On the Ball: Man Versus Machine

 Pages 1 2 NEXT
 

On the Ball: Man Versus Machine

When will StarCraft get its Deep Blue? Probably never, if Blizzard has its say.

Read Full Article

@Xersues

A software bot can beat people at Starcraft easily, so therefore, it's not a sport? Bullshit. Even in a human body, a robot mind could easily beat and human mind at any traditional sport. Computers do shit better than humans do. Deal with it.

In addition people don't seem to realize just how difficult it is to program good AI that doesn't cheat. The human player has the advantage of being able to adapt and more importantly do new things. No AI is capable of doing something its not coded to do.

Can't be any less of a sport than Chess-Boxing.

Robots can be made to do anything better than a human.

Simply because an artificial superior exists, it does not negate the competition.

Sports are all about human vs human, and the relative skill of the two competitors. We don't enjoy a little league game (in it or as parents) less because we know a Major League team could wipe the floor with us. We don't enjoy a running track race less because cars can go faster.

I would just like to say that computers, even A.I., only do what you tell them to do, they just do it really fast. The advantage comes from speed, so at the moment, any Starcraft master A.I. would suck up so much memory that it wouldn't be able to play effectively. good micro only goes so far, and once you can figure out the basic pattern, it's easy to get around.

Also, an A.I. is never going to nuke rush someone.

Although the idea is extremely appealing the execution of such a program would be incredibly difficult - true a computer can execute commands much faster than a human, however giving the program the ability to pick which commands are the best to take at a given moment would be a lot of work.

The reason Deep Blue was competitive is because chess has a finite number of possible games and the computer was allowed a reasonable amount of computational time between moves.

Now in Starcraft the possibilities are infinite and the computer would have to decide what actions to take on a second on second basis - each of which could potentially require a hefty number of calculations!

I think it would be possible to design such a program - but it would require a lot of work and a reasonably powerful computer to execute!

Yes, you can make an AI that can beat any human player. Yes, you can make an AI that can nuke rush people. Does that ruin starcraft for non-AI competitors? Not really. Trying to say that an AI would ruin the 'sport' of starcraft is like saying that the invention of the car would forever obsolete the 100m dash. People still run foot races despite the fact that there are cars that can move that distance faster.

Now whether I consider starcraft to be a sport or not in the first place? That's another question entirely. One I haven't entirely answered for myself.

As randommaster points out, at this point it just isn't feasible to make an A.I that is on par with the "pro" players. Let me take a related game that I played in the "pro" divisions for a while, namely Company of Heroes. Company of Heroes involved a great deal of mind battles, just as I imagine StarCraft II does. It wasn't that you couldn't counter a Quick Ostwind or a Wehrmacht T3-Puma Rush if you only knew that was what was taking place. But could you be sure? Perhaps I wasn't going for the T3 Rush but the Ostwind, and could you afford to wait another three minutes to get an M10 instead of that AT-gun if I was about to send in my first Puma?

Modern game AIs can't handle the level of complexity and game reading that goes on in "Pro" level strategy gaming. There is so much that a human player does that isn't related to micro but is very important to how the game plays out. This is what AIs can't do yet and it is why a human can still beat them. A good StarCraft match (or Company of Heroes match) is just as much about what you build and what you do with your units as what you could build or could do.

Really good replies in this thread. I don't think we should laugh at e sports, competition is competition no?

There are several mods of StarCraft to simulate pro-Gamer difficulty of play. The AI is specifically tailored to a select map. Utilizing the best strategies, micro managing battles without ever slowing down the macro machine. Suck programs are use by pro-gamers to train and hone their skills.

Replays have been long datamining for every tiny bit of information. Build order, unit composition, specific clicks and even field of view. So when Stork was ruling the championship scene, the programmers set to work and created an AI mod that simulated his strategies and timing almost to perfection. Now this became a weapon in the hands of his competitors to try and figure out a counter on dissect it to millisecond to find a weak chain.

