My favorite game has been invaded by killer AI bots and Elon Musk hype

Elon Musk is a hype merchant, and this weekend his Twitter account served up a generous serving of hoopla for the OpenAI bots that were destroying the best Dota 2 players in the world. As a veteran Dota player and inveterate contrarian, I couldn’t let Musk’s exaggeration go by unchallenged. What we saw the OpenAI bots achieve was awe-inspiring for anyone who’s ever dabbled in Dota, but it’s still only scratching the surface of the competitive complexity of this game.

According to Musk, the achievement by the OpenAI team eclipses those of previous artificial intelligence projects that created computers capable of defeating the world’s best chess and Go players. He presents Dota as a vastly more complex game, which it is, but he omits the point that the OpenAI bots were playing a thoroughly simplified variety of Dota.

The best way to think of this multiplayer online game is like a version of basketball: five players on each team, a playing field of fixed and (broadly) symmetrical dimensions, and a ton of strategic and tactical depth. There are specialist support players acting like point guards in assisting their more powerful teammates, who are expected to carry the team to victory much in the same way that a high-scoring NBA star might be. Now imagine all the intricacy of team play and positioning scythed off and having just a simple one-on-one pickup game: that’s what the OpenAI Dota 2 bots were involved in. It’s still a competitive form of the game, but it bears only a fraction of the complexity of the full team contest.

The Dota 2 bot that was shown off this weekend was basically the world’s best pickup-game baller. It can play only one in-game character, out of a selection that now numbers more than 100, and it can only play against the exact same hero on the other side. That hero, named Shadow Fiend, requires accurate timing and placement of his magical attack ability, the Shadowraze, and we know from aimbots that superhuman mechanical precision is a trivial thing for computers to accomplish. Furthermore, none of the pro players had time to practice against, observe, or understand the play patterns of the bot before playing against it (which they’d usually do when taking on a fellow human).

I note all these limitations and constraints so as to properly couch the excitement I’m about to express. You need to understand that this AI advancement is a gradual step rather than an abrupt leap forward.

With all of that out of the way, let me say that the Dota player and tech enthusiast in me was mesmerized by the game between pro player Danil “Dendi” Ishutin and the Shadow Fiend bot (in the video above). Dendi is one of the best known and loved players, a Shadow Fiend expert in his own right, and he was defeated by the AI in a dominant fashion. If I were to go up against Dendi and he was feeling utterly ruthless, he’d probably have done the same to me that this bot did to him. The bot baited Dendi into over-aggression and it faked Shadowraze attacks to scare him off. It delivered a clinical display of tactical supremacy.

For me, that match ranks as the most evocative demonstration of artificial intelligence yet. This is because the bot learned by itself. The OpenAI developers let it out onto the map, and over the course of two weeks of intense self-training — which OpenAI claims amounted to “lifetimes” of practice time — the system went from walking dumbly to its death to becoming a cold-blooded killer. The whole process is a little bit magical to me and feels like intellectual evolution accelerated to a timescale where we can actually perceive it.

The OpenAI Shadow Fiend figured out how to zone out the enemy opponent. It learned the mechanics of health potions, which regenerate the hero over time, and how they can be disrupted by attacking the enemy during the regenerative period. And it adapted to experimental human trickery like an opponent trying to confuse it by dropping destructible items on the ground.

What I saw this weekend was the Dota equivalent of an android beating Steph Curry, LeBron James, and all other NBA stars in a one-on-one game. During the same weekend, Dota players competed for a grand prize in excess of $10 million in the final of the International tournament, so you can rest assured that the level of skill and competition was high.

Only the bot left the venue undefeated.

As wowed and delighted as I am by this OpenAI demo, I write to caution against rushing to conclusions about superhuman bots. The full-fat Dota 2 is a game that took me months to understand, let alone master, and its full variety of possibilities is orders of magnitude more complex than the stuff AI has been shown capable of adapting to so far. If the OpenAI group fulfills its ambitious promise of bringing an entire team that can beat the world in proper Dota matches next year, I’ll gladly jump aboard the Musk hype train. Just not yet.