[ad_1]
This is the game for OpenAI, the research initiative co-founded by Elon Musk who put his AI bots against Dota 2 amateur players and wins.
Then, OpenAI will challenge the best players of the arena fighting game at the Dota 2 The International tournament, described by The edge as the biggest annual event on the e-sports calendar. Indeed, professional and highly motivated players train all year to try to win the annual prize pool of 40 million Dota.
What's happening here with OpenAI: the lab has already successfully trained its bots to beat the best Dota 2 players in one-on-one games. What the laboratory is trying to do now is to put them in place to compete against human teams in 5-on-5 competitions.
It is, needless to say, an extremely complex business. OpenAI teaches AI robots through reinforcement learning, putting them in a virtual world and using trial and error to help them find themselves how to achieve a goal. The reward functions are identified, where the bots score points when they perform a certain task, but otherwise they are basically released to play themselves.
For this last workout, the OpenAI bots have played the equivalent of 180 years of games against them every day. Greg Brockman, co-founder of OpenAI and CTO The edge it starts at random, like doing something like walking around the map. Basic skills are picked up in a few hours. What all this equates to – if you assume that it takes a human at least 12,000 hours to become a gaming professional, the OpenAI systems "play 100 lives of experience each and every day. "
OpenAI bots have a lot of advantages that you can guess. Faster reaction times, for example, plus instant access to data such as inventories and health of heroes that human players should manually check. To be sure, they also do not have a taste of the fullness of Dota 2, with limitations that only include the use of five of the 115 heroes – all of whom have different play styles – which are available. Some aspects of the game like invisibility have been completely disabled.
OpenAI Five – a team of five neural networks – "averages 150-170 actions per minute (and has a theoretical maximum of 450 because of the observation of every fourth image)," reads an OpenAI ticket released today. On the news. "The perfect timing for skilled players is trivial for OpenAI Five – OpenAI Five has an average reaction time of 80ms, which is faster than humans."
All this is not just for pleasure or bragging rights. There is certainly an application here that may prove useful in the future. An AI system that can learn to master a complicated video game could be used in the real world, for example, by optimizing something like a municipal utility grid.
"Our underlying motivation goes beyond Dota," says the OpenAI blog. "AI deployments in the real world will face challenges raised by Dota that are not reflected in chess, Go, Atari or Mujoco benchmarks." In the end, we will measure the success of our Dota system in its application to real-world tasks. "
Before the August competition, OpenAI plans to hold a match with the best players on July 28th. You can follow Twitch to watch live or request an invitation to participate in person.
Source link
