Elon Musk’s OpenAI previewed the power of its own artificial intelligence at Valve’s yearly DoTA 2 tournament. The AI agent dominated several DoTA 2 professionals in 1 vs 1, showing the early capabilities of AI inside the MOBA.
For those that don’t follow DoTA 2, it is multiplayer online battle arena, primarily played in a 5 vs 5 format. The aim for each team is to destroy towers to reach the opposing team’s base.
OpenAI spent two weeks training the bot inside the game, constantly running through simulations against itself. It amassed “lifetimes” of gameplay, learning intricate tactics to beat its opponent. The bot learned how to creep block, zone, and bait, considered key skills for DoTA 2 players, which surprised most of the professionals.
At the International, DoTA 2 pro Danylo “Dendi” Ishutin was defeated twice by the bot.
OpenAI first ever to defeat world's best players in competitive eSports. Vastly more complex than traditional board games like chess & Go.
— Elon Musk (@elonmusk) August 12, 2017
It should be noted that the 1 vs 1 was extremely limited compared to a normal game of DoTA 2. Bottle, runes, neutrals, raindrops, shrines, and soul rings were not allowed, limiting the scope of the game.
That said, OpenAI is looking to have a full bot team for the International next year, which may compete with some of the teams that make it to the tournament.
It is not the first AI agent to show proficiency in video games. DeepMind, owned by Google, has perfected Atari Breakout, beat the number one player in Go, and is working on an agent to compete in Blizzard’s Starcraft 2.
Musk, the founder of OpenAI, has said that AI is an existential threat to humanity, and last week on Twitter he called it “vastly more risky than North Korea”. He asked governments to actively regulate the sector, to the annoyance of some AI experts and Mark Zuckerberg.