Brood war is real time strategy not turn based, also this is not a chessboard...maps aren't small square with highly limited amount of possibilities...
The AI would simply fail at the scouting part... which is basically the most important. Fog of war implies you need to think, not calculate. It would not take long until a human recognize patterns and gets his way...
The macro aspect isn't really relevant, it's been a long time progamers can manage an entire game without having more than 500 minerals even on 5 bases assuming he isn't maxed out... While I agree AI can have the edge on that, it is in no way a game changer...
On top of that there is no such ultimate strategy in brood war that allows you to counter anything. If you choose strategy A it will protect you from strategy B and C but not from D. There is no escape, if you scout or choose wrong, you are dead no matter what. Statistic and probability won't help coze at the time you made your choice you did not have a chance to scout. And this is the definition of real time strategy, if the AI can't scout, what is it gonna do? Guess? No it won't guess anything, it will simply act based on the information it has, human will quickly catch up, and what's next? EXPLOIT.
I can only imagine proud scientist bringing the so called ultimate AI zerg vs Flash just to die in 3min from rush bunker... Imagine the length of an algorithm that need to anticipate "hold lurker" good luck...
I believe they are still plenty of arguments, I can't predict future and everything is possible, but present day, come on...
On March 13 2016 04:01 Eternal Dalek wrote: I would be impressed by an AI that can win in DotA, where decision-making matters far more than mechanics. In Starcraft, it would be no contest. We've seen all the macro and micro hacks and bots over the years, and they're unstoppable by most players. Bear in mind that these hacks weren't made by Google, who has pretty much unlimited money to spend on development. Imagine a brilliant hacker who doesn't have to worry about paying his/her bills, free to devote everything into AI development.
That's what we're up against, and I doubt the Dominion will win.
Hey, Dustin Browder, i just had an idea for Starcraft 3: Bio vs. Mech. Terran, Zerg, and Protoss vs. an unstoppable race of intelligent machines.
There are some disadvantages to DotA though: there are still patch changes, so that you can't easily train a bot on one specific patch; and it's not so easy for an AI to mass practice games vs itself since you might be stuck with Steam or with 45 mins per game and those sort of things.
And there's the question of whether you should have an AI control all five heroes at once, since that might be seen as cheating since it'll have perfect coordination.
It would be fair for the same AI control all 5 heroes, however, it would be necessary for it to be 5 separate instances of the same program. Otherwise, it wouldn't be fair at all.
I would be extremely impressed if an AI team could even beat 5 random 4k players thrown together. It's not going to happen for a really long time. Like... we'll all be dead.
On March 13 2016 06:07 NonY wrote: Unlike turn-based board games, where inputting moves is trivial and thus the method can be ignored, playing SC is intrinsically tied to keyboard and mouse control. If the AlphaGo team wants to tackle SC, then they have a significant robotics challenge in front of them that I'm not really sure is going to be worth their time as AI researchers. It's always a bad idea to bet against technology when technology is allowed unlimited time to develop, but SC presents some very significant increases in difficulty just for the AI, robotics aside. It's far more complex because in addition to a "mirror match" you've got to be able to beat two completely different sets of "game pieces" and there isn't just one simple game board. And after all that, games can hinge on luck like in a poker game. The human can pick randomly, like glance at his mineral count and do one extreme if it ends in an even number and do another extreme if it ends in an odd number, and there might simply be no solution for both possibilities. Avoiding all such situations seems unlikely. Because of this, it could possibly be a top player if it avoids predictability, but it seems just as likely as a poker AI to consistently win tournaments. Nonetheless I'm excited to see how it progresses. I wonder if the Korean BW players have a renewed sense of purpose seeing as how an AI might be entering one of their tournaments someday.
While I agree there *could* be a degree of robotics involved with this, it's much more likely they would just write (or find) drivers to run off of virtual keyboard/mouse.
I'm not sure I understand why professional players think they could beat AI (even today). Just rotating through a few timing attacks that deny enough scouting to make the player coin flip a defend would make the AI pretty hard to consistently beat; especially given perfect micro potential.
On March 13 2016 06:41 summerloud wrote: i think boxer has no idea what hes talking about. it would prolly be possible to program an AI that would beat everyone just using blink stalkers
Broodwar doesn't have blink stalkers. Even then the AI has to get there in either an even position or ahead, you can have perfect blink stalker micro, won't matter if your opponent just has a lot more units/economy.
This question has an easy answer, if they cap eapm it can't beat humans. Taking for granted the AI needs to use a cursor and the same kind of tools as a human. If they the don't cap eapm the ai will beat humans easy.