Even though there hasn't been anything even close to the effort and money spent to develop Deep Two, it still shows that we are not that far from seeing StarCraft super computer. The key difference is time. In the past most of the focus was turned into creating the super machine, the solver to all problems. Experiments like Deep Two allowed engineers to push the boundaries of computing power. Nowadays computers are everywhere and the focus is more and making them easier and more comfortable to use.

Nimbus:
@Xersues

A software bot can beat people at Starcraft easily, so therefore, it's not a sport? Bullshit. Even in a human body, a robot mind could easily beat and human mind at any traditional sport. Computers do shit better than humans do. Deal with it.

that's not exactly true.

Human brains are still more complex then any computer system we've ever created (even the new Nvidia chip with its 3 billion transistors pales in comparison).

What Deam's saying is that computers have the edge MECHANICALLY in a game like starcraft.

When your brain thinks of something, it sends an impulse to your fingers, and then your fingers act, and then something happens on the computer, its sent to a screen, the screen broadcasts it to your eyes, your brain reinterprets the data it's fed from the screen, your brain responds and sends another impulse to your fingers.

The computer AI merely has to be aware that something is happening and it can react instantaneously. It's sort of like if your bloodsugar spikes, your body will send out more insulin to bring it under control. Think of bloodsugar as a zerg rush, and insulin as a squad of marines and that's basically how a computer AI of the Deep Blue variety would act. With no real thought, just reactions.

In that case, your interface with a computer IS slower then the computer's interface with itself.

But your brain still beats the computer almost every single time.

Cars can move faster than humans, so clearly running isn't a sport either. Machines can also throw farther, so take out shot-put and javelins. A boat can sure as hell move faster through the water, so there goes swimming, and long jump doesn't stand a chance when machines can actually fly. Clearly we should all just stand around doing nothing since we're so useless at everything.

randommaster:
I would just like to say that computers, even A.I., only do what you tell them to do, they just do it really fast. The advantage comes from speed, so at the moment, any Starcraft master A.I. would suck up so much memory that it wouldn't be able to play effectively. good micro only goes so far, and once you can figure out the basic pattern, it's easy to get around.

Also, an A.I. is never going to nuke rush someone.

There's no reason an AI with the proper training couldn't know to do that in certain situations.

@Jordan Dean: I wonder what you think about the people who create AI's compeditivley? Could that be considered a sport? (Think chess bots/game bots *eg DEFCON*) Could this be the real e-sports? (I think not until these ai can conistantly defeat human competition.)

Anything that a computer has a chance of doing better I can't ever claim to be a "sport" or even "pro" at.

By that argument, nothing in the world is a sport, because some day (maybe a LOOONG time into the future, but still some day) we will probably have nano-engineered robots that can play every physical "sport" better than humans too.

Does that mean Soccer, Rugby or whatever sport ceases to be a sport? No. With the risk of a ban for saying this, his quoted opinion above is, quite frankly, retarded, because if you follow his logic, we might as well terminate the term "sport" because we can create something that would make it cease to exist.

Also, have chess stopped being played professionally because computers beat humans now (which they do. The worlds best chess program, Rybka, is almost virtually unbeatable even by the best grandmasters, and have bested grandmasters even while giving them handicaps)? No. It still lives. Instead, computers brought a lot to chess, without destroying human chess. Centaur Chess (Human and computer working together, using the humans strategical thinking with the computers tactical precision and calculating power), or simple computer-vs-computer tournaments are quite popular these days.

Xerseus is simply unable to discern between "humans" and "AI's" and think that the competition has to be between those two, when they can be kept seperate. Even if someone created an AI that could beat every human on the earth, human StarCraft would still be played professionally, because humans are humans, and as long as we are humans, we can use competition to test each others skills, regardless of whether or not a computer can do it better.

The Great JT:
Can't be any less of a sport than Chess-Boxing.

