Blizzard and DeepMind Turn StarCraft II Into An AI Research Lab (techcrunch.com) 52
Last year, Google's AI subsidiary DeepMind said it was going to work with Starcraft creator Blizzard to turn the strategy game into a proper research environment for AI engineers. Today, they're opening the doors to that environment, with new tools including a machine learning API, a large game replay dataset, an open source DeepMind toolset and more. TechCrunch reports: The new release of the StarCraft II API on the Blizzard side includes a Linux package made to be able to run in the cloud, as well as support for Windows and Mac. It also has support for offline AI vs. AI matches, and those anonymized game replays from actual human players for training up agents, which is starting out at 65,000 complete matches, and will grow to over 500,000 over the course of the next few weeks. StarCraft II is such a useful environment for AI research basically because of how complex and varied the games can be, with multiple open routes to victory for each individual match. Players also have to do many different things simultaneously, including managing and generating resources, as well as commanding military units and deploying defensive structures. Plus, not all information about the game board is available at once, meaning players have to make assumptions and predictions about what the opposition is up to.
It's such a big task, in fact, that DeepMind and Blizzard are including "mini-games" in the release, which break down different subtasks into "manageable chunks," including teaching agents to master tasks like building specific units, gathering resources, or moving around the map. The hope is that compartmentalizing these areas of play will allow testing and comparison of techniques from different researchers on each, along with refinement, before their eventual combination in complex agents that attempt to master the whole game.
It's such a big task, in fact, that DeepMind and Blizzard are including "mini-games" in the release, which break down different subtasks into "manageable chunks," including teaching agents to master tasks like building specific units, gathering resources, or moving around the map. The hope is that compartmentalizing these areas of play will allow testing and comparison of techniques from different researchers on each, along with refinement, before their eventual combination in complex agents that attempt to master the whole game.
Working backwards (Score:5, Interesting)
Re: (Score:2)
Yeah, I'll be interested to see an AI capped at human levels of APM, an AI with un-capped APM vs human and un-capped AI vs un-capped AI.
I wonder if the un-capped versus will in anyway reflect the current game strategies or just be its complete own kind of meta.
Anything that does splash damage would be less useful if your opponent can split perfectly every time without fail.
Re: (Score:3)
What you'd want to do is look at how the effectiveness of a unit (or rather a cont
Re: (Score:2)
Re: (Score:2)
Just simulate all the things.
It's what we do now, look at 500,000 player hours and say "lol no one above 10th percentile is using ice bolt oops we wasted time developing it"
Simulating them overnight lets you polish it before even launching. And save money from making QA slaves play for 5,000 hours.
Re: (Score:2)
A lot of the actions in SC2 is micromanaging units to avoid damage and keep units in perfect range all the while responding to your opponent doing the same. So an AI would still need to make all those quick actions to keep up.
Re: (Score:2)
But that presumes the gameplay between a perfect AI and a human is anywhere near similar. The "too good" AIs that run into range, shoot once, then run out of range before your defenses activate can twitch at that point until the lowest unit in the game has defeated the best. Those types of gameplay "hacks" might be found and eliminated, but those t
Re: (Score:2)
sure. past game data can be analyzed to create better games. nothing new.
but what is this "they"? do you mean human data analysts,using same dumb but fast computers? lol
ai - another name for data analysis.
this news actually means ibm's deepmind subsidiary got cooperation from blizzard to use their game datasets. rest is hype.
We all know the next logical step.. (Score:4, Insightful)
... from controlling units in a RTS to controlling drones in the field of war.
Personally I welcome our robotic overlords
Re: We all know the next logical step.. (Score:2)
Zerg rush! kekeke (Score:2)
I'm not sure if this is good or bad for the game itself. Currently Blizzard uses cheats for "AI" opponents and in the original Starcraft game, many bugs and quirks that rudimentary classifiers found useful were then replicated by humans to make the game rather linear and boring.
Re: (Score:3)
It takes a bit of talking at the EBGames store before the young ones realize I'm not there to buy something for a grandchild.
