Google DeepMind's next target is Starcraft II

[Guest]

AI has taken many strides in beating humans at games, and one of the most famous examples was the IBM Deep Blue vs Garry Kasparov chess rivalry which made us look at AI in awe

A more recent example was when Google DeepMind beat Lee Se-dol in a game of Go last March, a feat that most if not all computer systems have difficulty with due to its complexity

Now AI has a new challenge ahead of it, Starcraft II, the fastest selling strategy game of all time

 

Starcraft II, Blizzard's RTS,  was released on July 27, 2010 and saw success with critics and consumers praising it from the then newbies to the veterans of Starcraft

6 years on, the game still has a large community of players around the world  

 

Google's DeepMind is one of the most well known AI in the world 

Starcraft II is one of the most popular and fiercely competitive games in the world

Together? Well, we're close to finding out how they will interact

 

Quote

"StarCraft is an interesting testing environment for current AI research because it provides a useful bridge to the messiness of the real-world," says Oriol Vinyals, a DeepMind research scientist who was once the top-ranked StarCraft player in Spain. "The skills required for an agent to progress through the environment and play StarCraft well could ultimately transfer to real-world tasks." Creating strong StarCraft AIs would obviously also have major implications for the game itself, with the potential for customized virtual coaches to raise the level of play across the board.

 

Though the developers of DeepMind look forward to winning the battle they still see many problems to overcome

 

Quote

StarCraft is a complex strategy game that involves resource management, scouting, and battle tactics. Unlike Go, players are unable to see the entire field of play at once, meaning they have less information to work with when planning their moves; it’s a test of memory and adaptation as much as it is long-term planning. "The thing about Go is obviously you can see everything on the board, so that makes it slightly easier for computers," DeepMind founder Demis Hassabis told me earlier this year, in an interview that Blizzard executive producer Chris Sigaty says "got some discussions going" between the two parties.

 

So what's the difference between Starcraft's AI and Google's?

Well, Starcraft's AI can see the whole playing field

Meanwhile Google will play it just like any player, it will need to scout the areas so that it gains visibility on the playing field 

 

Quote

Blizzard and DeepMind both caution that we’re far away from the day where an AI will be able to defeat a top-ranked StarCraft player. Neither would offer a rough date when asked, though it’s worth pointing out that AlphaGo’s victory came a lot sooner than most observers predicted.

 

Though we won't see AI try to beat a top-ranked player any day soon, it's an exciting prospect for gamers and AI researchers alike

It's definitely an awesome spectacle to see games become the battleground for AI research

I'm personally looking forward to seeing AI's capabilities and I'm fairly optimistic towards the technologies advancements 

 

The battle of Man vs Machine.....to be continued 

 

Sources: http://www.theverge.com/2016/11/4/13518210/deepmind-starcraft-ai-google-blizzard

https://deepmind.com/blog/deepmind-and-blizzard-release-starcraft-ii-ai-research-environment/

[Maslofski]

if this'll lead to better game ai then i'm all for it

[Sauron]

Good luck... starcraft 2 is a lot more complex than chess from a computational standpoint, way too many variables.

[tlink]

27 minutes ago, Sauron said:

Good luck... starcraft 2 is a lot more complex than chess from a computational standpoint, way too many variables.

computers and AI have also become way more advanced and complex. the computer used back then is probably as fast as my high school graphical calculator.

[Sauron]

1 minute ago, tlink said:

computers and AI have also become way more advanced and complex. the computer used back then is probably as fast as my high school graphical calculator.

1996, so no. Still, even the one that plays go is probably not fast enough for this. The fundamental difference between a board game and a videogame is that in the second the rule set is a lot wider and allows for plenty more variation by the player's hand. You don't have a predetermined set of squares you can move units to, nor are the possible paths set in stone. Maybe you can get it to beat the campaign though.

[AresKrieger]

Honestly this isn't going to be as impressive as alpha go since starcraft is at least 50% speed of action as a determining factor

[Bouzoo]

41 minutes ago, Sauron said:

Good luck... starcraft 2 is a lot more complex than chess from a computational standpoint, way too many variables.

To be fair, there are guides and strategies which make most variables obsolete and just point it down on how fast you can evolve. It is complex but people have dumbed it down a lot. 

