|
On March 14 2016 04:47 Big J wrote:Show nested quote +On March 14 2016 04:15 necrosexy wrote:On March 14 2016 00:58 redviper wrote:On March 14 2016 00:45 Cuce wrote:On March 13 2016 23:11 NonY wrote:
It also cannot gather any info except by looking at a monitor and listening to headphones. Any other method is clearly a hack. I wonder how ai would react of it will react in time to dt. since there wont be an alert. or how it will scan the screen agains nukes. we as players "feel" something wrong and react to these stuff. well thats just simplified way of saying "players been doing this sutff for so long that neuroplasticity forms a neural reflex akin to muscle memory". The same goes for sublte changes in macroing stuff to fit slight changes in timing due to socuting information, or deviations from meta throught out the game. these stuff are not really conscious desicions, they are more muscle memory. Thats the crux of the issue I thing. yeah sure, it can pull workred rushes everygame to win. But wining is not the result such a development team wants. they want to develop AI, not get gimmicky wins. I wonder how much variables an AI would have to go through to figure out what some to a pro player intuativly. Not sure if we can pull that off in realtime right now. thats another thing, realtime vs turnbased. You are truly underestimating an advanced AI. AIs build on DNN can certainly create make different decisions based on slightly different scouting information. After a million games the AI would have a much better understanding of the game's strategy than any human being could. This is like saying if an AI analyzes flipping a coin a million times, it has a better idea of what side the coin will land from this knowledge. The AI cannot reliably determine what course the human will take from the limited scouting input. Moreover, the AI can be duped by the human player with bad info. I disagree. One proper scout plus a very precise estimate of how many resources a player could have mined deducted from previous scouting can tell you nearly everything at any point in the game. And an AI may be capable to just force scouts by sheer amounts of micromanagement. Not to mention that such an AI might be aware that it will perform more efficiently in battles, even when outnumbered to some degree and therefore it might just choose to sacrifice more power for information, since it may realize that getting duped is the only way to lose against a mechanically inferior player. There's a lot possible.
But it doesn't tell the AI what the human player is spending with those resources.
If the human diligently denies scouting, the AI must ultimately guess.
|
I will take a bet against Boxer, fine for him that he was a pro, but i dont think he realise what a pc can do nowadays 
|
you never bet against boxer.
|
On March 13 2016 18:39 Grumbels wrote:Show nested quote +On March 13 2016 17:46 rabidch wrote:
it will take a while for google to come up with this though (a few years or more), mostly because of hashing out plans with blizzard (assuming theyll cooperate), designing how theyll train and do decisions, and then getting enough computing power and time to train. even with google's cloud i think it will take a massive amount of computing to train the thing, assuming they choose neutral networks like they did with alphago Why should they need permission from Blizzard though? Their other AIs would work with just visual input, it is not like they would need access to the game state.
having cooperation with blizzard would be able to speed up development time for google. the amount of computing power to run a go game versus computing power starcraft is far far less, and google would certainly be interested in cutting it down
|
On March 14 2016 05:11 Befree wrote:
So clearly a computer programmed to do an activity of brute force is entirely unimpressive when it beats a human. For example a car being faster than a human, or a calculator doing extremely large calculations that a human can't do.
The more abstract and less mechanical, the more impressive it would be for a computer obviously.
I guess I just wonder where StarCraft falls on this spectrum. The real-time aspect seems to me like it would give a huge advantage to AI which has very little to do with its "intelligence" and more to do with the brute power of computers.
I know BW AI tournaments put APM caps to eliminate at least part of this issue, but just the brute power of its awareness seems like it diminishes the meaningfulness of its competitive ability. It's a huge advantage that doesn't come through any sort of creativity or intelligence, just brute strength.
To me, that kind of collision is the entire point of exercises of this sort, which is why I am against imposing restrictions or handicapping the AI any more than absolutely necessary to ensure it doesn't straight up violate the rules. As I see it, the question here is whether human creativity, adaptability and capacity for abstract thinking is enough to overcome the raw computational power and mechanical precision of a computer. If the answer is no, then the answer is no and the human loses. Taking away the AI's main edge seems to render the whole project pointless. In the end, the ability of an AI to analyse and process massive amounts of data in a very short period of time is what gives it a fighting chance, and that must on some level include things like perfect unit control and mechanics.
It's a bit like if I insist a cheetah is only allowed to use two legs in a race, because otherwise it's going to be much faster than me. While true, it also defeats the basic premise of the competition, which is to discover if a human can compensate for a natural disadvantage through training and strategy. In this example the answer will clearly be no, and that's fine. It may be the same way in SC2(I suspect that it is), and that's fine too.
