Last year, OpenAI, a company founded by Elon Musk, showcased one of its super advanced AI creations at the Dota 2 World Championship (The International 2017), an esports tournament with a prize pool of 24.7 million dollars.
At the time, the company’s super intelligent AI bot, which had acquired the equivalent of tens of thousands of years of experience in just a few short months by playing over and over against itself, defeated the best players in the world in a 1v1 (Shadow Fiend vs. Shadow Fiend) match-up.
With only 10 days left to go until The International 2018, OpenAI is testing today (August 5th) another one of its creations. This time an entire team of AI bots will be taking on a team of human players (top Dota 2 analysts and professionals, 99.95 percentile) in an attempt to push the boundaries of Artificial Intelligence and prove that it can master hypercomplex systems with 1000+ notions (such as Dota 2) and learn how to perform within them better than the most knowledgeable experts.
Artificial Intelligence has already conquered difficult and complex games like Chess and Go, defeating the highest skilled players in those games with relative ease. But Dota 2 is in a class of its own, being much more difficult and complex than these two board games, mainly due to its huge number of notions, mechanics and options. If Artificial Intelligence can beat humans in such a sophisticated game, then its level of intelligence truly is extraordinary.
First reaction
Upon hearing about the project, I was a bit skeptical about the whole thing and dissatisfied with the limitations imposed upon the match-up. These limitations were designed to put humans at a severe disadvantage and allow the AI to thrive, by having it play under ideal conditions.
Basically, the whole situation would have involved a mirror match-up (5 predetermined heroes vs. the same 5 heroes) and in such a case, the human team would have faced an AI team that had already played this combination of heroes against itself literally millions of times. Furthermore, the 5 heroes that had been chosen (Viper, Sniper, Lich, Crystal Maiden and Necrophos) were very simple and didn’t allow for sophisticated maneuvers during the game.
So the drafting knowledge possessed by humans and their high-level skill with various heroes that are difficult to master would have been neutralized.
New conditions
A few weeks ago, OpenAI changed its mind about the conditions under which the OpenAI vs. Humans match will take place. The new conditions allow both teams to draft heroes from a pool of 18 possibilities. Thus, the number of options has dramatically increased, allowing humans to make use of a higher percentage of their Dota 2 knowledge and exploit potential weaknesses in the AI’s decision making.
A number of other limitations have been lifted, such as the use of wards, the ability to kill Roshan and the option to use invisibility items.
The human team
The human team is made up of 5 strong players: Blitz, Cap, Fogged, Merlini and MoonMeander. The first 3 are top Dota 2 analysts, the 4th is a former Dota 2 analyst, while the 5th is a professional player. All of them are very high MMR competitors (top 0.05%). To be fair, they don’t actually constitute a professional team, but their shared knowledge is still a big asset and should allow them to come up with a winning strategy.
Betting prediction
Betway puts the odds for this match at 1.45 – 2.55 in OpenAI’s favor. There’s little faith in humanity for this match, but the bookmaker’s assessment is wrong in my opinion. That’s because it relies on previous results obtained by OpenAI against much lower skill players, and also because it doesn’t take into account the new set of conditions under which the match is taking place.
The fact that humans are allowed to draft 5 heroes of their choosing gives them a big advantage, even though the hero pool is still quite limited. That’s because it allows them to create highly specific compositions designed to win the game via what is called a cheese strat, which is a kind of strategy that looks dumb but works simply because the opponent has never seen it before or simply doesn’t expect it. And this kind of strategy has already been successfully used against the 2017 OpenAI bot do bring about its demise.
So in spite of what the bookmakers believe, the human team actually has at least a 50% chance to win this match. And at these odds, this creates a great Dota 2 betting opportunity.
Humans to win