Thats a sport? ;o

And true, if Korea can make gaming looks something from professional athletes, nothing says gaming cant have its own league

Gethsemani:
Modern game AIs can't handle the level of complexity and game reading that goes on in "Pro" level strategy gaming. There is so much that a human player does that isn't related to micro but is very important to how the game plays out. This is what AIs can't do yet and it is why a human can still beat them. A good StarCraft match (or Company of Heroes match) is just as much about what you build and what you do with your units as what you could build or could do.

The difficulty in getting an AI to execute good long term strategies, and counter those of its opponents, is also chess computers' greatest weakness. I don't think it matters anymore these days, because chess computers are amazingly strong now, but at least back in the Deep Blue era the best way to beat a computer was to play for long term advantage, beyond the maximum move search depth of the AI.

Altorin:

randommaster:
I would just like to say that computers, even A.I., only do what you tell them to do, they just do it really fast. The advantage comes from speed, so at the moment, any Starcraft master A.I. would suck up so much memory that it wouldn't be able to play effectively. good micro only goes so far, and once you can figure out the basic pattern, it's easy to get around.

Also, an A.I. is never going to nuke rush someone.

There's no reason an AI with the proper training couldn't know to do that in certain situations.

You can't train an AI, though, you have to program it. And while you can program actions to take in specific situations (early build order, effective countermeasures, etc.), there are too many situations that can come up, so you can easily miss something, and even if you were able to program every single possible scenario that could ever occur, it would take up so computing power that it would slow down to a crawl. If you just go with general cases then it's easier to exploit the patterns.

It's the same reason Chess hasn't been "solved," unlike Checkers. The decision trees are too complex and can't be implemented with todays technology.

Considering the only AI available at the moment is "Very Easy" it's kinda hard to tell if the computer knows what they're doing. I did a late rush with 12 zerglings and killed 3 other Terrans allied against me solely because they had 3-4 units by the time I hit their base.

Blizzard has stated it's looking toward Korea for ideas in how to improve their AI, but I don't see them developing an uber-AI that can destroy even the best Koreans.

randommaster:

Altorin:

randommaster:
I would just like to say that computers, even A.I., only do what you tell them to do, they just do it really fast. The advantage comes from speed, so at the moment, any Starcraft master A.I. would suck up so much memory that it wouldn't be able to play effectively. good micro only goes so far, and once you can figure out the basic pattern, it's easy to get around.

Also, an A.I. is never going to nuke rush someone.

There's no reason an AI with the proper training couldn't know to do that in certain situations.

You can't train an AI, though, you have to program it. And while you can program actions to take in specific situations (early build order, effective countermeasures, etc.), there are too many situations that can come up, so you can easily miss something, and even if you were able to program every single possible scenario that could ever occur, it would take up so computing power that it would slow down to a crawl. If you just go with general cases then it's easier to exploit the patterns.

It's the same reason Chess hasn't been "solved," unlike Checkers. The decision trees are too complex and can't be implemented with todays technology.

Yes, but the "nuke rush" strategy would be incredibly easy to program into an AI. As far as I could tell from your video, the nuke rush didn't really require that you react too much to what the enemy does. You build your nuke and ghost as fast as you can along with your science vessel and a few other things, then storm your way into their base and watch the enemy die. That's the kind of thing that a computer would be amazing at. It's the long term defensive strategy that the computer would have trouble with.

Holy crap, it's pretty awesome that a knee-jerk response to an article got posted. Its like 15 minutes of anonymous fame. I love how most of these responses are arguing the essence of sport.

Athinira:

Anything that a computer has a chance of doing better I can't ever claim to be a "sport" or even "pro" at.

By that argument, nothing in the world is a sport, because some day (maybe a LOOONG time into the future, but still some day) we will probably have nano-engineered robots that can play every physical "sport" better than humans too.

Does that mean Soccer, Rugby or whatever sport ceases to be a sport? No. With the risk of a ban for saying this, his quoted opinion above is, quite frankly, retarded, because if you follow his logic, we might as well terminate the term "sport" because we can create something that would make it cease to exist.

