DeepMind’s AlphaStar: A Grandmaster Level StarCraft 2 AI!

📝 The paper "#AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning" is available here:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Haro, Anastasia Marchenkova, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Benji Rabhan, Brian Gilman, Bryan Learn, Christian Ahlin, Claudio Fernandes, Daniel Hasegan, Dan Kennedy, Dennis Abts, Eric Haddad, Eric Martel, Evan Breznyik, Geronimo Moralez, James Watt, Javier Bustamante, John De Witt, Kaiesh Vohra, Kasia Hayden, Kjartan Olason, Levente Szabo, Lorin Atzberger, Lukas Biewald, Marcin Dukaczewski, Marten Rauschenberg, Maurits van Mastrigt, Michael Albrecht, Michael Jensen, Nader Shakerin, Owen Campbell-Moore, Owen Skarpness, Raul Araújo da Silva, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh.

Károly Zsolnai-Fehér's links:

[Wecoc]

Me: I don't have to worry about driving well because my car has anti-collision AI
The AI in my car: _playing DOTA2_

[quahntasy]

*Eight minute paper.*

But hey I loved it.

[YashKMusic]

Two Minute Papers: _uploads video_
.
.
Me: "What a time to be alive!"

[ThalCast]

"its kind of like a drunken kung fu master, but in StarCraft2" Lol
Meanwhile i burn my brain 200% to be a low plat player

[DavenH]

A small correction: 2:05 you say it's "well above the 99% winrate mark" -- it's rather above the 99 percentile of players. You win approximately 50% of games due to the ladder matchmaking you with similar skill, except at the very top, and Alphastar is not at the top (~7350 mmr, whose winrates is about 80-90%).

[dpie]

I love your enthusiasm and excellent videos. Please don’t stop making them!

[thiagobarbosa]

Yes. Teach them more games to defeat enemies. Nothing can go wrong.

[admthrawnuru]

Next up: DeepMind defeats US military.

[vesacksi]

AlphaStar has no manners tho, they should progam it to say GG after the opponent says GG

[MattRoszak]

Another example of: "Oh, maybe [some problem] wasn't that complicated after all, but AI will never manage to do [some new problem]."
At some point the machines will be able to do everything, and some people still deny that that's even possible.

I think maybe Ray Kurzweil's timeframe is more realistic than we think.

[multicrogamer]

When i first saw Alphastar replays it was impressive, watching them here confirms importance and scale of this research

[FiguraCinque]

When 2 of my favorite topic mix togheter:
Two Minute Papers and Starcracft 2 AlphaStar. EPIC CROSSOVER!

[alexisfy]

I sneezed when the "no pun intended" came up. Almost died right there.

[ARavingLobster]

EYY! So happy to see SC2 and Alpha Star represented here! That's super cool. Shoutouts from the community, I love this channel and we love what we've been able to showcase with Deepmind and Alpha Star.

It's been a super cool research project that the community has been absolutely thrilled to be a part of this past year. We hope to see more of this kind of thing in the future, thank you so much for covering it!

[Gartenguy]

Exciting News.

I wish this learning would be done for so many games. It would be very interesting to see, what a self learning AI can come up in games like Rocket League or the very complex Age of Empires 2.
And while we are on it, we should also not limit the AI APM just to see, how crazy a game can get.

[spuriousc]

Can we get a grandmaster level Cities: Skylines AI and then use it to assist urban planners?

[TheJysN]

Great video, and i am glad you took more than 2 Minutes for this one :P

[Ketroc]

1:58 Going well over the 99% win rate would be impressive. Unfortunately this chart's math is very wrong. Going 18/30 is not 99.76%. I also sense all the data in this chart is off. Alphastar played 50 matches on ladder with each race (not 30), and it's win rate was about 85-90% (~100% while climbing the ladder, then ~50% when it reached a high rank)... so in its first 30 games it should be around 29-30 wins, not 18-25.

[koozdra]

I've been watching commentary on the released replays. So exciting!!

[KyranFindlater]

Super awesome. We can definitely learn some neat tricks which the AI explores and reveals in action, which no human could have thought of /have had time and determination to test properly.