cool hit counter AI can also team up with each other to play Dota 2 to beat humans_Intefrankly

AI can also team up with each other to play Dota 2 to beat humans

Recently, OpenAI, the AI non-profit organization founded by Musk and Sam Altman, announced that the AI it developed was able to team up to beat top Dota 2 amateur players in 5V5 matches, and that these human players had an average ladder score of over 4,200. This is another milestone event after Alpha Dog vs. Kojie and the AI challenge of Starcraft. Because, unlike Go and Starcraft, Dota 2 is a team game that requires cooperation between different players.

it is reported that OpenAI has developed a set of algorithms called "OpenAI Five". While in isolation this algorithm is not groundbreaking, just a kind of neural network for playing Dota 2, and there have been previous AI algorithms that have beaten human players in 1-on-1 Dota 2 matches, theThe importance of this AI victory is that the 5 sets of algorithms learned to collaborate with each other and their teammates between them.

This time, the AI is facing off not against top pros but OpenAI employees, and the AI will only use 5 of the 115 available heroes in the game, and there are restrictions on human players from using props like the True or False Eye and Magic Bottle, but that's just the beginning for the AI.

After the showdown, Greg Brockman, one of the founders of OpenAI, said : "We could sense from the competition that the intention to collaborate between AI algorithms seemed to be a very natural and instinctive expression. And, in a 5V5 team match in Dota 2, we also tried replacing one of the five AI players with a human player, and the human player reported being able to appreciate the support he received from his AI teammates. "

In addition, OpenAI creates artificial intelligence using also machine learning, a technique of reinforcement learning. While it may seem simple, its ability to allow AI to learn complex behaviors. And this process of allowing the AI to reinforce learning is putting the AI in a virtual environment and learning to achieve the goal through the AI's own attempts. These AI bots, after a lot of training, accumulate 100 times more experience per day than a person can accumulate in a lifetime.

It is understood that MOBA (multiplayer online competitive game) games like Dota 2 have very complex game mechanics in themselves and require perfect cooperation between teammates to achieve the ultimate victory. The victory between these 5 AI players proves that artificial intelligence can work with each other.

Therefore, if we extend the cooperation between multiple AIs in the game to AI technology commercialization application scenarios, we can think of such multiple AI algorithms that can be applied to scenarios such as online trading, advertising bidding ranking, and diversified manufacturing on production lines. In addition to this, AI with teamwork and humans can also cooperate.

In the future, OpenAI's current AI development will compete in the top Dota2 tournament The International this fall, going up against a battle jedi of five professional players.


1、Explanation of Constructor Methods in Swift primary
2、iOS Development Calling the System Phone Function
3、translated How mobile game developers are improving their strategies for breaking into the market
4、AnalysisLi Qiang Industry 40 is not just for industrial companies
5、BizTalkadditional modify removingXmlDocument An efficient way to name space

    已推荐到看一看 和朋友分享想法
    最多200字,当前共 发送