[ad_1]
Google DeepMind today shared the results of research and experiments in which several AI systems were trained to play Capture the Flag on Quake III Arena, a first-person shooter multiplayer. A AI trained in the process is now better than most human players in the game, as she plays with a human or mechanical teammate.
The IA, named For the Win (FTW), played nearly 450,000 games of Quake III Arena to gain dominance over human players and establish his understanding of how to work effectively with d & # 39; other machines and humans. DeepMind refers to the practice of training multiple independent agents to conduct collective action as multi-agent learning.
"We train agents who learn and act as individuals, but who must be able to play with and against other agents, artificial or human," said the company today. in a blog. "In a multi-agent perspective, [Capture the Flag] requires players to successfully cooperate with their teammates and compete with the opposing team, while remaining steadfast in any style of play that they might encounter."
DeepMind is perhaps best known the creator of AlphaGo, an AI system that beat the world's best Go player in May 2017. AlphaGo Zero, a descendant of AlphaGo, has then was made to improve by playing games against himself.
The DeepMind experiment involved 30 agents simultaneously playing four at a time against humans or machines.
In a tournament with and against 40 humans Capture the flag players, machine teams only remained undefeated in the games against exclusively human teams and had a 95 percent chance of winning against teams in which humans played with a machine partner.
Machine crews captured 16 flags per game less than a team of two FTW agents.
Agents were found to be effective in tagging by humans, reaching tactics 80% of the time versus 48% for humans. FTW continued to keep its advantage over human players even when its tagging abilities were removed at levels comparable to those of humans.
Interestingly, a survey of human participants found FTW more cooperative than his human teammates. The founder and CEO of DeepMind, Demis Hassabis
The search was performed with unique challenges
Capture the Flag was played in environments with random map layouts rather than a static and consistent environment so to train the systems. for better results. Indoor environments with flat terrain and outdoor environments with varying altitudes were also introduced. Agents also functioned in slow or fast mode and developed their own internal reward systems
The only signal used to teach agents was whether their team won the game by capturing the most flags within five minutes.
No rules of the game were given to the machines before, but over time, FTW learned basic strategies like basic defense, a teammate or a camp in the base of a opponent to mark them. Tagging, the act of touching an opponent to send it back to its point of fraternization, has been incorporated into the tactics used to win the matches.
DeepMind is the most recent study of researchers on AI. to form a machine strategy, memory, or other features common in humans but that do not occur naturally with computers.
Last month, OpenAI revealed that it was using reinforcement learning to train the AI to beat talented human teams playing Dota 2. [19659002Theideasthatcanbelearnedfrommulti-agentenvironmentscanbeusedtoinformtheteam;man-machineinteractionandformatheIAsystemstocompleteorworktogether
SRI International, for example, as part of DARPA The Lifelong-Learning Machines training program consists of training AI systems to play the StarCraft role-playing game: Remastered in order to train them to act collectively and to act in groups or groups as the characters in the game do.
also found a lot of value in StarCraft. In August, DeepMind announced the release of the StarCraft II API for reinforcement learning as part of a partnership with Blizzard.
Source link