Personally, I am more interested to see that happens when an AI plays a fast-paced, decision-making centric game like Dota 2. I have no idea if an AI that could beat a top5 team could even be written, but if it could, it would be far more impressive than beating pros in starcraft.
|
On March 14 2016 03:37 nimdil wrote:
I don't think any progamer is suited to say anything about potential strength of well tuned AI in SC. They just have no idea what they talk about. In the same breath, a lot of people ITT that think experience trumps knowledge and facts about AI have no idea what they're talking about.
It seems many Wikipedia'd "AI" and "DeepMind" and are now experts in the field.
|
Seems like boxer is offering a challenge so that he gets the first call up =p
|
pretty sure AI has already won gsl, aka Innovation.
|
On March 14 2016 02:14 Liquid`Nazgul wrote:Show nested quote +On March 14 2016 01:21 Scarlett` wrote:
its 4 actions per ling in the AOE per tank shot (select; move; (wait) select; attackmove); say theres even only 5 tanks and with natural clumping of lings this by itself would require 6000+ APM (assuming theres ~15 lings in the aoe)
& dodging with the individually targeted ling also wouldnt work if the units attacking are somewhat clumped up (which is better vs certain armies and worse vs others so deciding how to pre-split is a huge issue in itself) unless the ling is at the edge of the army other than the front ~ one that is indeed very unlikely to be targeted even if the opponent is not controlling their tanks at all This is accounting for perfection. My view is that AI can be incredibly effective with a lot less than perfection. Pulling out individual units to counter the autotargetting of aoe units is something that can be done without thousands of apm. It won't be perfect, but it will be many times superior to human micro even at low apms. Just imagine your own apm but now with perfect selection onto individual units, perfect ability to predict where automated shots are going to go, and instant reactions and decisions. The difference between that and what you can do now is enormous. Show nested quote +
it also ignores the fact that the other player can micro the tank -> tell it which unit to target; so the AI would need to have the tank and every possible unit it can be targeting (based on the turret angle) on screen at the time it decides to micro while also knowing whether the tank will have vision of the target long enough to fire
because of the delay in the tank shot that makes this micro even theoretically possible, if the terran is controlled by an AI itself it could even re-target the tank after the zerg micros which is much simpler > if i take the previous example thats 2 actions compared to ~60 for a shot
all these micro bots built in sc2 are built upon information given directly form the game engine making them trivial to code rather than having to go by the visual and audio output the players are limited to
Right, but I think when you introduce an AI into the fold you have to assume its reactions to audio/visual are going to be pretty much instant. What I'm saying though is if audio/visual is instant, the behavior of the tanks can be 'read'. It is all behavior that is standardized and predictable. You can introduce human behavior to battle it of course, that is what will happen when an AI plays a professional, but I don't think it would be close. Just my perspective. Show nested quote +On March 13 2016 23:11 NonY wrote:
I don't get everyone who is assuming that the AI is allowed to cheat. The hardware rules are pretty clear for SC. Anything that is supposed to require one button press in the game must take one button press in the real world. Any "macro" that turns a combo of presses into one press is banned. And certainly any system that completely bypasses having to press anything is not going to be allowed. It also cannot gather any info except by looking at a monitor and listening to headphones. Any other method is clearly a hack. These are basic things for anyone who is played competitive SC to understand. Certainly by tournament rules an AI wouldn't be allowed to participate, and neither is it allowed to actually enter Go tournaments. For there to be a fruitful discussion about an AI playing there should be an understanding it wouldn't be playing through headphones and pressing buttons on a keyboard. If we are going to enforce those rules onto an AI we may as well not bother.
While it's true that it's impossible to subject an AI to the limitations of a human, there is a lot that can be done. Cap keys per second to lower than a human to account for extra efficiency, cap how quickly it can process information when changing screens (so it can't scan an area, look at it, and go back to where it was before in a couple of frames) and cap mouse accuracy so that it if it wants to click on something a certain distance away from its current cursor position it has to do it at a lower mouse speed, give it human-level reaction time with some predicitive capabilities. They could even give the AI 'eyes' so it has to think about when to observe the minimap, and doing so would restrict what kind of micro it could be doing for that brief instant.
The main issue is that I think doing stuff like this just increases the complexity of the model by a lot which would make things a lot harder and unappealing to AI researchers because they may feel they are workong on stuff that is too domain-specific to StarCraft. If they're going to let the AI do whatever crazy micro it wants then the goal would be for the AI to teach itself that superhuman micro is the optimal strategy, rather than teaching itself how to actually outsmart a human in StarCraft, in which case it'd be boring and pointless to make it fight Boxer or Flash. I do think with enough effort there could be a 'fair' AI, but it might be too much effort.
