One of the simplest ways to test the intelligence of a machine is to pit them against humans – in some sort of objective playing field. Games are, indeed, the most prevalent platform that measures the capabilities of Artificial Intelligence (AI) as compared to the intelligence of humans. While playing complex games, if an AI defeats its human counterpart, then it has demonstrated superior strategic thinking and decision-making abilities. At present, Dota 2 is one of the most advanced strategic games available on the market – with around 10 million players worldwide. The International 2018, the annual e-sports tournament that brings together the very best professional Dota 2 players in the world, is set to match the strategic abilities of the winning team – by pitting them against a team of bots developed by OpenAI, called OpenAI Five.
Since the development of AI kick-started, they have been trained to acquire the necessary skill set required to master various strategic games. One such strategic board game, Chess, was the weapon of choice for many AI developers. By playing several hundreds of games against itself, it was predicted that an AI bot could gather enough experience to master the game. Indeed, in 1997, IBM’s Deep Blue AI was able to defeat then Chess Grandmaster, Gary Kasparov, in a highly publicized match. Since then, AI developers have expanded their domain to teach their bots other games as well. DeepMind’s AlphaGo, in 2016, beat Lee Sedol, a world master at the traditional Chinese board game Go. Since then, AI bots have managed to defeat humans in numerous other games as well – rendering decades of human practice and skill worthless against mechanical computation.
So why are AI researchers so curious with the outcome of The International’s Dota 2 game? The fast-paced multiplayer strategy game is set to pose a different set of challenges for the AI bots – requiring collaboration capabilities and managing unpredictability. Dota 2 teams usually consist of five players – so it’s up to 5 independent AI bots to defeat the best 5 member human team in the world. Researchers are intrigued by the ability of AI to indulge in teamwork, an essential component that’s required to win the game.
OpenAI, the AI research lab founded by Elon Musk and Sam Altman, have used reinforcement learning to train their bots. This is similar to the methods used to train bots in chess, only advanced. By rewarding the system for every correct maneuver it makes, the bots indulge in positive reinforcement, learning to take the right decision at the right time. While training the OpenAI Five in Dota, the bot’s reward is also shifted from getting points for themselves to increasing the overall team’s score.
When the bots were first introduced two months ago, they acquired 180 years’ worth of experience in just one day. At the start, OpenAI Five lost to amateur players. Within a month, they had ‘matured’ enough to defeat casual players. It’s no wonder, that the anticipation of the matchup between the world’s best Dota 2 team and OpenAI Five is immense.
So, what does a win at Dota 2 entail for AI developers? They believe that the same algorithm that can match humans in creating a game-winning strategy can be used to solve real-world problems. Coming up with solutions to the numerous problems facing the world today is extremely tricky, and require cost analysis of several factors. If AI manages to beat humans in an advanced strategy game, maybe they can provide solutions to real-world problems as well.