After AlphaGo, the artificial intelligence played the turn to defeat world champions Fan Hui, Lee Sedol and most recently Ke Jie, another artificial intelligence developed by Google DeepMind recently. 2 professional gamers Dario “TLO” Wünsch and Grzegorz “MaNa” Komincz in the game StarCraft II. It is worth mentioning that, in 11 matches that are streamed directly on YouTube (you can review it here), the AI wins in 10 consecutive matches, and the last match is the only one to win first. artificial intelligence.
To be fair, AI has more practical applications in life, science and research than playing games against people. But it is not easy to create an AI to win a human game. Unlike Go, StarCraft II is a strategy game with a different depth in gameplay. Just because AI can’t have a lot of time analyzing moves like playing chess, having to reflect and make real-time strategy choices makes it more difficult to create a AI for StarCraft II than for Go. . Along with that, instead of calculating a single move, StarCraft AI must pay attention to many other aspects, from resources, how to develop troops, and even macro units in battle.
In December, matches between AI named AlphaStar and Dario “TLO” Wünsch were held at DeepMind’s office in London, and he proved that AI can play games against people. The professional StarCraft commentators after watching the replay of the match said that AlphaStar has the ability to “superman”, and has amazingly precise macro skills for each unit.
Even when human opponents have more powerful units, DeepMind’s AI still knows how to take down those units. For example, in a match, AlphaStar attacked Maa’s main house with the Stalker of the Protoss, and the commentators said that if humans were to control these units, it would not be as smooth as AI. DeepMind’s development.
Previously, OpenAI of Elon Musk had DOTA 2 matches with 5 professional gamers. Although the victory of this game is for humans, AI knows how to learn and reflex very quickly and accurately in subsequent matches, something that people cannot do because of psychological causes. In other words, it is the “coldness” of AI that makes it extremely effective, and also what makes people who create deep learning systems feel anxious.
AI handling is too accurate and voyeuristic, so you may suspect that the player likes to click how many times once, while the limited power, APM (action per minute) depends on the muscle. can’t press the keyboard and click the mouse as fast as a robot. However, this capability is eliminated because AI developers in DeepMind do not allow AI to be faster than humans, meaning that APM is limited to a specific number. But not because of that, AI has an unfair advantage for professional gamers.