DotA 2 Finals Bring In Huge Numbers And OpenAI Challenge

It has been a busy weekend for one of the most popular video games in the world.

DotA 2 is an online game with about 15 million unique players. This weekend the game’s creators Valve, hosted their seventh annual world championships, dubbed The International, or TI for short. Team Liquid, an esports mainstay, managed to snag the top spot, running through a gauntlet of other powerhouse teams including a finals defeat of Newbee.

With a win at TI, Team Liquid took home a whopping $10.8 million, the largest ever prize pool for an esports tournament. Perhaps the most shocking stat is that only $1.6 million of that came from the game’s creators themselves; the other $9.2 million comes from fans purchasing in-game content and some of the proceeds going towards funding the overall prize pool. This creates an incentive for players to buy premium content, because they know at least some of the profit will go back into the competitive scene.

The total prize pool dwarfs other major sporting events, including every major PGA event this year, the Tour de France, and even the Super Bowl—although the NFL’s championship game still draws in much more revenue from advertising.

Total viewership for this year’s event has not been announced, but it’s likely the number will top last year’s ratings. In 2016’s TI, over 5.8 million people tuned in to watch the finals.

Fans and companies alike have begun to notice how large the esports industry is becoming, so it was not a surprise to see Elon Musk’s company, OpenAI, use TI as a platform to test machine-learning interactions with a human player. A bot was trained for two weeks to learn the intricacies of DotA 2 and then challenge a pro (Dendi from Natus Vincere) to a one on one match. A normal competitive DotA match is five on five.

DotA 2 is an extremely complex and challenging game. There are 113 characters to choose from, each with unique stats and abilities. Couple this with dozens of items that can augment a player’s capabilities and the combinations become almost endless. This means OpenAI’s challenge was far more daunting than beating a chess grandmaster and even a Go champion.

“We didn’t program it to understand the rules of DotA. We let it play lifetimes of 1v1s against itself then coached it on what’s good and bad,” said Greg Brockman, cofounder and CTO of OpenAI.

The demonstration was limited, but the OpenAI bot crushed Dendi in back-to-back games. Dendi was supposed to play more games, but resigned after the losses, saying it felt like a human, but also felt like “something else.”

It’s also important to note that after the showmatch, OpenAI allowed the audience to play against the bot, offering prizes to the first 50 to beat it. Those prizes were given away quickly, as human combatants utilized what some call “cheese” strategies of unique item builds, pulling enemy creeps to destroy the first tower or simply being very good. This again proves how complex a bot to play DotA 2 can be, as there are so many ways to manipulate the game that are seemingly “unfair” to AI.

Elon Musk cofounded the nonprofit OpenAI to prevent AI from taking over the world, something he has been increasingly vocal on over the past months. OpenAI will now try to field a five man team at next year’s TI to see if they can complete with world champions.