EPTit should be Flash in SC1
DeepMind sets AlphaGo's sights on SCII - Page 2
| Forum Index > SC2 General |
|
BeStFAN
483 Posts
it should be Flash in SC1 | ||
|
Taf the Ghost
United States11751 Posts
On March 28 2016 14:35 lpunatic wrote: BW AIs with no APM limits still can't beat decent humans, despite super muta micro and whatnot. Well, Google also dumped somewhere in excess of several million dollars worth of equipment and probably more money in salaries at the issue. To the point that it wouldn't surprise me if the total "Man-Hours" applied to producing a Go-playing AI wasn't actually more total hours than it took for the Master to learn the game. (While it's impressive that it won, you can do a lot of impressive things with massive money expenditures.) Though I did come up with a funny thought: what if someone only agreed to play the AI if it had to actually physically use a Keyboard & Mouse? Be sort of funny to see the complaints about insisting on that. | ||
|
ZigguratOfUr
Iraq16955 Posts
On March 28 2016 14:35 lpunatic wrote: BW AIs with no APM limits still can't beat decent humans, despite super muta micro and whatnot. The people programming the BW AIs don't have Google's resources. Besides the goal really isn't to win at Starcraft, it's to develop AI research. They want to find new approaches and algorithms that would enable a computer to "out-think" humans at Starcraft. Figuring out ways to abuse the AIs mechanical advantages isn't all that meaningful. | ||
|
BronzeKnee
United States5218 Posts
On March 28 2016 14:35 Taf the Ghost wrote: Thus, the first thing an "AlphaStar" would need to learn would be how to use the mouse. Haha. Yeah this is what I argued in the other thread. The AI would need the constraints of a mouse, keyboard and monitor or it is just cheating. The mouse, keyboard and monitor makes our play quite sloppy and limit our vision. I still think the more interesting match is to let humans control the game with their mind and let the AI face off against that. Imagine perfect forcefields, and suddenly they appear. The only constraint would be the monitor (the AI wouldn't be able to view more than one area of the map at a time). I can't see the human ever losing in my lifetime with that constraint, the human mind would be too fast, too perfect, too powerful. | ||
|
purakushi
United States3301 Posts
Still wish it is BW, but since Blizzard is involved, they would obviously push for SC2. | ||
|
Die4Ever
United States17721 Posts
On March 28 2016 15:07 BronzeKnee wrote: I can't see the human ever losing in my lifetime with that constraint, the human mind would be too fast, too perfect, too powerful. This AI already beat Go world champions, that's a turn based game with no mechanics. | ||
|
BeStFAN
483 Posts
On March 28 2016 15:07 BronzeKnee wrote: I can't see the human ever losing in my lifetime with that constraint, the human mind would be too fast, too perfect, too powerful. hahahahaha | ||
|
Valen55
2 Posts
| ||
|
insitelol
845 Posts
| ||
|
JimmyJRaynor
Canada17151 Posts
On March 28 2016 13:58 kingjames01 wrote: That's a weird way to defend a point... Are they playing Chess, Go or StarCraft? here is why Alpha will have a tough time without heuristics in a game like Chess or Starcraft https://webdocs.cs.ualberta.ca/~cdavid/pdf/aiide12-combat.pdf years from now .. who knows.. maybe no heuristics are required. but for now.. Alpha will have a tough time. | ||
|
plogamer
Canada3132 Posts
And frankly, I don't think increase apm will make too much difference as long as it's not unreasonably restrictive. | ||
|
FFW_Rude
France10201 Posts
| ||
|
Mendelfist
Sweden356 Posts
On March 28 2016 15:39 insitelol wrote: In Go or Chess you can go the extensive way, just loading in all possible situations into the memory and the only thing AI should do in that case is to choose between a relatively small amout of choices. You don't know what you are talking about. This doesn't work for Go, and this is why Alphago is a breakthrough. [B]There are a lot of situations in this game where you literally have to guess, and this is where common sense comes to help, which is obviously a problem for any AI. So i guess the only possible solution for this whould be working on some common "principles" or "basics" for AI to persue while playing. Yes that is a bit artificial as these "rules" would not be of AI origin, but this is how i see this. The problem is the same for Go, but they did NOT give Alphago a set of common principles. It has discovered these by itself. | ||
|
insitelol
845 Posts
