| OpenAI Bot Triumphant Playing Dota 2 |
| Written by Sue Gee |
| Saturday, 12 August 2017 |
|
The latest game in which AI has beaten a top human player is Dota 2, a battle arena video game published by Valve, as a sequel to Defense of the Ancients. OpenAI's self-trained bot made a surprise appearance at this year's International Dota 2 tournament held in Seattle and had a decisive victory.
Dota 2, which is played by over 10 million people monthly is normally played between two teams of 5 players with the goal of collectively destroying a large structure defended by the opposing team known as the "Ancient", whilst defending their own. The ten players each control one of the game's 113 "heroes" each of which have their own strengths and weaknesses. The 1v1 format is also a complex game with hidden information. Agents must learn to plan, attack, trick, and deceive their opponents. The International is Valve's annual week-long tournament for professional players with a prize pool of over $24 million. OpenAI took just two weeks to train a robot to play pro Data 2 players at this year's event and it was undefeated against many top professionals including SumaiL, the top 1v1 player in the world) and top overall player in the world, Arteezy. The match that has hit the headlines was intended to have three rounds against the crowd favourite player, Danylo “Dendi” Ishutin. Open AI's bot beat Dendi in the first match in about ten minutes. Dendi resigned from the second match and declined to play a third and is reported to have said: "Please, stop bullying me" Afterwards he expressed surprise that a bot could outplay a human and said that the bot: “feels a little like [a] human, but a little like something else.”
OpenAI is the non-profit founded by Elon Musk, Sam Altman, Greg Brockman and Ilya Sutskever in December 2015 with a $1 billion investment, see AI Goes Open Source To The Tune Of $1 Billion. It quickly began to implement research projects, see OpenAI Gym Gives Reinforcement Learning A Work Out and last year partnered with Microsoft in order to use Azure. While Elon Musk has been outspoken about the threats posed by AI he took to Twitter to recognize OpenAI's achievement:
In this video, made in advance of the exhibition match Greg Brockman, Open AI's CTO and part of the research team for this project explains why a bot was trained to play a battle game: Open AI’s goal is to build safe artificial general intelligence. We know A I can be very beneficial to humanity, and it's going to require fundamental Advances to see what it's really capable of. Dota is a great test bed for artificial intelligence
Referring to the challenge of teaching a robot to play this game which has very complicated rules he points out that the bot is entirely self trained. It started out in a completely random state with no knowledge of the game's world and simply played against a copy of itself. This meant it always had an evenly matched opponent. Learning by playing itself proved very effective as within a fortnight it had acquired the behaviors and strategies to defeat human players:
Greg Brockman, himself, appeared to be impressed by the success of the method.
In a subsequent tweet Brockman mentions that OpenAI is now working towards Dota 5v5 and is hiring. |
| Last Updated ( Saturday, 12 August 2017 ) |
