Till date, there has been a general perception that AI algorithms operate independently. Question marks have been raised over their ability to collaborate to perform complex tasks. Researchers at OpenAI have been working on this problem for some time now, and they seem to have found the answer. A team of AI algorithms called the OpenAI Five have managed to beat a team of human video game players in Dota 2 – the popular battle arena game.
OpenAI had previously developed an algorithm which was capable of competing against human players in the single-player mode in Dota 2. This latest achievement using a team of similar algorithms modified to factor in both individual and team success has proved to be quite evolutionary. These algorithms do not communicate directly, but only through gameplay.
How OpenAI Five beat the human Dota experts
The OpenAI Five mastered the game of Dota 2 by initially playing against different versions of themselves. Over a period of time, they managed to learn different strategies which human players generally use – figuring out ways to attack, defend and perform a variety of other tasks. Most importantly, they learnt the art of collaboration and working as a team – something that eventually led them to beat some of the world’s top Dota 2 players.
One of the founders of OpenAI, Greg Brockman thinks that this is a milestone achievement for AI – with great implications that could help humanity in a positive way. “What we’ve seen implies that coordination and collaboration can emerge very naturally out of the incentives”, he says. He added that substituting a human player for an algorithm to play Dota 2 in a team mode worked out very well.
What is Dota 2?
Dota 2 is one of the world’s most popular strategy games, played by millions across the world. In the team mode, five players collaborate to control a building or a structure by planning attacks and engaging in real-time combat. Each of the players have different strengths, weaknesses and roles within the team, and they have to optimize their capabilities to work with the team in the best possible way.
Games continue to be the perfect test-bed for AI
The tradition of pitting AI algorithms against expert game players has been an ongoing tradition. Last year DeepMind developed an AI algorithm AlphaGo that beat the world’s best human Go player, while another program AlphaGo Zero perfected its Go and Chess skills simply by playing against itself iteratively.
Collaborative AI algorithms could be the future
Beating humans in a Dota 2 team game is a rather important achievement for AI. With the commercial applications of AI on the rise, this collaborative approach used by the AI algorithms can prove to be invaluable. These algorithms, for example, can collaborate to outperform humans in a bidding war, or give faster, more accurate predictions related to certain events.
One cannot rule out the possibility of them collaborating even with humans and helping them with their day to day activities in the near future. However, could there be a downside to this? Could human effort be replaced by a combination of AI algorithms working together? We will find out in due course of time, but there seems to be no evidence to suggest this…just yet.