On March 28 2016 15:53 Mendelfist wrote: You don't know what you are talking about. This doesn't work for Go, and this is why Alphago is a breakthrough. The problem is the same for Go, but they did NOT give Alphago a set of common principles. It has discovered these by itself. Hm, ok, thanks. Didn't know that. It is hard to imagine though, i guess i should dig this a bit. edit: oh, its ANN right, should have guessed. So she whould accumulate these "principles" from enldess replay/vod base. It solves the "common sense" problem at least partially. Would be fun to watch. | ||
|
OtherWorld
France17333 Posts
| ||
|
TheoMikkelsen
Denmark196 Posts
All you need is to make AlphaGo Terran, give it medivacs, stim, and the ability to micro each individual marine perfectly. Seriously it should be the easiest task ever. Just make it like open reaper expand every game, then it only has to counter some cheeses. Go does not have mechanics, Starcraft 2 does, and there the AI would be infinitely superior. | ||
|
mierin
United States4943 Posts
[QUOTE]On March 28 2016 14:35 Taf the Ghost wrote: I can't see the human ever losing in my lifetime with that constraint, the human mind would be too fast, too perfect, too powerful. [/QUOTE] Isn't that the real issue here, though? Human brains are powerful BECAUSE they can think quickly, but the fact that we can think quickly means we're okay with being wrong if we come to a decision fast enough. The computer is on a totally different spectrum...it takes a long time to come to a conclusion, but in its mind it's always the move that has the highest probability of winning. The whole clash of fundamental ideals there is what makes watching AI vs. humans so interesting to me...really a 2nd race, if we could make the comparison. | ||
|
spinesheath
Germany8679 Posts
On March 28 2016 15:07 BronzeKnee wrote: I can't see the human ever losing in my lifetime with that constraint, the human mind would be too fast, too perfect, too powerful. China already has a supercomputer that has about the same computing power as the estimated computing power of the human brain. In about 10-20 years this computing power will cost like a couple thousand dollars (this is just based on a continuation of moore's law, which has held true for a very long time). On March 28 2016 17:04 mierin wrote: Isn't that the real issue here, though? Human brains are powerful BECAUSE they can think quickly, but the fact that we can think quickly means we're okay with being wrong if we come to a decision fast enough. The computer is on a totally different spectrum...it takes a long time to come to a conclusion, but in its mind it's always the move that has the highest probability of winning. The whole clash of fundamental ideals there is what makes watching AI vs. humans so interesting to me...really a 2nd race, if we could make the comparison. You can easily code an AI to take say 1 second to come up with a bunch of potential moves and then pick the one that looks the most promising so far. By no means does it have to evaluate all possiblities or any such nonsense. Obviously you can tinker with that decision making process any way you want; there could be thousands of decisions for different aspects of the game being evaluated in parallel. Also humans don't think quickly when compared to computers. Brain cells operate at a much lower frequency as a modern cpu and signal transmission between cells is much slower than on a cpu. | ||
|
hellokitty[hk]
United States1309 Posts
| ||
|
mierin
United States4943 Posts
On March 28 2016 17:09 spinesheath wrote: China already has a supercomputer that has about the same computing power as the estimated computing power of the human brain. In about 10-20 years this computing power will cost like a couple thousand dollars (this is just based on a continuation of moore's law, which has held true for a very long time). You can easily code an AI to take say 1 second to come up with a bunch of potential moves and then pick the one that looks the most promising so far. By no means does it have to evaluate all possiblities or any such nonsense. Obviously you can tinker with that decision making process any way you want; there could be thousands of decisions for different aspects of the game being evaluated in parallel. Also humans don't think quickly when compared to computers. Brain cells operate at a much lower frequency as a modern cpu and signal transmission between cells is much slower than on a cpu. I think we're kind of defining 'quickly' differently. Either way I think we agree with each other. | ||
| ||