[dalekphalm]

8 minutes ago, Sauron said:

1996, so no. Still, even the one that plays go is probably not fast enough for this. The fundamental difference between a board game and a videogame is that in the second the rule set is a lot wider and allows for plenty more variation by the player's hand. You don't have a predetermined set of squares you can move units to, nor are the possible paths set in stone. Maybe you can get it to beat the campaign though.

It'll be interesting to see how it does. On one hand, I'm going to assume DeepMind will have difficulty with intuition and strategy/tactics, that humans are innately good at.

 

On the other hand, it's reaction time and actions per minute (or even per second) will be able to far surpass even the best SC2 Pro gamers. And as anyone who's ever watched the big pro tournaments knows, a single moment of good reaction can make all the difference, and can turn an obvious loss into a surprise win.

[tlink]

1 minute ago, Sauron said:

1996, so no. Still, even the one that plays go is probably not fast enough for this. The fundamental difference between a board game and a videogame is that in the second the rule set is a lot wider and allows for plenty more variation by the player's hand. You don't have a predetermined set of squares you can move units to, nor are the possible paths set in stone. Maybe you can get it to beat the campaign though.

humans don't need those variables to win, so why would a computer? winning is not about seeing all the options, you just have to see one step ahead of your opponent, if he doesn't need all variations than neither does the AI. 

[Sauron]

4 minutes ago, Bouzoo said:

To be fair, there are guides and strategies which make most variables obsolete and just point it down on how fast you can evolve. It is complex but people have dumbed it down a lot. 

Guides give you a general idea of what people are doing and consider effective, but they don't tell you "place this building here. Move that unit there. Attack this." and they only apply to the early game, what happens beyond that depends on what your opponent is doing (and good players will tell you that even the early game is highly variable depending on what the other guy does).

4 minutes ago, dalekphalm said:

a single moment of good reaction can make all the difference, and can turn an obvious loss into a surprise win.

That's assuming the guy with the good reactions is able to capitalize on his momentary advantage.

[dalekphalm]

Just now, Sauron said:

Guides give you a general idea of what people are doing and consider effective, but they don't tell you "place this building here. Move that unit there. Attack this." and they only apply to the early game, what happens beyond that depends on what your opponent is doing (and good players will tell you that even the early game is highly variable depending on what the other guy does).

That's assuming the guy with the good reactions is able to capitalize on his momentary advantage.

True, but we've already seen DeepMind is able to successfully employ strategy with other games, like Chess and Go.

 

I doubt it's gonna stomp pubs immediately, but after learning the game, I would not be surprised if it's able to capitalize on those advantages. Only time will tell! I'd be interested in watching Day[9] and TB cast a match of DeepMind vs whoever is the "favourite" for best SC2 player at the moment

[Sauron]

3 minutes ago, tlink said:

humans don't need those variables to win, so why would a computer? winning is not about seeing all the options, you just have to see one step ahead of your opponent, if he doesn't need all variations than neither does the AI. 

First of all, your brain is a lot more powerful than you give it credit for. You process hundreds of variables all the time without even noticing. When you think to yourself "hmm, this looks like a good time to expand" you have subconsciously taken into account a huge variety of factors. You have also translated what you see in detailed information. An AI doesn't come with this sort of ability built in. It has to be programmed (or learn) to do it. "Just" seeing one step ahead of your opponent is exactly the sort of task an AI would struggle with.

[Sauron]

Just now, dalekphalm said:

True, but we've already seen DeepMind is able to successfully employ strategy with other games, like Chess and Go.

 

I doubt it's gonna stomp pubs immediately, but after learning the game, I would not be surprised if it's able to capitalize on those advantages. Only time will tell! I'd be interested in watching Day[9] and TB cast a match of DeepMind vs whoever is the "favourite" for best SC2 player at the moment

Meanwhile, we have this pure gold:

 

[dalekphalm]

3 minutes ago, Sauron said:

Meanwhile, we have this pure gold:

 

Lmfao is that an AI vs AI casted match? Hells yes, I'm saving that tab for later today.

[Sauron]

1 minute ago, dalekphalm said:

Lmfao is that an AI vs AI casted match? Hells yes, I'm saving that tab for later today.

