by Nash Bussieres
In Artificial Gamer, Dota is described as Basketball meets Chess – and there’s merit in that. But it’s more like if in your basketball/chess game, you and everyone on the court also got a gun. Dota is a mechanically intensive, heavily strategic team game of mutually assured destruction. Every character in play has abilities and powers that can have devastating consequences and completely shift the tide of battle if used perfectly. So the question posed by Artificial Gamer is an inevitable one: would a computer be able to play Dota more perfectly than a human?
The answer – if you were to ask your average Dota player – would be “obviously no.” Dota is a 5-on-5 team game where players take turns drafting characters, all with unique abilities and attributes, to form a cohesive squad. Your goal is to take down your enemy’s base called “the ancient” (Dota stands for Defense of the Ancients) by coordinating attacks on your opponents’ team and marching forward. You collect gold for killing your opponents and small computer-controlled swarms of enemies that are spawned in waves. Gold allows you to power up your character through buying items with the goal of becoming so powerful that your opponent can’t defend any longer as you waltz into their base and claim victory. It’s very much a war of attrition – even the fastest games can take over 20 minutes to complete.
So that’s the real rub here, a game this complex with this many variables in a real-time setting doesn’t immediately seem like it’s ripe for the taking from our eventual robot overlords. In fact, AI that plays Dota has existed since its inception as an in-game tutorial. And the AI teams, called “bots”, are incredibly bad; even on the hardest setting new players can easily overcome computer controlled opponents.
This concept of not only competent AI, but powerful AI in Dota being a laughable idea in the eyes of the wider community serves as the main narrative of Artificial Gamer. It follows the journey of OpenAI, a company who sets out to make a bot strong enough to beat any Dota team – even the world champs. We first see it take on Dendi – the best player in the world at the time – one-on-one and demolish him. But one-on-one Dota isn’t really the draw; it’s a team game and the complex decisions, coordination and human intuition needed to perform at a top level is completely incongruent with what is needed in a single player game. So can OpenAI do it?
The majority of the film focuses on the trials and tribulations of OpenAI as they try to get their bot ready to fight in time for The International 2018: the Dota world championships. There they will play exhibitions versus real human teams and attempt to prove that their bot can hang with the best. It’s a visually engaging story with lots of fascinating illustrations and fun graphics and is edited in a way that (mostly) nails really difficult segues and topic shifts without feeling too jarring or compartmentalized. The lack of a main narrator and an occasional inability to truly describe the concepts being talked about in laymen’s terms can make it a bit dry if you don’t already have at least a casual understanding of Dota, machine learning or both. The film is built up to The International as if it were to be the climax of the story, but this grand battle happens an hour in and turns out to only be a stepping stone in a much longer journey, which in turn hurts the pacing of the last third of the film.
Compelling and endearing interviews from the team at OpenAI do a lot to emphasize how much the current field of artificial intelligence and machine learning is a wild west; no one knows if anything is actually going to work or how it will work or when it will work. Spending this amount of time and energy on a project that has an unknown chance of success is unforgiving work and you can easily see the toll it takes on the team despite their determination.
Ultimately, Artificial Gamer is a deeply human story about a team of passionate and desperate pioneers trying to accomplish something they’ve been told is impossible. If you’re a fan of Dota or of machine learning in general you’ll get a lot out of it, but your eyes might glaze over a bit from time to time if you’re completely uninitiated.