[snip]

Xerseus is simply unable to discern between "humans" and "AI's" and think that the competition has to be between those two, when they can be kept seperate. Even if someone created an AI that could beat every human on the earth, human StarCraft would still be played professionally, because humans are humans, and as long as we are humans, we can use competition to test each others skills, regardless of whether or not a computer can do it better.

My logic never said nothing in life can ever be a sport, I just said I can't claim its a sport if a computer can consistently do it better.

You know why? Computers are tools. No one ever got an award for wrenching the fastest. No one gets an off on shaping metal anymore, that's what happens to society and technology advances. Flying planes used to be globally renown, now its simplified and turned into a hobby. It ceases to be amazing. If you think what I said was retarded, then by literal context of course it is. It has little substance, because you gave and so did most of the people on here sans the article that much thought that went into it.

Sports are what, events of enjoyment for spectators, and enjoyment for those that do them. When something no longer becomes an event of wonder its socially downgraded to recreation or hobby. No grand stands dedicated to its monument anymore.

What happens when a tool can consistently outperform its human counterparts? Humans adapt, into a realm the tool or AI cannot perform. That in itself is an awesome sport.

Things are not black and white ladies and gentlemen, and things are not stagnate. The "On a long enough time line everything turns to nothing, this logic sucks!" argument is just silly. Think about things just a little bit more, just like I should have when I mentioned that I can't find that(e-sports) amazing. :)

"Sport" and "pro" are what we make them. Have some one tell you they're a professional lamplighter (a real old job) and you'll laugh at them. That's what a light switch is for.

Anyway, you all know the only thing that really matters is what you want to do and have fun with. Everyone only gets one!

It would be very difficult to keep a "Master AI" up to date with a game's meta-game. If the AI hasn't been coded to deal with a new strategy, then it will be beaten. Therefore it would need to be updated constantly, which I s'pose over time would turn it into an unbeatable beast once all variant strategies have been discovered. Really, it just isn't feasible.

Just an important note to your post: I only read the article. I didn't read into other (potential) comments you made in the issue of the Escapist.

Now that you have explained yourself a bit better, i overall agree. But as for your example with the "lamplighter", that would only be relevant if lamplighter had some grounds for competition. One of the reasons starcraft is so popular despite it's age is that it's a great game for competition, so therefore, being labeled as a pro at "lamplightning" might not be anything special, but being labeled as a pro in StarCraft can actually carry benefits as long as people are willing to be spectators for the circus.

"Sport" (or E-sport) is in itself a terrible word. "Competition" is much better. You compete in a discipline versus other humans to test who is the better man. Computers, robots or AI's can't change that. It's like saying that sprinting competitions aren't fun because a Cheetah can sprint 60+ MPH. Sure, a Cheetah can run much faster than a human, but you aren't going to have Cheetah's running versus humans as a spectator sport anytime soon, while human sprinting competitions are still a popular discipline.

My point is that what non-human elements can achieve is irrelevant as long as the competition stays between humans. Sure, you can do non-human competitions (Kasparov vs. Deep Blue), if nothing else just for the laughs, but in the end, it shouldn't be compared to human vs. human, and it probably shouldn't be made into a spectator event either.

randommaster:

Altorin:

randommaster:
I would just like to say that computers, even A.I., only do what you tell them to do, they just do it really fast. The advantage comes from speed, so at the moment, any Starcraft master A.I. would suck up so much memory that it wouldn't be able to play effectively. good micro only goes so far, and once you can figure out the basic pattern, it's easy to get around.

Also, an A.I. is never going to nuke rush someone.

There's no reason an AI with the proper training couldn't know to do that in certain situations.

You can't train an AI, though, you have to program it. And while you can program actions to take in specific situations (early build order, effective countermeasures, etc.), there are too many situations that can come up, so you can easily miss something, and even if you were able to program every single possible scenario that could ever occur, it would take up so computing power that it would slow down to a crawl. If you just go with general cases then it's easier to exploit the patterns.