They have 4 or 5 in the playlist  it's hilarious

[dalekphalm]

Just now, Sauron said:

They have 4 or 5 in the playlist  it's hilarious

Well... I know what I'm gonna do today at work instead of working

[SpaceGhostC2C]

1 hour ago, Sauron said:

Good luck... starcraft 2 is a lot more complex than chess from a computational standpoint, way too many variables.

I'd like to point out: we still don't have a fast enough computer to "solve chess" in any reasonable amount of times (of course, neither can humans, hence why computers can win, and why we even continue  playing chess to begin with).

[tlink]

12 minutes ago, Sauron said:

First of all, your brain is a lot more powerful than you give it credit for. You process hundreds of variables all the time without even noticing. When you think to yourself "hmm, this looks like a good time to expand" you have subconsciously taken into account a huge variety of factors. You have also translated what you see in detailed information. An AI doesn't come with this sort of ability built in. It has to be programmed (or learn) to do it. "Just" seeing one step ahead of your opponent is exactly the sort of task an AI would struggle with.

i would completely disagree with that, an AI has direct access to the memory registers in your computer. we have an inferior interface, an AI can directly look into RAM to see whats going on. and just because your brain handles more inputs doesn't make those inputs any more relevant. if there are 6 extremely important factors that we are kinda okay at, and 20 not really relevant factors that we're good at, and the ai is amazing at those 6 important factors, while its really pretty bad at those 20 other factors. complexity by obscurity is not really complexity.

[CommandMan7]

As far as strategy games go, Starcraft 2 is a good starting point but I would think it's on the lower end of complexity. I think a more difficult challenge would be playing the sorts of games that have multiple paths to victory, so the AI would have to evaluate their enemy and choose the correct path rather than just compete on a single static path every time. I can't think of any examples except Civilization (5 ways to win) and AotS (2 ways to win).

[dalekphalm]

2 minutes ago, CommandMan7 said:

As far as strategy games go, Starcraft 2 is a good starting point but I would think it's on the lower end of complexity. I think a more difficult challenge would be playing the sorts of games that have multiple paths to victory, so the AI would have to evaluate their enemy and choose the correct path rather than just compete on a single static path every time. I can't think of any examples except Civilization (5 ways to win) and AotS (2 ways to win).

Well any 4X game would have the same alternate paths as Civ and AOTS.

[Sauron]

51 minutes ago, tlink said:

i would completely disagree with that, an AI has direct access to the memory registers in your computer. we have an inferior interface, an AI can directly look into RAM to see whats going on. and just because your brain handles more inputs doesn't make those inputs any more relevant. if there are 6 extremely important factors that we are kinda okay at, and 20 not really relevant factors that we're good at, and the ai is amazing at those 6 important factors, while its really pretty bad at those 20 other factors. complexity by obscurity is not really complexity.

That's cheating, not what this is supposed to do. The whole idea is to have it play with the exact same resources as a human. Interpreting on screen movement instantly is something we're extremely good at and is perhaps the most impactful thing in the game.

[samcool55]

Also apart from intelligence, are they going to give it superhuman reflexes?

If it's an AI it can have a much higher APM than a human.

[DXMember]

record a couple of championships, analyze the strategies used

execute them at insane APM - gg no re

I don't think it's any harder than what modern artificial "intelligence" is doing in chess lol

 

I believe a videogame is both more favorable and challenging place for AI to be because of the real-time aspect, unlike a turn based board game you have limited time for calculations but unlike a turn based board game your human oponent also has to react much quicker

[Sauron]

5 minutes ago, DXMember said:

record a couple of championships, analyze the strategies used

execute them at insane APM

I don't think it's any harder than what modern artificial "intelligence" is doing in chess lol

Chess engines use an insane amount of recorded matches to be any good. Executing the same strategy over and over just makes you exploitable, sc2 is not rock/paper/scissors.

[DXMember]

1 minute ago, Sauron said:

Chess engines use an insane amount of recorded matches to be any good. Executing the same strategy over and over just makes you exploitable, sc2 is not rock/paper/scissors.

it's a matter of having enough flexible strategies at hand that you lose the apparent patterns ))