|
On March 14 2016 04:15 necrosexy wrote:Show nested quote +On March 14 2016 00:58 redviper wrote:On March 14 2016 00:45 Cuce wrote:On March 13 2016 23:11 NonY wrote:
It also cannot gather any info except by looking at a monitor and listening to headphones. Any other method is clearly a hack. I wonder how ai would react of it will react in time to dt. since there wont be an alert. or how it will scan the screen agains nukes. we as players "feel" something wrong and react to these stuff. well thats just simplified way of saying "players been doing this sutff for so long that neuroplasticity forms a neural reflex akin to muscle memory". The same goes for sublte changes in macroing stuff to fit slight changes in timing due to socuting information, or deviations from meta throught out the game. these stuff are not really conscious desicions, they are more muscle memory. Thats the crux of the issue I thing. yeah sure, it can pull workred rushes everygame to win. But wining is not the result such a development team wants. they want to develop AI, not get gimmicky wins. I wonder how much variables an AI would have to go through to figure out what some to a pro player intuativly. Not sure if we can pull that off in realtime right now. thats another thing, realtime vs turnbased. You are truly underestimating an advanced AI. AIs build on DNN can certainly create make different decisions based on slightly different scouting information. After a million games the AI would have a much better understanding of the game's strategy than any human being could. This is like saying if an AI analyzes flipping a coin a million times, it has a better idea of what side the coin will land from this knowledge. The AI cannot reliably determine what course the human will take from the limited scouting input. Moreover, the AI can be duped by the human player with bad info.
Its not how things would work. The AI would have a graph of possible paths that can be taken. Just as a person does. Having seen what the current state is, it will be able to predict the path the opponent is taking. Just the way a person can. There is nothing special about how pros play, its experience, insight and mechanics. DeepMind-like AIs could probably surpass them on almost all of these factors given enough computation. It could probably determine better than a human what the opponent is doing.
And don't assume that the AI can only react, it can have its own strategy.
|
On March 14 2016 05:53 pr1de wrote:I will take a bet against Boxer, fine for him that he was a pro, but i dont think he realise what a pc can do nowadays
Blasphemy!!! As soon as find my torch and pitchfork, I will come for you! 
But then again, I'm actually kind of hungry...I'll probably have lunch first...but as soon as I've finished my lunch..meh nevermind...
|
8748 Posts
On March 14 2016 02:14 Liquid`Nazgul wrote:Show nested quote +On March 13 2016 23:11 NonY wrote:
I don't get everyone who is assuming that the AI is allowed to cheat. The hardware rules are pretty clear for SC. Anything that is supposed to require one button press in the game must take one button press in the real world. Any "macro" that turns a combo of presses into one press is banned. And certainly any system that completely bypasses having to press anything is not going to be allowed. It also cannot gather any info except by looking at a monitor and listening to headphones. Any other method is clearly a hack. These are basic things for anyone who is played competitive SC to understand. Certainly by tournament rules an AI wouldn't be allowed to participate, and neither is it allowed to actually enter Go tournaments. For there to be a fruitful discussion about an AI playing there should be an understanding it wouldn't be playing through headphones and pressing buttons on a keyboard. If we are going to enforce those rules onto an AI we may as well not bother.
History does not set a precedent one way or the other because using a robot to execute the moves in Chess and Go would be purely symbolic. For SC it actually matters that the moves are performed physically. I'm not sure what significance we can take from it defeating the best human player if it is allowed to cheat. If their goal is to simply do AI research, then they can do it without the framework of games and without challenging human players. For us to put their accomplishment in terms we can understand, which I assume is a reason why they play these games and challenge their champions, then acknowledging the reality of the game is a necessity. They don't have to choose SC, but if they do and they want their results to continue to be meaningful, then they have to accept what SC is.
There are some very advanced robots. I think a very capable machine could be built and it would be all the more impressive and clear to people what they're accomplishing. Take a moment to imagine the SC-playing robot and how cool that would be. If they choose not to do so, then I guess personally all I can say is that it's no longer interesting to me, the same way I'm not interested in trying to judge how good of a player a hacker is, or who the best micro tournament player is, or whatever.
