Artificial intelligence has a long history of challenging players in games like Go or chess, however the cross over into video games has yet to be shown off…until now. During The International 2017 Dota 2 world championship, the OpenAI engine backed by Elon Musk has been beating pro players in quick fashion.
The OpenAI engine took on three of the best Dota 2 players after learning the game in just two weeks. The AI played itself over and over again working out the best approach given any encountered situation. The makers of the AI, claim that it isn’t any better than the average human in regards to inputs. The strengths lie with predicting player movements and improvisation based on those predictions.
One of the most high profile matches pitted the OpenAI bot against Danil Ishutin, better known as “Dendi”. The match was a 1-on-1 match with the first player who scored two kills or destroyed an in-game tower claiming victory. The first game had Dendi on the defense as he attempted to adjust to the bots capabilities. The bot was able to grab a quick kill. Later on the players traded kills awarding the match to OpenAI. The second match saw a very quick kill from OpenAI. Seeing the writing on the wall, Dendi conceded the match.
Based on data from the makers, OpenAI can beat regular Dota 2 bots in just an hour. For those pro players of the world, give it two weeks of training. All hope is not lost however; during additional tests during The International 2017, players discovered ways to confuse the bot. Distracting the bot from the main objectives in the game, allowed for a little human retribution!
While the 1-on-1 match simplifies the game which allowed the bot to learn the basic game without all the team based strategy, that is just the beginning of what the OpenAI team has plans for. In a blog post, they stated they hope to build a group of bots that can take on a top human team.