The issue is that there are so many ways to abuse an AI, the number of variables means that if the AI has one single weakness, then humans can exploit. In Go and chess, there are much less options. Even trying to design an AI that can survive a perfect cannon rush would probably take years.
As long as AI matches are made available for players to analyse, then human players will beat the AI (which is fair given that the AI developer has access to pro games). Maybe in 50 years time, an true AI which can learn as it plays can win, but right now, the AI still very much depends on human programming, hence there will be tons of weaknesses.
Why does everyone assume computers would have perfect micro? I think it was Letabot in the other thread who said that micro was one of the hardest parts of building an AI. Just because a computer might be able to input commands without execution errors, the actual decision element of microing units isn't at all trivial, nor would it generalize well, I think.
I'm curious how back propagation works in deep mind, if anyone knows. Like, for example in a game like Go, is it possible to evaluate a move and do reinforcement/error correction without having to wait until the final result of the game?
On March 13 2016 06:41 summerloud wrote: i think boxer has no idea what hes talking about. it would prolly be possible to program an AI that would beat everyone just using blink stalkers
Only if an AI can survive a cannon rush/bunker rush/nydus! An AI would be much worst at scouting, and you could easily dupe it by hiding stuff or faking stuff. It will take years to develop an AI that can consistently hold off the most basic rushes.
On March 13 2016 07:09 nbaker wrote: Why does everyone assume computers would have perfect micro? I think it was Letabot in the other thread who said that micro was one of the hardest parts of building an AI. Just because a computer might be able to input commands without execution errors, the actual decision element of microing units isn't at all trivial, nor would it generalize well, I think.
I'm curious how back propagation works in deep mind, if anyone knows. Like, for example in a game like Go, is it possible to evaluate a move and do reinforcement/error correction without having to wait until the final result of the game?
I don't think most people know what perfect micro means. In order to have perfect micro you need to calculate the Nash Equilibrium each fight (at every single frame). With only 50 ms per frame, there is no way computers today can calculate the perfect micro.
There is one aspect involving micro-management where bots are better than human players. Since the default mining from worker units is not optimal, it is possible to calculate a better path for the worker units to take. Executing this in a pixel perfect matter will shave off several milliseconds of travel time each trip. Example video below:
On March 13 2016 06:41 summerloud wrote: i think boxer has no idea what hes talking about. it would prolly be possible to program an AI that would beat everyone just using blink stalkers
Only if an AI can survive a cannon rush/bunker rush/nydus! An AI would be much worst at scouting, and you could easily dupe it by hiding stuff or faking stuff. It will take years to develop an AI that can consistently hold off the most basic rushes.
An AI could go the route of a safe build, holding an all in is hard because you are greedy and you are greedy to get an economical macro edge. An AI doesn't need an eoconomical advantage if they have perfect macro and micro.
Also you are all underestimating the possiblity of what an AI can do, as soon as something is in vision an AI could calculate exactly how to micro to get away or evade. For example you say an AI would be worse at scouting, I think thats pretty funny because imagine getting rid of a scouting probe/drone/scv that microes perfectly in your base. It could survive for ages especially if its a probe, it could surive against msc and zealot and only get taken out when you make a stalker or blowing an overcharge. It can micro against slow lings and a queen forever as well. The reason pros doesn't do this is because it takes too much energy and time for too lite gain, an AI would get all those small edges that pros doesn't have time to prioritize.
This is so much different than a boardgame though. Boardgames are paused states, while Starcraft is time dependant, and humans are limited by having to use our hands and eyes to play the game.
I mean it might beat Boxer if it's allowed crazy apm, and reading data from the game so it can do inhuman micro and multitasking. But is that really fair if it has access to more data input and a way faster and more precise output than the real life player?
Honestly it would only be a fair beat if they had to use visual recognition software on an actual monitor for input data, and some sort of mechanically automated mouse and keyboard to click on stuff and scroll the screen.
I think it's a ways off before an AI would be able to beat a top player in a BoX. Strategy in real time is a lot different from turn based strategy, it needs to make optimal decisions in a split second with limited information. Mechanically the AI will have a pretty big advantage even if it's APM is capped at say 400-500 because it will have the most efficient APM of any player and can act across the whole map not just on a single screen within microseconds if necessary, which is an unfair advantage over 'real' players. Obviously AI will eventually get there though, and by the time DeepMind is ready to make such a challenge it will probably have surpassed humans.
EPT