|
its not so impressive if an AI can outmuscle a human
its more impressive if it can out-think a human
so you need to have some restrictions in place to make it a thinking competition
|
On March 14 2016 11:06 redviper wrote:Show nested quote +On March 14 2016 04:15 necrosexy wrote:On March 14 2016 00:58 redviper wrote:On March 14 2016 00:45 Cuce wrote:On March 13 2016 23:11 NonY wrote:
It also cannot gather any info except by looking at a monitor and listening to headphones. Any other method is clearly a hack. I wonder how ai would react of it will react in time to dt. since there wont be an alert. or how it will scan the screen agains nukes. we as players "feel" something wrong and react to these stuff. well thats just simplified way of saying "players been doing this sutff for so long that neuroplasticity forms a neural reflex akin to muscle memory". The same goes for sublte changes in macroing stuff to fit slight changes in timing due to socuting information, or deviations from meta throught out the game. these stuff are not really conscious desicions, they are more muscle memory. Thats the crux of the issue I thing. yeah sure, it can pull workred rushes everygame to win. But wining is not the result such a development team wants. they want to develop AI, not get gimmicky wins. I wonder how much variables an AI would have to go through to figure out what some to a pro player intuativly. Not sure if we can pull that off in realtime right now. thats another thing, realtime vs turnbased. You are truly underestimating an advanced AI. AIs build on DNN can certainly create make different decisions based on slightly different scouting information. After a million games the AI would have a much better understanding of the game's strategy than any human being could. This is like saying if an AI analyzes flipping a coin a million times, it has a better idea of what side the coin will land from this knowledge. The AI cannot reliably determine what course the human will take from the limited scouting input. Moreover, the AI can be duped by the human player with bad info. Its not how things would work. The AI would have a graph of possible paths that can be taken. Just as a person does. Having seen what the current state is, it will be able to predict the path the opponent is taking. Just the way a person can. There is nothing special about how pros play, its experience, insight and mechanics. DeepMind-like AIs could probably surpass them on almost all of these factors given enough computation. It could probably determine better than a human what the opponent is doing. And don't assume that the AI can only react, it can have its own strategy.
Making a prediction and making the right prediction is the difference here. Again, the AI must guess.
|
On March 13 2016 14:06 BronzeKnee wrote:Show nested quote +On March 13 2016 14:03 writer22816 wrote:On March 13 2016 13:57 BronzeKnee wrote:On March 13 2016 03:19 Brutaxilos wrote:
As a programmer, I'm actually quite skeptical that Boxer would be able to beat an intelligent AI given that a team of scientists are given enough time to develop one. It's inevitable. If the constraints are the same, as in the computer must use a keyboard and mouse and can only view one part of the screen at a time I don't think an AI could ever win against a top player, at least in my lifetime. And if we were able to control the game using the human brain, it would be another no contest, we'd have near perfect micro and macro too. Only if the computer is unrestrained by a keyboard and mouse and the human is restrained by those factors will the human lose. Humans are far too innovative. If you simply deny scouting and the AI will either guess what you are doing or go for some standard safe play, and either of those things could be exploit. What the hell does it even mean for a computer to use a keyboard and a mouse? A computer doesn't have two hands and 10 fingers. The computer with the AI should be playing the game using another computer using a keyboard, mouse and monitor because that is how Starcraft is played. I'm didn't create Starcraft so don't blame that that is how the game is played. Having those limitations is what makes Starcraft difficult. If the AI can at once be blink microing Stalkers while warping in units at pylon off screen (off field of vision) then that is cheating.The AI in Chess or Go can do nothing a human cannot, the AI is literally outthinking the players. So even if the APM is limited, the field of vision must be limited also. The AI has no chance given equal constraints. If we are talking about no keyboard, mouse or monitor for the computer, then it should be the same for humans. I can imagine perfect forcefields, if the game responded to my mind, I'd never miss a forcefield. And my macro would be on point too, just subtle sounds would be all I would need to know to send a worker to a mineral line.
I know this is a few pages further back in the thread, but I really would like to address the artificial physical limitations people are trying to impose.
I'm pretty certain that no matter HOW MUCH WE ATTEMPT TO RESTRICT the AI in terms of physical playing limitations, it's still going to VASTLY outperform a human in terms of raw mechanics. SC is just much too mechanics-heavy of a game to properly design and implement an AI that "wins by out-thinking instead of out-muscling."
As a specific example, sure you can limit the AI to only perform actions within view, so that it cannot micro perfectly while also macro'ing "off screen". However this immediately becomes a moot point, because good multitasking is highly prized among SC players, and since true multitasking is impossible, it's no more than screen flipping using location and army hotkeys and trying to maximize the benefits of one's limited attention and actions. Considering inherent human limitations such as reaction time and muscle activation time, human multitasking is FAR from perfect. So the AI can be made to "multitask" as well, except we'll give it active snapshots and predicative algorithms as to what is and what would be happening on screen, so every millisecond the AI is flipping back and forth between the 2 important micro/macro locations, and performing actions as perfectly as your silly "limitations" allow.
Of course in the end people were getting ridiculous with ideas of extreme limitations, going as far as saying computers should be limited to playing with a keyboard and a mouse, thus almost completely removing any mechanical advantages the AI would have by nature, because "that's how SC is meant to be played." If such bizarre limitations were to be put in place so the "AI plays just like a human", then I guess AlphaGO is a cheater by definition, and its wins are meaningless as well right? Considering the human brain can only access limited parts of its limited memory (by comparison) at any given time, AlphaGO should have been made to only access 2KB of its 56MB (just random numbers here for illustration) database while playing right?
What makes these matches interesting is seeing how the imbalances (memory size vs. overall awareness, etc.) between AI and humans play out. If you attempt to artificially handicap the AI to exactly resemble a human, then what's the point of competing in the first place?
