|
http://sports.sina.com.cn/go/2016-03-27/doc-ifxqswxn6442945.shtml
It looks like we have confirmation that DeepMind will attempt to enter into the StarCraft arena next.
As of this moment, there are no details as to how long AlphaGo (AlphaStar?) will need before it's ready. There are also no representatives from the professional StarCraft II scene named as of yet.
Against Lee Sedol, Google had AlphaGo using 1920 CPUs and 280 GPUs. It remains to be seen how much more computing power Google will devote to this next challenge and the effect that large jump in resources has in AlphaGo's decision-making ability.
Edit: Updated hardware usage.
|
The real-time aspect will be critical. Go, being turn-baseed, is purely a matchup of mind against "mind". If They allow the machine unlimited APM, then this won't be much of a match.
Also, you'd need representatives from all 3 races, wouldn't you?
|
On March 28 2016 12:53 Circumstance wrote:
The real-time aspect will be critical. Go, being turn-baseed, is purely a matchup of mind against "mind". If They allow the machine unlimited APM, then this won't be much of a match.
Also, you'd need representatives from all 3 races, wouldn't you?
And if you don't allow the machine to have arbitrary APM, what APM will you allow? The choice is completely arbitrary.
|
On March 28 2016 13:04 Fran_ wrote:Show nested quote +On March 28 2016 12:53 Circumstance wrote:
The real-time aspect will be critical. Go, being turn-baseed, is purely a matchup of mind against "mind". If They allow the machine unlimited APM, then this won't be much of a match.
Also, you'd need representatives from all 3 races, wouldn't you? And if you don't allow the machine to have arbitrary APM, what APM will you allow? The choice is completely arbitrary.
It could match its opponent's APM
|
On March 28 2016 13:04 Fran_ wrote:Show nested quote +On March 28 2016 12:53 Circumstance wrote:
The real-time aspect will be critical. Go, being turn-baseed, is purely a matchup of mind against "mind". If They allow the machine unlimited APM, then this won't be much of a match.
Also, you'd need representatives from all 3 races, wouldn't you? And if you don't allow the machine to have arbitrary APM, what APM will you allow? The choice is completely arbitrary.
The only way I see APM being limited would be if you did the following
1) Choose a designated 'esports keyboard' (some razer/steelseries/ect product)
2) Find the maximum amount of inputs the keyboard recognizes per minute
3) Make that the APM 'limit'
Each keyboard will stop responding after a certain number of inputs per second, and said input limit is probably much lower than the input limit that AlphaGo would have. At least I assume so.
|
Kinda disappointing they went for SCII instead of BW. BW is much more stable than the current version of SCII, especially that this is a new expansion that was out for only couple of months compared to 18 years of a stable version, heck, BW even have a stable map pool (or even a most single used map). It would have been more suitable for the AI to train on instead of a changing map pool and balance updates from time to time.
|
On March 28 2016 13:04 Fran_ wrote:Show nested quote +On March 28 2016 12:53 Circumstance wrote:
The real-time aspect will be critical. Go, being turn-baseed, is purely a matchup of mind against "mind". If They allow the machine unlimited APM, then this won't be much of a match.
Also, you'd need representatives from all 3 races, wouldn't you? And if you don't allow the machine to have arbitrary APM, what APM will you allow? The choice is completely arbitrary.
I don't think an apm cap would be completely arbitrary. Somewhere around the average apm of top pros or maybe a little higher I think is reasonable.
|
1001 YEARS KESPAJAIL22272 Posts
Choosing to cap APM based on top pros makes sense because it makes the measurement a lot more meaningful.
We already know that machines can have an unbelievably high APM. Clearly, machines can out micro humans. However, what we want to know is if machines can out-think humans. Capping APM based on human ability makes sense because it adds another layer of decision-making. What will the computer choose to do with that APM? Will some of that APM be wasted on micro with minimal gains? Or will the computer choose more efficient ways to use a limited resource? We could learn a lot if we limit APM that we would not learn if we let the machine use as many commands as possible.
APM is a limited resource in strategy games, just as turn based games have 1 move per turn. Having unlimited APM defeats the purpose of testing a computer's ability to learn, adjust, and out-think a human being.
|
Having the APM based on top pros would be pretty advantageous for DeepMind. Pros are trying to do things as fast as possible, not with the least APM possible, so there is obviously a large amount of "wasted" APM (though for a human it isn't wasted since it isn't like you could use it elsewhere). So a computer with a certain amount of APM can do a lot more than a human with the same APM.
A computer who is as "smart" as a human should be able to get by with less APM though exactly how much is hard to evaluate. Also computers can't misclick.
