Last year, the non-profit organization OpenAI pitted its artificial intelligence system against Dota 2 professionals. And then the machine was unable to outperform humans. Now the system has taken revenge.
The OpenAI Five Championship took place in San Francisco over the weekend, during which the AI ββmet with five e-sportsmen from the OG team. This team took the highest award in eSports in 2018, taking first place in The International Dota 2 tournament with a prize fund of $25 million. Team members met with OpenAI bots, who were trained using the same methodology. And the people lost.
OpenAI bots are reported to have learned reinforcement and independently from each other. That is, they got into the game without prior programming and settings and were forced to learn by trial and error. OpenAI co-founder and chairman Greg Brockman said that in 10 months of its existence, artificial intelligence has already played 45 thousand years of Dota 2 gameplay.
As for the game itself in San Francisco, each squad had 17 heroes to choose from (there are more than a hundred of them in the game). At the same time, the AI ββchose a mode in which each team can prohibit the selection of those heroes that it chose. This allows you to build on your strengths and minimize your weaknesses. Illusions and the functions of summoning new heroes were also disabled, although it was possible to resurrect the fallen.
The AI ββwas reported to have used tactics that resulted in short-term gains, but they paid off. At the same time, the system revived dead heroes even at the beginning of the battle. In general, the machine used a very aggressive approach, a kind of βblitzkriegβ, which people were unable to repel, since the first match lasted only half an hour.
The second one was even shorter, as the AI ββdestroyed the humans very quickly, focusing on attack rather than defense. In general, it turned out that the reinforcement learning scheme gives results. This will allow it to be used in the future for various tasks.
Source: 3dnews.ru