It's the same reason Chess hasn't been "solved," unlike Checkers. The decision trees are too complex and can't be implemented with todays technology.

Yes, you CAN train an AI. See Artificial neural network
The problem is that it will be much too slow to have the reaction needed to beat a pro SC player.

In the case of Deep Blue, since IBM couldn't out think a human, they went for the brute force way with a huge parallel computer that had access to a database of pro chess player matches.

I think it may be possible to make a pro SC AI today, but there's no way it can run on a PC running SCII at the same time. It may take 30 octo-core, but it is possible. It's just not practical enough to put in your living room.

Athinira:
Just an important note to your post: I only read the article. I didn't read into other (potential) comments you made in the issue of the Escapist.

Now that you have explained yourself a bit better, i overall agree. But as for your example with the "lamplighter", that would only be relevant if lamplighter had some grounds for competition. One of the reasons starcraft is so popular despite it's age is that it's a great game for competition, so therefore, being labeled as a pro at "lamplightning" might not be anything special, but being labeled as a pro in StarCraft can actually carry benefits as long as people are willing to be spectators for the circus.

"Sport" (or E-sport) is in itself a terrible word. "Competition" is much better. You compete in a discipline versus other humans to test who is the better man. Computers, robots or AI's can't change that. It's like saying that sprinting competitions aren't fun because a Cheetah can sprint 60+ MPH. Sure, a Cheetah can run much faster than a human, but you aren't going to have Cheetah's running versus humans as a spectator sport anytime soon, while human sprinting competitions are still a popular discipline.

My point is that what non-human elements can achieve is irrelevant as long as the competition stays between humans. Sure, you can do non-human competitions (Kasparov vs. Deep Blue), if nothing else just for the laughs, but in the end, it shouldn't be compared to human vs. human, and it probably shouldn't be made into a spectator event either.

This *points up* though with one exception: Human vs. Machine matches make for great spectator events, at least the first time it's done. Once it's established that one can consistently beat the other with little to no difficulty it goes right back to Human vs. Human and maybe Machine vs. Machine again.

Starcraft is difficult to automate for the same reason poker is, there is incomplete information and an element of bluffing involved. Computer scientists have taken an interest in hold'em bots recently, and the research is significant because it is on the cutting edge of science and is trying to gain new knowledge in regards to how to process and make good decisions with incomplete information.

Mr. Mike:
It would be very difficult to keep a "Master AI" up to date with a game's meta-game. If the AI hasn't been coded to deal with a new strategy, then it will be beaten. Therefore it would need to be updated constantly, which I s'pose over time would turn it into an unbeatable beast once all variant strategies have been discovered. Really, it just isn't feasible.

Indeed, this is why research into Starcraft AI would be in the same field as research into poker AI.

dochmbi:
Starcraft is difficult to automate for the same reason poker is, there is incomplete information and an element of bluffing involved. Computer scientists have taken an interest in hold'em bots recently, and the research is significant because it is on the cutting edge of science and is trying to gain new knowledge in regards to how to process and make good decisions with incomplete information.

Unless the programmer cheated and programmed the AI to query the hidden location of enemy units and base decisions on that. >.>

Brett Staebell's article, on the other hand, offers a portrait of a more determined competitor, one who went on to become a sort of StarCraft Grandmaster: Lim Yo-Hwan, better known as SlayerS_'BoxeR'. BoxeR practiced religiously for years until he reached the top of the professional StarCraft scene, becoming a high-profile and well-paid gaming celebrity in the process. It's an extraordinary achievement that hasn't been replicated since - one could argue that BoxeR wasn't just the pinnacle of competitive StarCraft, but of e-sports in general.

Hasn't been replicated since? I suggest you look up the term "bonjwa"

However, taking the logic a step further wouldn't it be interesting to see a game which pitted humans against AI while giving both a chance to win.

