AI continually takes the fight to human designers. Neural networks such as OpenAI and DeepMind have successfully defeated some of the world’s best esports players with apparent effortlessness.
Artificial intelligence: The Deus Ex Machina of Esports
Google’s in the habit of breaking a few eggs when it has to. In 2017, AlphaGo faced off with Ke Jie, the world’s Go champion.
If you are not sure what Go is, it’s a strategy game played with multiple pieces, invented 2,500 years ago in China. The objective is to surround your enemy and gain more ground from them. It’s a perfect test of strategic thinking.
OpenAI was learning Dota 2 at 186 years a day, managing to master it to the point where the AI could prevail in single player and controlling an entire team.
So what are the odds of a computer beating the world’s champ? Solid – AlphaGo defeated the “meat bag”, crowning itself the new supreme entity in the Go universe.
Now, something similar is happening in video gaming where, apart from strategy, there’s a lot of fine motor skills to acquire and master as well. It’s the ultimate challenge, and the human makers are losing the game.
From Dota 2 to StarCraft II, what other games do we suck at when faced with the ultimate enemy?
The Importance of Keeping an OpenAI
Entrepreneur Elon Musk partnered up with a fellow inventor Sam Altman to create the OpenAI initiative intended to develop Artificial Intelligence (AI) and machine learning.
Apart from that noble goal, it must have seemed funny to Mr. Musk and Mr. Altman to have a go at a bunch of human Dota 2 pros, because they challenged a number of decorated players from Danil “Dendi” Ishutin to community analysts, commentators and former competitors.
So, the game started and OpenAI drew first blood. By the time the dust had settled, accompanied by the occasional gasp from human players, OpenAI had schooled a ragtag band of former Dota 2 professionals, gaining true legitimacy among competitive outfits.
Thankfully the OpenAI squad is benched indefinitely, having successfully achieved its goal of defeating the humans, and perhaps bringing a few existential questions to fore. OpenAI played at a staggering rate, cramming 180 year of practice in a single day. Quite the pace indeed.
The AI also triumphed over a team of established Dota 2 personalities, including:
- Ben “Merlini” Wu
- William “Blitz” Lee
- Ioannis “Fogged” Lucas
- David “MoonMeander” Tan
- Austin “Capitalist” Walsh
The defeat was disheartening. In fact, Mr. Walsh shared his despair at the outcome: Never felt more useless in my life but we’re having fun at least so I think we’re winning in spirit.
Ground Control to Major Tom
Meanwhile, AlphaStar’s DeepMind has managed to school two StarCraft II players in Protos vs. Protos mirror match. DeepMind, though took a little while longer to prepare and defeat Dario “TLO” Wünsch and Grzegorz “MaNa” Komincz, two renown players from Team Liquid, an esports organization.
DeepMind completely dismantled opposition from TLO, who admittedly was a Zerg player (one of the three possible factions in StarCraft II), but it didn’t falter either when faced with a professional Protos player.
Following five straight losses, MaNa joked that he had considered himself to be a fairly decent player (he’s currently the highest winning Protos vs. Protos player in StarCraft II, or at least one thereof). We took a closer look at the entire game in our piece “DeepMind Played 200 Years to Defeat StarCraft Pros”.
The computer wasn’t unbeatable as an 11th game played a while after the first few indicated. DeepMind ran into a classical do-not-compute hurdle when MaNa started dropping units and systematically harassing the economy of the Protos AI.
DeepMind was at a loss with MaNa having taken the fight to the enemy. What followed was a sweeping deployment of MaNa’s troops, taking out the stunted AI. Not the finest triumph, but one that at least helped team H secure a nominal progress against the machine.
HAL Your New Overlords
Putting AI to the test in highly competitive video games seems like a proper undertaking. However, scientists haven’t narrowed down their research to just these games. The Carnegie Mellon University developed its Libratus AI against poker professionals, with the computer ending up with $1.7 million in net worth.
In a similar vein, back in 2016 Facebook tried to see if an AI is any good at negotiating, teaching bots to play Monopoly. You might have thought that negotiation isn’t something that comes naturally to AI, but as the experiment proved, the AI were able to fake interest and mislead players.
It Doesn’t Have to Empty Esports from Meaning
Just because “computers” are able to beat humans at two of the most demanding games, and soon, dominate any conceivable competition out there, doesn’t mean that esports is any less entertaining or seeing individuals excel at games such as StarCraft II and Dota 2 excites less admiration.
Still, now that a PC can learn at a pace that no human player can ever achieve, it does seem a bit dispiriting.