TL;DR: SC is just too mechanics heavy to allow for meaningful AI development. Coupled with the fact that it's a game of imperfect information with a significant RNG aspect to it, it's just not a good candidate game for AI exploration.
|
On March 14 2016 13:06 EngrishTeacher wrote:Show nested quote +On March 13 2016 14:06 BronzeKnee wrote:On March 13 2016 14:03 writer22816 wrote:On March 13 2016 13:57 BronzeKnee wrote:On March 13 2016 03:19 Brutaxilos wrote:
As a programmer, I'm actually quite skeptical that Boxer would be able to beat an intelligent AI given that a team of scientists are given enough time to develop one. It's inevitable. If the constraints are the same, as in the computer must use a keyboard and mouse and can only view one part of the screen at a time I don't think an AI could ever win against a top player, at least in my lifetime. And if we were able to control the game using the human brain, it would be another no contest, we'd have near perfect micro and macro too. Only if the computer is unrestrained by a keyboard and mouse and the human is restrained by those factors will the human lose. Humans are far too innovative. If you simply deny scouting and the AI will either guess what you are doing or go for some standard safe play, and either of those things could be exploit. What the hell does it even mean for a computer to use a keyboard and a mouse? A computer doesn't have two hands and 10 fingers. The computer with the AI should be playing the game using another computer using a keyboard, mouse and monitor because that is how Starcraft is played. I'm didn't create Starcraft so don't blame that that is how the game is played. Having those limitations is what makes Starcraft difficult. If the AI can at once be blink microing Stalkers while warping in units at pylon off screen (off field of vision) then that is cheating.The AI in Chess or Go can do nothing a human cannot, the AI is literally outthinking the players. So even if the APM is limited, the field of vision must be limited also. The AI has no chance given equal constraints. If we are talking about no keyboard, mouse or monitor for the computer, then it should be the same for humans. I can imagine perfect forcefields, if the game responded to my mind, I'd never miss a forcefield. And my macro would be on point too, just subtle sounds would be all I would need to know to send a worker to a mineral line. I know this is a few pages further back in the thread, but I really would like to address the artificial physical limitations people are trying to impose. I'm pretty certain that no matter HOW MUCH WE ATTEMPT TO RESTRICT the AI in terms of physical playing limitations, it's still going to VASTLY outperform a human in terms of raw mechanics. SC is just much too mechanics-heavy of a game to properly design and implement an AI that "wins by out-thinking instead of out-muscling." As a specific example, sure you can limit the AI to only perform actions within view, so that it cannot micro perfectly while also macro'ing "off screen". However this immediately becomes a moot point, because good multitasking is highly prized among SC players, and since true multitasking is impossible, it's no more than screen flipping using location and army hotkeys and trying to maximize the benefits of one's limited attention and actions. Considering inherent human limitations such as reaction time and muscle activation time, human multitasking is FAR from perfect. So the AI can be made to "multitask" as well, except we'll give it active snapshots and predicative algorithms as to what is and what would be happening on screen, so every millisecond the AI is flipping back and forth between the 2 important micro/macro locations, and performing actions as perfectly as your silly "limitations" allow. Of course in the end people were getting ridiculous with ideas of extreme limitations, going as far as saying computers should be limited to playing with a keyboard and a mouse, thus almost completely removing any mechanical advantages the AI would have by nature, because "that's how SC is meant to be played." If such bizarre limitations were to be put in place so the "AI plays just like a human", then I guess AlphaGO is a cheater by definition, and its wins are meaningless as well right? Considering the human brain can only access limited parts of its limited memory (by comparison) at any given time, AlphaGO should have been made to only access 2KB of its 56MB (just random numbers here for illustration) database while playing right? What makes these matches interesting is seeing how the imbalances (memory size vs. overall awareness, etc.) between AI and humans play out. If you attempt to artificially handicap the AI to exactly resemble a human, then what's the point of competing in the first place? TL;DR: SC is just too mechanics heavy to allow for meaningful AI development. Coupled with the fact that it's a game of imperfect information with a significant RNG aspect to it, it's just not a good candidate game for AI exploration.
I couldn't agree more.