The more I think about it the more I'm surprised that Starcraft was chosen. Since there's so many mechanical advantages that a computer can leverage that are very hard to control for, any win by a computer will be beset by complaints and excuses, legitimate or not, that the computer isn't actually strategically or tactically superior.
|
with APM limit it will be quite a tough challenge to beat the top pros
|
without heuristic functions i think AlphaStar will have trouble "learning" the game.
and with heuristic functions it'll play with the style of dictated by them.
|
On March 28 2016 13:39 JimmyJRaynor wrote:
without heuristic functions i think AlphaStar will have trouble "learning" the game.
and with heuristic functions it'll play with the style of dictated by them.
That was the prevailing stance before AlphaGo fought Lee Sedol... It didn't hold up to reality though.
|
APM is not the point. A large portion of human APM is meaningless and just for warming up purpose toward spikes in big fights.
SC is a totally different game to Go. People call it not "perfect information", meaning AI does not know exactly what its opponent doing in a SC game, not as completely as black and white stones on a Go board.
Suppose two medivacs are approaching the zerg AI's 2nd base, the AI does not know what's inside, marines, marines and mines, or empty? AI's need to predict the possible drop location, it is main base, or 2nd base, or one for each? All the possible scenarios requires defense strategies accordingly with consideration of limited resource and efficiency. All these scenarios are developing in real time and could shifting from each other within milliseconds as the medivacs make a boosted turn.
|
It scares me if I match up with a guy named deepmind or alphastar on ladder.
|
On March 28 2016 13:46 kingjames01 wrote:Show nested quote +On March 28 2016 13:39 JimmyJRaynor wrote:
without heuristic functions i think AlphaStar will have trouble "learning" the game.
and with heuristic functions it'll play with the style of dictated by them. That was the prevailing stance before AlphaGo fought Lee Sedol... It didn't hold up to reality though.
who is the winner?
the best Chess AI with heuristics or Alpha?
|
On March 28 2016 13:50 JimmyJRaynor wrote:Show nested quote +On March 28 2016 13:46 kingjames01 wrote:On March 28 2016 13:39 JimmyJRaynor wrote:
without heuristic functions i think AlphaStar will have trouble "learning" the game.
and with heuristic functions it'll play with the style of dictated by them. That was the prevailing stance before AlphaGo fought Lee Sedol... It didn't hold up to reality though. who is the winner? the best Chess AI with heuristics or Alpha?
That's a weird way to defend a point...
Are they playing Chess, Go or StarCraft?
|
I would like to see the ai learn without any APM restraints to see what kind of micro it learns is the best. Will it learn when it's best to stand and fight for a trade and when it's best to run away?
I think limiting the APM puts a limitation on the ai that it shouldn't have. The question shouldn't be whether or not a human can beat a handicapped ai, it should be wether ai can leverage all of it's advantages to beat humans just as how humans will leverage their advantages to beat the ai.
|
On March 28 2016 14:22 ThunderBum wrote:
I would like to see the ai learn without any APM restraints to see what kind of micro it learns is the best. Will it learn when it's best to stand and fight for a trade and when it's best to run away?
I think limiting the APM puts a limitation on the ai that it shouldn't have. The question shouldn't be whether or not a human can beat a handicapped ai, it should be wether ai can leverage all of it's advantages to beat humans just as how humans will leverage their advantages to beat the ai.
Except with no APM limitations you don't really need any strategy or intelligence to beat humans. Just program it with a few rushes that will almost always succeed due to perfect micro.
|
On March 28 2016 14:30 ZigguratOfUr wrote:Show nested quote +On March 28 2016 14:22 ThunderBum wrote:
I would like to see the ai learn without any APM restraints to see what kind of micro it learns is the best. Will it learn when it's best to stand and fight for a trade and when it's best to run away?
I think limiting the APM puts a limitation on the ai that it shouldn't have. The question shouldn't be whether or not a human can beat a handicapped ai, it should be wether ai can leverage all of it's advantages to beat humans just as how humans will leverage their advantages to beat the ai. Except with no APM limitations you don't really need any strategy or intelligence to beat humans. Just program it with a few rushes that will almost always succeed due to perfect micro.
BW AIs with no APM limits still can't beat decent humans, despite super muta micro and whatnot.
|
The Mouse-movement and Keyboard entry limitations would have to be imposed. Giving a computer full API access is to cheat, as the API could do things no mouse is actually capable of. We've seen some of the custom AIs over the years, but those functionally produce a new unit-movement control that goes around the game's actual functionality.
Thus, the first thing an "AlphaStar" would need to learn would be how to use the mouse. Haha.
|
|
|
|
|
|
EPT
