قالب وردپرس درنا توس
Home / Technology / OpenAI's Dota 2 AI is a champion in the e-sports team, winning consecutive victories

OpenAI's Dota 2 AI is a champion in the e-sports team, winning consecutive victories



OpenAI, the nonprofit research organization founded by Elon Musk, can claim a world first: its artificially trained intelligence system trained for the complex strategy game. Dota 2 has defeated an e-sports world champion team. The competition took place today in San Francisco and called itself the OpenAI Five Final. This ended the organization's public demonstrations for its Dota technology at a high level.

The contests on the human side included five Dota 2 professionals from Team OG, which won the world's most coveted e-sports award last year when it became number one in The International, the premiere annual Dota 2 tournament with prizes now totaling $ 25 million. OG beat the OpenAI Five-Bots in one of the top three competitions, all of which are trained using the same depth enhancement techniques and independently controlled by different shifts of the same system. Enhanced learning is effectively a self-improvement attempt and error in which the AI ​​falls into the game environment without understanding how the game works and has been extensively trained using reward systems and other incentive mechanisms.

Today's performance is by far the highest quality demonstration of OpenAI Five's capabilities, with the system losing two games to less powerful e-sports teams last August. According to Greg Brockman, OpenAI's Chief Technology Officer, OpenAI Five improves by playing itself in an accelerated virtual environment. "OpenAI Five is based on intensive learning, which means we did not program it. We programmed it the way you learn it, "Brockman told the contest. "In the 1

0 months of its existence, Dota 2 has already played 45,000 years. That's a lot – it has not gotten boring yet.

Dota 2 is an extremely complex strategy game that involves more than 100 unique characters, deep skill trees and item lists, and a dizzying array of variables displayed on screen at a specific time in a match become. Therefore, OpenAI sets certain limits when its AI system plays professional players, especially by limiting the number of heroes used by both five-player teams.

In this case, each squad had 17 heroes to choose from. OpenAI also chose the so-called "Captain's Draft" mode, where each team can strategically ban heroes to prevent the other team from selecting those characters before using a specific picking order. In this way, the captain can build strengths between heroes' combinations and take advantage of the weaknesses of enemy heroes through strong counters once the teams begin to gradually fill up the lineup. As with previous games, the summon and illusion features of OpenAI have been disabled. In both cases, extra variables are used in the form of hero copies and unique creatures for which OpenAI has not trained the system to account.

In addition, the game is played like a normal match Dota 2 with the ultimate goal of destroying the "old" of the opposing team or a large tower at the end of each team's territory only vulnerable when the opposing team has successfully destroyed smaller towers between Heroes 'Heroes' battles throughout the game.

In the first game of the day, OpenAI surprised Five OG and won the game through a series of aggressive tactics, including a special decision to spend the earned in-game currency to instantly revive heroes after death. OpenAI Chief Technology Officer Greg Brockman says OpenAI loves strategies that favor short-term profit and uncover its shortcomings in mastering the kind of long-term planning people can rely on in such strategic competitions. In the match, however, the early buy-backs paid off and OpenAI Five gained an advantage OG simply could not overcome as the match drew in the 30-minute space.

In the second game, OpenAI was even better, earning an early advantage over OG The first few minutes and then ruthlessly advanced on the human players, until they achieved the victory in just over half the time it took to win the first game. Mike Cook, an avid Dota 2 player and viewer specializing in the mix of AI and game design, noted how unusually aggressive OpenAI Five began to play in the second game and how little OG was against his progress was fought over the map. In particular, Cook noted how well OpenAI Five could use his specific hero favorites.

For OpenAI, winning here is not just a reason to celebrate in and of itself, but a proof that he's approached Reinforcing learning and its general philosophy of AI will lead to milestones. The company will stop hosting public demonstrations of its AI bot, and is now working on software that allows people to collaborate in real-time alongside the OpenAI Five software, play in a team with the bots, and their unique, unprecedented strategies and strategies can learn behavior.

OpenAI says this collaboration software may never make it available to the public, though I could try it out for myself here at the event. (Although I had world class Dota 2 AI on my team, unfortunately, I was downplayed in a less dramatic manner than OG.) But Sam Altman, co-founder and chairman of OpenAI, says that this is some sort of work proof that cooperation with AI agents can bring enormous benefits in the future.

"This is an important lesson for the world as this world is trained to train these things and to let them work in parallel an interview with The Verge . "Collaboration is one of the more positive visions we have for the future of the world – AI works with people to make people better and more fun and more effective."

According to Altman, OpenAI is likely to continue battling with . Dota 2 and other video game environments, especially because they are such good test environments for the AI ​​and good benchmarking tools for measuring progress. But he tells me that there's probably no video game that can not handle a system like OpenAI Five on a level beyond human capabilities. For the broader AI industry, mastering video games may soon become a passé. Simple table inserts required to prove your system can learn quickly and act in a way that is needed to handle harder real tasks with meaningful benefits.

Ultimately, OpenAI wants to take its Dota 2 learning and expand it to new realms beyond games and finally into the real world. To this end, the organization is working to use reinforcement learning and other techniques to provide robotic hands with skilful, skilful and human-like movements.

"What OpenAI is trying to do is build common artificial intelligence and share those benefits with the world and make sure it's safe," Altman says, pointing to the drive to create a multi-purpose AI system. that can do any job a human can "We were not here to hit video games, as funny as that is. We are here to uncover secrets on the way of AGI. "


Source link