I for my part I would find it beautiful if an AI could show us that the game is played completely wrong and possibly broken beyond repair. It shows the possibilities for the game, the gaming industry and e-sports and should be embraced rather than feared. It has no impact whatsoever on the human achievments in this game so far.
|
On March 14 2016 13:15 Big J wrote:Show nested quote +On March 14 2016 13:06 EngrishTeacher wrote:On March 13 2016 14:06 BronzeKnee wrote:On March 13 2016 14:03 writer22816 wrote:On March 13 2016 13:57 BronzeKnee wrote:On March 13 2016 03:19 Brutaxilos wrote:
As a programmer, I'm actually quite skeptical that Boxer would be able to beat an intelligent AI given that a team of scientists are given enough time to develop one. It's inevitable. If the constraints are the same, as in the computer must use a keyboard and mouse and can only view one part of the screen at a time I don't think an AI could ever win against a top player, at least in my lifetime. And if we were able to control the game using the human brain, it would be another no contest, we'd have near perfect micro and macro too. Only if the computer is unrestrained by a keyboard and mouse and the human is restrained by those factors will the human lose. Humans are far too innovative. If you simply deny scouting and the AI will either guess what you are doing or go for some standard safe play, and either of those things could be exploit. What the hell does it even mean for a computer to use a keyboard and a mouse? A computer doesn't have two hands and 10 fingers. The computer with the AI should be playing the game using another computer using a keyboard, mouse and monitor because that is how Starcraft is played. I'm didn't create Starcraft so don't blame that that is how the game is played. Having those limitations is what makes Starcraft difficult. If the AI can at once be blink microing Stalkers while warping in units at pylon off screen (off field of vision) then that is cheating.The AI in Chess or Go can do nothing a human cannot, the AI is literally outthinking the players. So even if the APM is limited, the field of vision must be limited also. The AI has no chance given equal constraints. If we are talking about no keyboard, mouse or monitor for the computer, then it should be the same for humans. I can imagine perfect forcefields, if the game responded to my mind, I'd never miss a forcefield. And my macro would be on point too, just subtle sounds would be all I would need to know to send a worker to a mineral line. I know this is a few pages further back in the thread, but I really would like to address the artificial physical limitations people are trying to impose. I'm pretty certain that no matter HOW MUCH WE ATTEMPT TO RESTRICT the AI in terms of physical playing limitations, it's still going to VASTLY outperform a human in terms of raw mechanics. SC is just much too mechanics-heavy of a game to properly design and implement an AI that "wins by out-thinking instead of out-muscling." As a specific example, sure you can limit the AI to only perform actions within view, so that it cannot micro perfectly while also macro'ing "off screen". However this immediately becomes a moot point, because good multitasking is highly prized among SC players, and since true multitasking is impossible, it's no more than screen flipping using location and army hotkeys and trying to maximize the benefits of one's limited attention and actions. Considering inherent human limitations such as reaction time and muscle activation time, human multitasking is FAR from perfect. So the AI can be made to "multitask" as well, except we'll give it active snapshots and predicative algorithms as to what is and what would be happening on screen, so every millisecond the AI is flipping back and forth between the 2 important micro/macro locations, and performing actions as perfectly as your silly "limitations" allow. Of course in the end people were getting ridiculous with ideas of extreme limitations, going as far as saying computers should be limited to playing with a keyboard and a mouse, thus almost completely removing any mechanical advantages the AI would have by nature, because "that's how SC is meant to be played." If such bizarre limitations were to be put in place so the "AI plays just like a human", then I guess AlphaGO is a cheater by definition, and its wins are meaningless as well right? Considering the human brain can only access limited parts of its limited memory (by comparison) at any given time, AlphaGO should have been made to only access 2KB of its 56MB (just random numbers here for illustration) database while playing right? What makes these matches interesting is seeing how the imbalances (memory size vs. overall awareness, etc.) between AI and humans play out. If you attempt to artificially handicap the AI to exactly resemble a human, then what's the point of competing in the first place? TL;DR: SC is just too mechanics heavy to allow for meaningful AI development. Coupled with the fact that it's a game of imperfect information with a significant RNG aspect to it, it's just not a good candidate game for AI exploration. I couldn't agree more. I for my part I would find it beautiful if an AI could show us that the game is played completely wrong and possibly broken beyond repair. It shows the possibilities for the game, the gaming industry and e-sports and should be embraced rather than feared. It has no impact whatsoever on the human achievments in this game so far.
The point of the calls for limiting the "mechanics" of the AI is to show that is actually won because of tactics/strategy.
That's pretty much the point of these AI programms in the first place.
If you simply have it with no limitations it could win purely on mechanics alone, it would be almost trivial.
You could also argue that starcraft (or any game which is played in real time) might be simply a bad choice for this kind of thing.
