OpenAI, an artificial intelligence research institute, won a world championship e-sports team with an artificial intelligence system trained to practice complex strategy games Dota 2 . This contest was held in San Francisco today and ended Dota reclamation technology, which was called OpenAI Five Finals and has received much attention in the organization's public discussions.
Competition on the human side is the top five players of the OG team who won the world's most coveted e-sports awards last year at The International Dota 2 First premiere Dota 2 The contest awarded a total of $ 25 million. OG was trained using the same in-depth reinforcement learning techniques in all three competitions for the OpenAI Five bot and independently controlled at multiple layers of the same system. Reinforcement learning is actually a trial and error approach to self-improvement, and using the reward system and other incentive-granting mechanisms, AI is dropped into the gaming environment without understanding how the game works and is extensively trained.
Today's performance is a top-quality demonstration of the performance of OpenAI Five. The system lost much of its e-sports team, which lost two games in August last year. Greg Brockman, co-founder and chairman of OpenAI, is also the chief technology officer at the organization, but OpenAI Five improves by recycling himself in an accelerated virtual environment. "OpenAI Five has gained strength by strengthening learning, which means we have not coded the way we play, we have coded ways to learn," Brockman told the crowd ahead of the competition. "I have already played 45,000 Dota 2 gameplay in my 10 months of existence."
Dota 2 is a very complex strategy game with more than 100 unique characters. , A tree of deep skills and a list of items, various dizzying variables that appear on the screen at any moment of the match. Therefore, OpenAI imposes limitations on certain players when they play pro players. Most importantly, it counts the number of heroes used by five players.
In this case, each squad can choose from 17 heroes. OpenAI also opted for the so – called "Captain 's Draft" game mode, which prevented each team from strategically forbidding the hero to select that character before the other team used a different picking order. This allows the captain to build strengths between the hero combinations, and once the team starts building the roster, they can take advantage of enemy hero weakness through powerful counters. Like previous games, OpenAI also did not use the summon and fantasy features. Both OpenAI are not training the system and will fetch additional variables in the form of heroic copies and unique creatures.
Beyond that, the game will proceed like a normal Dota 2 game and the ultimate goal is to destroy the big tower at the end of each team's territory or the "ancient" This hero is vulnerable only when the hero team successfully destroys a small tower during a fight between teams.
OpenAI Five surprised OG and surprised OG, claiming victory by relying on aggressive tactics, including a unique decision to instantly revive the hero according to death, using calls made within the game. match. As Greg Brockman, Chief Technology Officer at OpenAI, pointed out, OpenAI relies on a strategy that favor short-term benefits, revealing its lack of understanding of long-term planning types, and generally relying on those strategic contests to win . However, early buying was favorable in this game, and OpenAI Five had the advantage that OG could not overcome when the game was dragged into the 30-minute range.
We always see this happening in a test game. Bots live again, humans laugh, humans lose. Greg Brockman (@gdb) April 13, 2019
In the second game, OpenAI got an initial advantage for OG. In the first few minutes I went ruthlessly to the players. I won just over half of the time I needed to win the first game. Dota 2 Dota 2 Specializing in the fusion of artificial intelligence and game design, players and viewers have pointed out how the unusually aggressive OpenAI Five began in the second game and how OG fights that development Across the map. Cook specifically mentioned how well OpenAI Five was able to take advantage of certain heroic picks.
Already sadly already finished. OpenAI rated four out of five heroes as net worth. After 10 minutes of running OpenAI, it really gets worse. April 19, 2019
OpenAI's victory is not only a celebration of its own, but its approach also includes reinforcement learning on AI, A general philosophy is a milestone. The research team will not make public protests about AI bots, but they are working on software that allows humans to collaborate in real time with OpenAI Five software, and they are teaming with robots to learn unique and unprecedented strategies. And behavior. The organization also unveils a platform for games with OpenAI Five, the Arena mode, which will be open to the public for three days starting April 18.
Special Announcement: Full Internet invitation to play OpenAI Five can be registered immediately (either competitors or team members).
Sign up now! I am very happy to see what I learned while observing the OpenAI Five in the wild. OpenAI said collaborative software might not be available to the public, but I could, but OpenAI says collaboration software may not be open to the public. peg.twitter.com/TaMhxdgVIt
– Greg Brockman (@gdb) April 13, 2019
Try it for myself at this event. (Despite having world-class Dota 2 artificial intelligence on my team, it unfortunately broke in a much less dramatic way than OG.) Sam Altman, co-founder and CEO of OpenAI, Evidence that collaboration with intelligent agents can bring great benefits in the future.
"An important lesson about how the world works, how to train these things, and how to work in parallel at the same time." Altman interviews The Verge It is one of the vision to think positively about the future.
Altman says that OpenAI will continue to be bolder in Dota 2 and other video game environments. Because he is a good benchmarking tool for measuring the AI test bed and progress, but he says that there is no video game that currently systems like OpenAI Five can not master at a level beyond human abilities. In the wider artificial intelligence industry, video game mastering will soon become a simple table-stake needed to prove that the system can learn quickly and behave in the way needed to perform more realistic and realistic work with more meaningful benefits
Ultimately, Ope nAI wants to extend Dota 2 into a new domain outside the game, eventually expanding to the real world. To do this, organizations use reinforcement learning and other technologies to create more handsome,
"What OpenAI is trying to do is build a common AI and share its benefits with the world and make sure it is safe." Altman is a multi-purpose intelligent intelligence system "You could not be here to win a video game." We have come to uncover the secrets along the way of AGI. "
Correction: An earlier version of this article stated that Sam Altman, co-founder of OpenAI, is the chairman of the organization, while CEO Greg Brockman is CEO.