Suppose that the speed of AI was balanced by the flexibility of human thinking.

Or even if a machine was pitted against a human in some sort of virtual race, then the reflexes of the machine could be compensated for by the superior terrain navigation by the buman.

It would require some balancing to get right, but that's all part of designing a good game, or even a good competition, right?

Okay, do motor boats make crew obsolete? Maybe crew is now just a historical curiosity, partaken of as a hobby or for general fitness?

To excel at a game you may be good but you will always be bound by the confines of the game. Sure, you might come up with a brand new strategy but you are still thinking inside the box. I say it is better to be the one that, metaphorically speaking, invents the game. Treading on new ground--computers cannot go there.

Athinira:

"Sport" (or E-sport) is in itself a terrible word. "Competition" is much better. You compete in a discipline versus other humans to test who is the better man. Computers, robots or AI's can't change that. It's like saying that sprinting competitions aren't fun because a Cheetah can sprint 60+ MPH. Sure, a Cheetah can run much faster than a human, but you aren't going to have Cheetah's running versus humans as a spectator sport anytime soon, while human sprinting competitions are still a popular discipline.

My point is that what non-human elements can achieve is irrelevant as long as the competition stays between humans. Sure, you can do non-human competitions (Kasparov vs. Deep Blue), if nothing else just for the laughs, but in the end, it shouldn't be compared to human vs. human, and it probably shouldn't be made into a spectator event either.

I agree completely and am particularly fond of the "cheetah vs human" sprinter comparison.

Xersues:
No one ever got an award for wrenching the fastest.

Considering there's a World Championship Lumberjack Competition, I'm not sure your "jobs aren't sports" definition works either. People can compete at just about anything that allows for variation in performance.

Hell, freaking weightlifting is a sport, and that's just picking things up and putting them down again. How much more mundane can you get than that?

Can some games be considered e-sports? I dunno, but "computers play themselves faster than players can" isn't a reason to discard the idea that computer games can be sports.

On a somewhat related note, I still don't think of poker as a sport - I don't care how often ESPN plays it. It's a gamble, not a direct competition. THAT I'd consider more of a reason to find Starcraft's sport status questionable than anything to do with the computer: sometimes it's just a shot in the dark that pays off or a cheese tactic that circumvents 90% of the game. I have a hard time considering a game where you can win by bluffing to be a sport. But then, that's just me, nothing technically defining about the term in it. (Afterthought: many sports can involve bluffing of a sort - acting tired or injured is semi-common in Tennis, for instance, as a way to get a breather or put an opponent's pace off a step; Soccer players exaggerate "injuries" all the time to get fouls called. I'd use a more specific term than bluffing if I could think of one.)

thimblyjoe:
snip

You COULD program an AI to blindly nuke rush, but the only reason it's remarkable in the video is because BoxeR new his opponent wasn't very good. The nuke rush strategy is incrediblt inefficient and risky, which is why you don't usually see it happen. An AI in that situation would have gone with a more conventional strategy.

lomylithruldor:
Yes, you CAN train an AI. See Artificial neural network
The problem is that it will be much too slow to have the reaction needed to beat a pro SC player.

In the case of Deep Blue, since IBM couldn't out think a human, they went for the brute force way with a huge parallel computer that had access to a database of pro chess player matches.

I think it may be possible to make a pro SC AI today, but there's no way it can run on a PC running SCII at the same time. It may take 30 octo-core, but it is possible. It's just not practical enough to put in your living room.

You are right that you can "train"AI, but I was indeed thinking only about conventional desktops, so I didn't mention the crazy supercomputers that scientists are so fond of.

 Pages 1 2 NEXT

Reply to Thread

Log in or Register to Comment
Have an account? Login below:
With Facebook:Login With Facebook
or
Username:  
Password:  
  
Not registered? To sign up for an account with The Escapist:
Register With Facebook
Register With Facebook
or
Registered for a free account here