Humans grab victory in first of three Dota 2 matches against OpenAI

Artificial intelligence has swept the board with humans in games like chess and Go, but taking on e-sports might be a step too far — for now. At The International tournament last night in Vancouver, a team of human pro gamers defeated a team of AI bots at the battle arena game Dota 2. The victory for team human was decisive but by no means inevitable, with the AI players putting up a valiant fight. And with two more games to play this week, machine might yet triumph over humanity.

The bots were the creation of OpenAI, a nonprofit research lab founded by tech luminaries such as SpaceX CEO Elon Musk. The lab’s main goal is to develop artificial intelligence that “benefits all of humanity,” but teaching bots to play Dota has been an important research task for some time now. Playing video games may seem trivial, but it’s a huge challenge for AI, requiring teamwork, long-term strategy, and complex decision-making. The hope is that these skills, once honed in video games, can be applied to real-life challenges.

In last night’s match, though, AI could not quite muster the goods. The two teams swung back and forth in the early stages of the game, with humans taking the lead, then AI, then humans. But ultimately, it was human strategy that carried the day.

The bot team, dubbed OpenAI Five, played well but made some key mistakes, including an unusual fixation with a neutral character in the game known as Roshan. This character lives in the middle of the map, and heroes can team up to kill it as a side-quest. Doing so is dangerous but brings serious rewards, including an item that allows heroes to respawn quickly. OpenAI repeatedly targeted Roshan, a tactic that sometimes paid off (giving key heroes extra lives), but other times left their home territory undefended, giving the humans — a team of pro gamers from South America named Pain Gaming — a chance to dig in.


A team of AI-controlled heroes take on Roshan. (Screenshot not from last night’s game.)
Image: OpenAI

This isn’t to say that the bots showed no strategic thinking at all. Far from it. They often made decisive plays, like sacrificing isolated heroes that a human team might have tried (and failed) to save. They also played an incredibly tight game at close quarters, with precision attacks in the heat of battle. As AI games researcher Mike Cook commented on Twitter: “The bots are still very good at moment-to-moment, but they seem bad on macro-level decisions.” (For a detailed blow-by-blow account of the match, it’s worth reading all of Cook’s tweets.)

This is as you might expect. The AI system, after all, plays like a machine. It doesn’t get flustered and it has access to the game’s backend, giving it a precise feed of numbers covering information like the health of each hero and distance between them.

The humans’ victory wasn’t a complete surprise either. Although OpenAI’s bots have previously beaten top pros in 1v1 matches and teams of skilled amateurs in 5v5 matches, these were comparatively simpler games. A 1v1 match emphasizes skills where AI excels (like precision timing), and the 5v5 matches were played with more restrictions. Dota 2 has a number of complex mechanics, including a huge roster of characters to choose from, each with their own skill set, and in-game systems such as couriers (NPCs which deliver items like health potions) which add to the strategy. In earlier games, OpenAI removed lots of these elements, but in the matches at The International this week it’s adding some back in, increasing the game’s difficulty.

Speaking to The Verge ahead of the game last night, OpenAI co-founder and chief researcher Greg Brockman said that an internal poll of employees had suggested there was “less than a 50 percent probability of winning.” “That was the general consensus,” said Brockman, before adding that what was really important was the rate that the AI team was improving. “Usually we start playing teams when we’re about at their level, then a week or two later we surpass them. And that has happened to us a number of times now.”

So the question is now, can OpenAI Five learn from this experience and champion in the two games to come? Can they adapt and overcome? Just imagine the pep-talk those bots are getting in their AI locker room. “They’re only flesh and blood.”

The OpenAI Dota 2 bots just defeated a team of former pros

A month and a half ago, OpenAI showed off the latest iteration of its Dota 2 bots, which had matured to the point of playing and winning a full five-on-five game against human opponents. Those artificial intelligence agents learned everything by themselves, exploring and experimenting on the complex Dota playing field at a learning rate of 180 years per day. Today, though, the so-called OpenAI Five truly earned their credibility by defeating a team of four pro players and one Dota 2 commentator in a best-of-three series of games.

There were a few conditions to make the game manageable for the AI, such as a narrower pool of 18 Dota heroes to choose from (instead of the full 100+) and item delivery couriers that are invincible. But those simplifications did little to detract from just how impressive an achievement today’s win was. The OpenAI team won against well-known Dota personalities Ben “Merlini” Wu, William “Blitz” Lee, Ioannis “Fogged” Lucas — all of them former professional players — along with current pro player David “MoonMeander” Tan and play-by-play commentator Austin “Capitalist” Walsh. Walsh sums up the despondency felt by Team Human after the bout neatly:

The OpenAI Five triumphed in convincing fashion in the first game, not allowing the human players to even destroy one of their defensive towers. The humans recovered a little in game two, conquering one tower, but they still got demolished. And finally, in a game three played purely for pride, the humans managed to squeeze out a win.

What stands out when you watch the matches is the apparent intelligence of the AI’s decisions and the inhuman absence of any indecision. The typical Dota 2 game, even on the professional tier, involves quite a bit of equivocation about whether to engage in a fight, try and shift it to a more favorable battleground, or run away from it completely. The OpenAI team just doesn’t need the processing time that humans require, which made its play appear unnatural — but only in the speed and crispness of the decision-making, not in the content of those decisions.

The developers of OpenAI noted that the OpenAI Five were losing to amateur players within their team back in May. By June, the AI had matured enough to defeat casual players, and today, it’s shown that it’s capable of overwhelming people who’ve been playing Dota 2 literally since its inception. The next goal for this rapidly evolving AI is to take on the very best Dota 2 players at Valve’s The International 8 later this month. That’s where the best teams will compete for the grand prize of being 2018’s Dota 2 champions, and one of the side shows will be a contest between them and the OpenAI Five.