deepmind's alphastar beats pros in starcraft ii

[Daniel Shawul]

https://deepmind.com/blog/alphastar-mas ... rcraft-ii/

[smatovic]

https://deepmind.com/blog/alphastar-mas ... rcraft-ii/

Moratorium for autonomous weapon systems anyone?

--
Srdja

[Werewolf]

https://deepmind.com/blog/alphastar-mas ... rcraft-ii/

Great presentation. Wonder why they trained on Protoss? Shorter matches?

[mar]

Great presentation. Wonder why they trained on Protoss? Shorter matches?

They actually explain it in the demonstration video.

What I find more interesting is why they only played one map (CatalystLE).

Also note that each of the games was played against a different agent, which I find really cool because in each game the style of the AI was different.
From what I saw the humans could've won more games, the AI cheated a bit in one game by micro-managing three groups of stalkers at once which would be physically impossible for humans, but this is also mentioned in the video.

In the last game MaNa embarassed the AI which got confused by repeated Immortal harassment (but they made it a bit harder for the AI which is also explained in the video); were it not for this mistake the AI would probably win though.
To be fair it's also worth noting that they had also limited average APM to match humans, so except for insane micro of the previous agents, there's no doubt about the fairness of the matches.
Interesting was also the second game against MaNa (you have to download the replay), where the AI used proxy tactics and catched the human player off guard.

It's a huge achievement and they'll likely do better in the future (hopefully all 3 races), looking forward to see more games against top humans.
They also said that the AI ran on commodity CPU+GPU during the matches, so I'd love to see some weaker version integrated into the game being able to play against it.

[Uri Blass]

I wonder how much time humans who know nothing about starcraft need to understand the rules of starcraft.
I failed to find a simple explanation about the game and I have no idea about this game.

[Nay Lin Tun]

I wonder how much time humans who know nothing about starcraft need to understand the rules of starcraft.
I failed to find a simple explanation about the game and I have no idea about this game.

Almost all RTS have three basic strategies. You can easily check with a good tech tree or diagram showing which unit do what task.
1. Build economy( collecting resources)
2. Build Army
3. Fight and conquer.
The decision depends on how long you should build economy and when should you start building army ! There is no cut off rule, but pure risks and skills!!
If you keep building economy, your enemy may destroy you. But if you build army early, you must effectively damage your enemy; otherwise your economy will be too lag behind and you will lose eventually.

[Ozymandias]

I wonder how much time humans who know nothing about starcraft need to understand the rules of starcraft.
I failed to find a simple explanation about the game and I have no idea about this game.

Almost all RTS have three basic strategies. You can easily check with a good tech tree or diagram showing which unit do what task.
1. Build economy( collecting resources)
2. Build Army
3. Fight and conquer.
The decision depends on how long you should build economy and when should you start building army ! There is no cut off rule, but pure risks and skills!!
If you keep building economy, your enemy may destroy you. But if you build army early, you must effectively damage your enemy; otherwise your economy will be too lag behind and you will lose eventually.

I would argue that you don't even need an army. The goal is to destroy your opponent before he destroys you, if you can do that with your resource-gathering units (and some people play that way), the army turns into an optional strategy, not a rule. I'm sure you can even find instances of wins, where no resources were collected by the victor at all (extreme rush cases).

[WinPooh]

Results in chess and go were much more impressive to me. Maybe because I always knew that computer can "click his mouse" faster than any human.

[mar]

Results in chess and go were much more impressive to me. Maybe because I always knew that computer can "click his mouse" faster than any human.

Quite the opposite. You seem to have no clue how hard it is.

Would you say that typing keys faster will make you a better writer?
A monkey can type faster than you but the sentences won't make any sense.

Anyway, that's not even how the AI inputs commands, the actions are not "clicks or key presses".
So you "always knew" nothing it seems.

[Nay Lin Tun]

Results in chess and go were much more impressive to me. Maybe because I always knew that computer can "click his mouse" faster than any human.

They restrict AI's action per minute for fair play with human. There is details discussion in reddit.
Afaik, human did 450 APM(action per minute) and AI did 250 APM in average. But people said only 50% of human APM would probably be effective action, cos usually human repeatedly click mouse or shift key several times without effective action or double or triple click for the same action.