|
On March 14 2016 13:53 The_Red_Viper wrote:Show nested quote +On March 14 2016 13:15 Big J wrote:On March 14 2016 13:06 EngrishTeacher wrote:On March 13 2016 14:06 BronzeKnee wrote:On March 13 2016 14:03 writer22816 wrote:On March 13 2016 13:57 BronzeKnee wrote:On March 13 2016 03:19 Brutaxilos wrote:
As a programmer, I'm actually quite skeptical that Boxer would be able to beat an intelligent AI given that a team of scientists are given enough time to develop one. It's inevitable. If the constraints are the same, as in the computer must use a keyboard and mouse and can only view one part of the screen at a time I don't think an AI could ever win against a top player, at least in my lifetime. And if we were able to control the game using the human brain, it would be another no contest, we'd have near perfect micro and macro too. Only if the computer is unrestrained by a keyboard and mouse and the human is restrained by those factors will the human lose. Humans are far too innovative. If you simply deny scouting and the AI will either guess what you are doing or go for some standard safe play, and either of those things could be exploit. What the hell does it even mean for a computer to use a keyboard and a mouse? A computer doesn't have two hands and 10 fingers. The computer with the AI should be playing the game using another computer using a keyboard, mouse and monitor because that is how Starcraft is played. I'm didn't create Starcraft so don't blame that that is how the game is played. Having those limitations is what makes Starcraft difficult. If the AI can at once be blink microing Stalkers while warping in units at pylon off screen (off field of vision) then that is cheating.The AI in Chess or Go can do nothing a human cannot, the AI is literally outthinking the players. So even if the APM is limited, the field of vision must be limited also. The AI has no chance given equal constraints. If we are talking about no keyboard, mouse or monitor for the computer, then it should be the same for humans. I can imagine perfect forcefields, if the game responded to my mind, I'd never miss a forcefield. And my macro would be on point too, just subtle sounds would be all I would need to know to send a worker to a mineral line. I know this is a few pages further back in the thread, but I really would like to address the artificial physical limitations people are trying to impose. I'm pretty certain that no matter HOW MUCH WE ATTEMPT TO RESTRICT the AI in terms of physical playing limitations, it's still going to VASTLY outperform a human in terms of raw mechanics. SC is just much too mechanics-heavy of a game to properly design and implement an AI that "wins by out-thinking instead of out-muscling." As a specific example, sure you can limit the AI to only perform actions within view, so that it cannot micro perfectly while also macro'ing "off screen". However this immediately becomes a moot point, because good multitasking is highly prized among SC players, and since true multitasking is impossible, it's no more than screen flipping using location and army hotkeys and trying to maximize the benefits of one's limited attention and actions. Considering inherent human limitations such as reaction time and muscle activation time, human multitasking is FAR from perfect. So the AI can be made to "multitask" as well, except we'll give it active snapshots and predicative algorithms as to what is and what would be happening on screen, so every millisecond the AI is flipping back and forth between the 2 important micro/macro locations, and performing actions as perfectly as your silly "limitations" allow. Of course in the end people were getting ridiculous with ideas of extreme limitations, going as far as saying computers should be limited to playing with a keyboard and a mouse, thus almost completely removing any mechanical advantages the AI would have by nature, because "that's how SC is meant to be played." If such bizarre limitations were to be put in place so the "AI plays just like a human", then I guess AlphaGO is a cheater by definition, and its wins are meaningless as well right? Considering the human brain can only access limited parts of its limited memory (by comparison) at any given time, AlphaGO should have been made to only access 2KB of its 56MB (just random numbers here for illustration) database while playing right? What makes these matches interesting is seeing how the imbalances (memory size vs. overall awareness, etc.) between AI and humans play out. If you attempt to artificially handicap the AI to exactly resemble a human, then what's the point of competing in the first place? TL;DR: SC is just too mechanics heavy to allow for meaningful AI development. Coupled with the fact that it's a game of imperfect information with a significant RNG aspect to it, it's just not a good candidate game for AI exploration. I couldn't agree more. I for my part I would find it beautiful if an AI could show us that the game is played completely wrong and possibly broken beyond repair. It shows the possibilities for the game, the gaming industry and e-sports and should be embraced rather than feared. It has no impact whatsoever on the human achievments in this game so far. The point of the calls for limiting the "mechanics" of the AI is to show that is actually won because of tactics/strategy. That's pretty much the point of these AI programms in the first place. If you simply have it with no limitations it could win purely on mechanics alone, it would be almost trivial. You could also argue that starcraft (or any game which is played in real time) might be simply a bad choice for this kind of thing.
That's... almost exactly what I said, I even had a TL; DR at the bottom but I guess it was still too long to read.
Still TL; DR: it would be nearly impossible to implement the "right" amount of limitations, and thus SC, or any other RTS game, are probably terrible candidates for AI development.
|
It can be done if you had money to throw away.
