Back in early July, we reported on A.I. players managing to defeat humans in the shooter Quake III Arena by making use of tactics and objective-focused strategies rather than just killing opponents. Artificial players have now managed to prove they can compete against one of the most hardcore competitive games around, Dota 2, and they were more than up to the challenge.
The artificial intelligence research company OpenAI created a team of A.I. Dota 2 players it dubbed the “OpenAI Five” to compete against current and former professional Dota 2 players ranked among the best in the world. The A.I. team drafted its players for the first game and projected that it had 95 percent change of beating its opponent. It came out of the gate strong, getting five kills in a row before the humans were able to get one, and it only got more lopsided from there.
The OpenAI Five took a tower, managed to eventually wipe the entire human team before taking their base, and declared victory a short time later. A similar scenario played out in the second game, with the A.I. players continuing to rack up kills.
With the third match, the OpenAI Five did start to run into problems. The Twitch viewers and audience members at the end got to pick the heroes the team played with rather than let them choose through a draft, and the OpenAI Five knew it was in trouble — the team only gave itself a 2.9 percent chance of victory. With the audience intentionally giving the A.I. players a less-than-ideal lineup of heroes, the humans were able to achieve victory, racking up more kills and showing that perhaps robots aren’t going to kill us all — provided we manipulate them a little bit.
This isn’t the last we’ve seen of the OpenAI Five. CTO Greg Brockman said the team will also compete in The International, the prestigious Dota 2 event taking place later this month. Impressive as the A.I.’s performance has been thus far, to really show its skill, the players will have to beat Dark Souls using only Donkey Konga drums. We’re not sure if that’s actually possible for artificial players, but OpenAI can probably figure it out. The team has managed to improve its A.I. players over just a few months, with significant progress made since May, and it will probably be unbeatable by the end of the year.
- Neuro-symbolic A.I. is the future of artificial intelligence. Here’s how it works
- Revisiting the rise of A.I.: How far has artificial intelligence come since 2010?
- Google CEO Sundar Pichai warns of dangers of A.I. and calls for more regulation
- Ahead of the 2020 presidential election, Facebook says it’s banning deepfakes
- Top drone racer takes on robot drone in first human-vs.-machine clash