It's a game for OpenAI, the research initiative founded by Elon Musk that uses and wins its AI bots against amateur Dota 2 players.
Next, OpenAI will challenge the top players of the Arena Battle game at the Dota 2 tournament The International, described by The Verge as the biggest annual event in the e-sports calendar. In fact, professional and highly motivated players train year round to earn Dota's $ 40 million annual prize pool.
What's happening here with OpenAI: The lab has already successfully coached its bots to beat top Dota 2 players. to-head games. What the lab is trying to do is set up human teams in competitions of 5 vs. 5.
It is a very complicated undertaking. OpenAI teaches the AI bots through learning by introducing them into a virtual world and using trial and error to help them find out for themselves how to reach a goal. Reward functions are identified in which the bots earn points when they perform a certain task, but otherwise they are basically free to play themselves.
For this final round of training, the bots of OpenAI played the equivalent of 1
The OpenAI bots have many advantages that you can guess. Faster response times, for example, and instant access to data such as inventory and hero health that human players would need to manually review. Of course, they also get no idea of the abundance of Dota 2, with limitations that only five out of 115 heroes – all of whom have different play styles – are using. Some aspects of the game such as invisibility have been completely disabled.
OpenAI Five – a team of five neural networks – "averaging about 150-170 actions per minute (and has a theoretical maximum of 450 due to watching every fourth frame)," reads an OpenAI blog post published today about the news. "Perfect timing is trivial to OpenAI Five, though it's possible for experienced players, and OpenAI Five has an average response time of 80 ms, which is faster than humans."
All this is not just for fun or to brag right. There is certainly one application that may prove useful in the future. For example, an AI system that can learn to master a complicated video game could be used in the real world, such as by optimizing an urban utility network.
"Our underlying motivation goes beyond Dota," he continues the OpenAI blog post. "Real-world AI engagements need to address the challenges Dota poses that are not reflected in Chess, Go, Atari games, or Mujoko benchmarking tasks, and ultimately we will see the success of our Dota system in its application for Measure real tasks. "
Before the August competition, OpenAI plans to host a match with top players on July 28th. You can follow Twitch to watch live or to request a personal invitation.