

The smart people of DeepMind did not explain how this AI research could be applied to the real world, b However, it highlights what can be done with reinforcement training and systems. IA. μ

[ad_2]
Source link