Re: (Score:2)
creimer has a similar problem when he goes to Victoria's Secret
Re: (Score:2)
The architects of our own doom (Score:2)
Things fall apart; the centre cannot hold;
Mere anarchy is loosed upon the world,
The blood-dimmed tide is loosed.....
Surely some revelation is at hand;
Surely the Second Coming is at hand.
And what rough beast, its hour come round at last,
Slouches towards Bethlehem to be born?
W.B. Yeats, The Second Coming
Not slouching towards Bethlehem but lurching ever closer to Judgment Day
Is this about Koreans? (Score:3)
Have we found, at long last, a possible way to beat the Koreans at StarCraft?
If so, will this Make America Great Again?
Re: (Score:2)
systemctl start maga
Re: (Score:2)
lol, no. The post isn't about Trump. I suppose the reason it seems incoherent to you is that you probably don't play StarCraft. It should have occurred to me that the article would be as likely to attract people interested in AI as much as StarCraft.
Mea Culpa, I should have provided some background.
Most StarCraft players know that the multi-million dollar professional StarCraft competitions are totally dominated by the Koreans, specifically Koreans from South Korea.
Here's some background.
https://en.wikipedi [wikipedia.org]
Re: (Score:2)
Using Trump's slogan makes it about Trump. So the Koreans are good at Starcraft? So what? Why bring Trump's slogan into it? Since when was America ever great at Starcraft? Since when did anyone but Koreans care about Starcraft? That game came out in the 90s for fuck's sake.
Still not about Trump.
Variations of "Make America Great Again" has been a staple of almost every politician running against an incumbent in the USA for quite some time. I've been hearing it for decades, and I can't imagine why anyone would be unaware of this slogan's long history.
I would not be surprised if they used it in the Adams-Jefferson election.
As for the questions about StarCraft, the article is about an AI that plays StarCraft. If the AI is going to be acknowledged to be any good, it will have to b
Re: (Score:2)
[citation needed]
1980:
http://www.iagreetosee.com/por... [iagreetosee.com]
I remember this one because I voted for the guy in 1980. GHW Bush also used that phrase when campaigning.
1991:
https://www.youtube.com/watch?... [youtube.com]
This is Clinton so it's not even a Republican phrase. It's just something people have been saying.
As for the rest of it, I'm not being coy. I wrote the post, and I know that I was not making a Trump reference when I wrote it.
After 16 presidential elections, Trump is just another one to me.
You can attempt to tell me what I meant w
Re: (Score:2)
Well it's worse than I thought.
I mentioned this thread to a friend in real life and found out that to a lot of people "Make America Great Again" is now code words for "Make America White Again", and saying that juxtaposed with beating Koreans in a video game only serves to hammer it home.
Well that's news to me, but nonetheless it's wrong to have said it.
I apologize for saying that.
wopr says (Score:2)
US first strike = SK mass damage and china unknown.
NK first strike = SK mass damage / Japan damage / 50-50 shot that 1 missile makes it all the way to the usa. China moves in to crush NK.
Re: (Score:2)
It'd be like a guy shining a flashlight in your face. Annoying, but not directly damaging. You can't shoot him for that. So what would the US response be to an EMP?
Re: (Score:2)
It's pretty obvious - a Korean nuke launched at the USA will mean an all-out attack on NK. That one is easy.
Personally I think the next Korean missile will be launched on a trajectory that makes it clear that NK can hit ALL of the USA. Just to make the point that trumpski can't feel too safe right now, just because he's relaxing on the East Coast.
Re: (Score:2)
You can't shoot him for that. So what would the US response be to an EMP?
Uh, it's more akin to pointing a laser at a driver and blinding him while at the wheel. People would die from that. The result would be war.
Even if we shot down the nuke before it hit a US territory (Guam) or ally would mean war. If NK demonstrates that they will shoot off nukes for whatever reason then all pretenses for diplomacy are off because they have demonstrated an act of nuclear aggression and we don't want to rely on THAAD to ensure millions don't die at the whims of dear leader.
can it play global thermonuclear war? (Score:3)
can it play global thermonuclear war?
What could go wrong? (Score:2)
So, we're going to develop and train AIs using a game that focuses on developing and processing resources, building weapons and defensive systems, and coordinating troops in an assault.
Yeah. That'll be fine.