5 programmers * $80k * 1 year = $400,000
|
On March 14 2016 14:00 EngrishTeacher wrote:Show nested quote +On March 14 2016 13:53 The_Red_Viper wrote:On March 14 2016 13:15 Big J wrote:On March 14 2016 13:06 EngrishTeacher wrote:On March 13 2016 14:06 BronzeKnee wrote:On March 13 2016 14:03 writer22816 wrote:On March 13 2016 13:57 BronzeKnee wrote:On March 13 2016 03:19 Brutaxilos wrote:
As a programmer, I'm actually quite skeptical that Boxer would be able to beat an intelligent AI given that a team of scientists are given enough time to develop one. It's inevitable. If the constraints are the same, as in the computer must use a keyboard and mouse and can only view one part of the screen at a time I don't think an AI could ever win against a top player, at least in my lifetime. And if we were able to control the game using the human brain, it would be another no contest, we'd have near perfect micro and macro too. Only if the computer is unrestrained by a keyboard and mouse and the human is restrained by those factors will the human lose. Humans are far too innovative. If you simply deny scouting and the AI will either guess what you are doing or go for some standard safe play, and either of those things could be exploit. What the hell does it even mean for a computer to use a keyboard and a mouse? A computer doesn't have two hands and 10 fingers. The computer with the AI should be playing the game using another computer using a keyboard, mouse and monitor because that is how Starcraft is played. I'm didn't create Starcraft so don't blame that that is how the game is played. Having those limitations is what makes Starcraft difficult. If the AI can at once be blink microing Stalkers while warping in units at pylon off screen (off field of vision) then that is cheating.The AI in Chess or Go can do nothing a human cannot, the AI is literally outthinking the players. So even if the APM is limited, the field of vision must be limited also. The AI has no chance given equal constraints. If we are talking about no keyboard, mouse or monitor for the computer, then it should be the same for humans. I can imagine perfect forcefields, if the game responded to my mind, I'd never miss a forcefield. And my macro would be on point too, just subtle sounds would be all I would need to know to send a worker to a mineral line. I know this is a few pages further back in the thread, but I really would like to address the artificial physical limitations people are trying to impose. I'm pretty certain that no matter HOW MUCH WE ATTEMPT TO RESTRICT the AI in terms of physical playing limitations, it's still going to VASTLY outperform a human in terms of raw mechanics. SC is just much too mechanics-heavy of a game to properly design and implement an AI that "wins by out-thinking instead of out-muscling." As a specific example, sure you can limit the AI to only perform actions within view, so that it cannot micro perfectly while also macro'ing "off screen". However this immediately becomes a moot point, because good multitasking is highly prized among SC players, and since true multitasking is impossible, it's no more than screen flipping using location and army hotkeys and trying to maximize the benefits of one's limited attention and actions. Considering inherent human limitations such as reaction time and muscle activation time, human multitasking is FAR from perfect. So the AI can be made to "multitask" as well, except we'll give it active snapshots and predicative algorithms as to what is and what would be happening on screen, so every millisecond the AI is flipping back and forth between the 2 important micro/macro locations, and performing actions as perfectly as your silly "limitations" allow. Of course in the end people were getting ridiculous with ideas of extreme limitations, going as far as saying computers should be limited to playing with a keyboard and a mouse, thus almost completely removing any mechanical advantages the AI would have by nature, because "that's how SC is meant to be played." If such bizarre limitations were to be put in place so the "AI plays just like a human", then I guess AlphaGO is a cheater by definition, and its wins are meaningless as well right? Considering the human brain can only access limited parts of its limited memory (by comparison) at any given time, AlphaGO should have been made to only access 2KB of its 56MB (just random numbers here for illustration) database while playing right? What makes these matches interesting is seeing how the imbalances (memory size vs. overall awareness, etc.) between AI and humans play out. If you attempt to artificially handicap the AI to exactly resemble a human, then what's the point of competing in the first place? TL;DR: SC is just too mechanics heavy to allow for meaningful AI development. Coupled with the fact that it's a game of imperfect information with a significant RNG aspect to it, it's just not a good candidate game for AI exploration. I couldn't agree more. I for my part I would find it beautiful if an AI could show us that the game is played completely wrong and possibly broken beyond repair. It shows the possibilities for the game, the gaming industry and e-sports and should be embraced rather than feared. It has no impact whatsoever on the human achievments in this game so far. The point of the calls for limiting the "mechanics" of the AI is to show that is actually won because of tactics/strategy. That's pretty much the point of these AI programms in the first place. If you simply have it with no limitations it could win purely on mechanics alone, it would be almost trivial. You could also argue that starcraft (or any game which is played in real time) might be simply a bad choice for this kind of thing. That's... almost exactly what I said, I even had a TL; DR at the bottom but I guess it was still too long to read. Still TL; DR: it would be nearly impossible to implement the "right" amount of limitations, and thus SC, or any other RTS game, are probably terrible candidates for AI development.
I read Big J's reply the wrong way i fear. I got confused why he would agree with you and at the same time say the AI would show us how the game is meant to be played (strategically) when it probably would just win by pure mechanics (against humans)
AI vs AI would be more interesting i guess.
So yeah my bad for already making posts early in the morning here -.-
TLDR: I partially agree with you, real time games (and sc2) are probably a bad place for AI development, or to be more precise a bad area to show that the AI beats humans because of superior 'understanding' of the game. (which is obviously only one step in the direction of a real 'intelligent AI')
|
|
|
|
|
|
EPT
