Google's DeepMind AI Plans To Take On StarCraft II (venturebeat.com)

An anonymous reader quotes a report from VentureBeat: Google and Blizzard are opening up StarCraft II to anyone who wants to teach artificial intelligence systems how to conduct warfare. Researchers can now use Google's DeepMind A.I. to test various theories for ways that machines can learn to make sense of complicated systems, in this case Blizzard's beloved real-time strategy game. In StarCraft II, players fight against one another by gathering resources to pay for defensive and offensive units. It has a healthy competitive community that is known for having a ludicrously high skill level. But considering that DeepMind A.I. has previously conquered complicated turn-based games like chess and go, a real-time strategy game makes sense as the next frontier. The companies announced the collaboration today at the BlizzCon fan event in Anaheim, California, and Google's DeepMind A.I. division posted a blog about the partnership and why StarCraft II is so ideal for machine-learning research. If you're wondering how much humans will have to teach A.I. about how to play and win at StarCraft, the answer is very little. DeepMind learned to beat the best go players in the world by teaching itself through trial and error. All the researchers had to do was explain how to determine success, and the A.I. can then begin playing games against itself on a loop while always reinforcing any strategies that lead to more success. For StarCraft, that will likely mean asking the A.I. to prioritize how long it survives and/or how much damage it does to the enemy's primary base. Or, maybe, researchers will find that defining success in a more abstract way will lead to better results, discovering the answers to all of this is the entire point of Google and Blizzard teaming up.

Comments

One huge difference

[Solandri]

Games like Chess, Go, Tic-Tac-Toe always let both players see the complete world state. Armed with that knowledge, it's easy to be systematic and deterministic.

Games like Poker and Starcraft hide part of the world state from each player, forcing them to guess at the parts they can't see. That opens up the possibility of one player bluffing - leading the opponent down the wrong decision tree because he's fooled into thinking the part of the world state he can't see is different from what it really is. I don't think this is something an AI can "solve". Certainly one could optimize it, so that it becomes damn good at guessing when a certain player is bluffing or not. But put it up against a different player and all that "learned" experience becomes useless, or even counter-productive. Or even pit it against the same player who's aware he's playing against the AI which beat him last time, and he'll simply do something he would never normally do to throw off the computer. It's a difficult enough problem that in pretty much all commercial computer games with a fog of war feature, the computer is just programmed to cheat by ignoring the fog and seeing everything.

Re:One huge difference

[Kjella]

Playing optimally does not mean you win every time. Take for example Texas Hold 'Em, no matter how poor a hand you have pre-flop (worst is 7-2 off-suite) against the best (pair of aces) you still have about 11-12% chance depending on colors and flush draws if you just shove every time and never see a flop. If it's your one-in-a-million lucky day you could do that six times in a row and win every time. Every poker pro - and most amateurs too - will have some bad beat story where they did everything right and still lost big. But in the long run it should work most of the time.

That's what should happen if you play optimally with a fog of war too, you don't have one static strategy where you build the same every time, at least not if the game follows basic rock-paper-scissors rules that should never work. You have a variation that mostly beats your opponent's variation and you hedge and switch strategies on the fly as your opponent's strategy is revealed to you. And of course that includes bluffing, but that's part of the call or fold decision. Assuming your opponent never bluffs and always bluffs both lead to very poor strategies, they bluff part of the time and you call part of the time on the same board.

It's not like chess where in position X you always want to do Y. But it's a lot more like the real world where you don't have perfect information and that's kinda the point here to make AI that can function instead of humans in the real world. They won't necessarily be better than our best, but they'll be more consistent in not doing things that are clearly sub-optimal, like say rear ending the car standing still in front of us. I'm not a good RTS player. I notice quite often that I'm simply not keeping up with the action, those should have retreated, those should have pushed the attack, that position should have been reinforced, those troops I build didn't get any orders and so on. Just consistently using what you have in a good way probably beats any strategy I got.

Civilization

[carvalhao]

I would be really interested to see what would the results be if you would get DeepMind playing a game like Civilization, in which cooperation and soft-power can be used to win the game. That could really give all of us some hints on how to manage diplomacy/belligerence in a way that could lead to some interesting thought experiments in the real world.

Re:Civilization

[drinkypoo]

I would be really interested to see what would the results be if you would get DeepMind playing a game like Civilization, in which cooperation and soft-power can be used to win the game. That could really give all of us some hints on how to manage diplomacy/belligerence in a way that could lead to some interesting thought experiments in the real world.

Not unless you made a game like Civilization which was more than a glorified board game. Civilization bears only the slightest passing resemblance to reality. As such, you can only learn the most superficial lessons from it.

EULA

[Anonymous Coward]

So it's fine if Google does it, but if I do it, I get a ban for running a bot? Okay then.

Obligatory Archer meme...

[johnnys]

"teach artificial intelligence systems how to conduct warfare."

Do you want Skynet?

Because that's how you get Skynet!

"teaching itself"

[Bobtree]

> DeepMind learned to beat the best go players in the world by teaching itself through trial and error.

AlphaGo was trained on databases of historical games. It looks for moves that are similar to what a human pro would play, and then reads out sequences to score the strength of the resulting position. It did not learn by itself from scratch. Once proficient, it was played against itself to improve.

Re:Interesting

[psycho12345]

They are addressing this directly, by limiting the APM of the AI to human levels. So it won't be able to rely on perfect micro to win.