|
Czech Republic12129 Posts
On March 17 2016 20:27 unholyflare wrote:Show nested quote +On March 13 2016 03:24 Charoisaur wrote:On March 13 2016 03:17 lordsaul wrote:I think people massively underestimate what perfect mechanics does to the game It depends on the rules/limitations placed on the AI, but imagine * Every Medivac always picking up units about to be hit by a stalker and immediately dropping it for the next shot * Marines that always maintain their range advantage on roaches * Tanks that always target the banelings first * Marines that always perfect split v banelings (you can find that online already) * Weak units that always rotate out of the front line * Medivacs healing the most important target in range, rather than the closest * Perfect charges vs tank lines (single units charging ahead of the main attack * ...to name a very few basic micro tricks And while all this happens, perfect macro? Humans overestimate themselves  . Computers won't even need "good" strategy to beat humans, just a large number of difficult to handle micro tricks and beastly macro. The "AI" that will need to be added is just to stop the computer glitching out against weird tricks (e.g. somehow tricking the AI into permanent retreat based on units trying to find perfect range. Edit: Humans are actually at an advantage in Chess and Go, because they are put under far less real time pressure people don't underestimate that. they know the AI would have to be limited for it to be a fair challenge. the point is to show that bots are more intelligent then humans not that they have better mechanics. This was never the point? Certainly as far as chess engines go, they are superior simply because they can brute force calculate in the way that humans can't. Humans have to "teach" engines strategy by assigning values to various strategic aspects. The brute force calculation power of machines in chess/go I would say is roughly the equivalent to mechanics in SC2. It's part of the deal.
For the love of me, READ ABOUT THAT THING ANYTHING. It haven't brute forced the game. Bruteforcing GO is technically near impossible, taht's why the net lost one game(IIRC the score is now 3-1)
|
Yea it was published in 1959. However its truly amazing how much we've gone, from literally not knowing how to calculate flow/shortest distance to being able to beat a human in Go.
|
Russian Federation93 Posts
FYI
Google self-driving cars has driven 1,011,338 in autonomous mode on life streets. It has been in an accident for 12 times and only once by it's fault. Considering that car has not to only decide how to drive, it has to scan the space around it, determine the objects (cars, humans, animals, garbage, marking, road signs, pits, traffic lights) and predict object's behavior in real-time. Starcraft isn't more complex that the real life. And the car AI isn't even 1% as smart as the AlplaGo.
|
On March 17 2016 20:33 Elentos wrote:Show nested quote +On March 17 2016 08:38 StarscreamG1 wrote:On March 17 2016 08:25 Quesadilla wrote:
Seems like most people are already echoing my thoughts too. Unless the AI is limited with the APM/mechanics, no way a person could win. Agree, but with restrictions the test wouldn't make sense. Why not? I'm pretty sure it's easier to create an AI that can beat the best SC players with humanly impossible mechanics than it is to make an AI that can beat them strategically with human-like mechanics. But the 2nd one is way more interesting.
I have a feeling the losing humans would get salty and claim the result isn't legitimate because of the machine's inhuman APM. Until, of course, it gets lowered to a level where they can overpower it mechanically themselves. Do you think they'll ever find an APM cap all parties can agree with?
|
On March 17 2016 20:39 deacon.frost wrote:Show nested quote +On March 17 2016 20:27 unholyflare wrote:On March 13 2016 03:24 Charoisaur wrote:On March 13 2016 03:17 lordsaul wrote:I think people massively underestimate what perfect mechanics does to the game It depends on the rules/limitations placed on the AI, but imagine * Every Medivac always picking up units about to be hit by a stalker and immediately dropping it for the next shot * Marines that always maintain their range advantage on roaches * Tanks that always target the banelings first * Marines that always perfect split v banelings (you can find that online already) * Weak units that always rotate out of the front line * Medivacs healing the most important target in range, rather than the closest * Perfect charges vs tank lines (single units charging ahead of the main attack * ...to name a very few basic micro tricks And while all this happens, perfect macro? Humans overestimate themselves . Computers won't even need "good" strategy to beat humans, just a large number of difficult to handle micro tricks and beastly macro. The "AI" that will need to be added is just to stop the computer glitching out against weird tricks (e.g. somehow tricking the AI into permanent retreat based on units trying to find perfect range. Edit: Humans are actually at an advantage in Chess and Go, because they are put under far less real time pressure people don't underestimate that. they know the AI would have to be limited for it to be a fair challenge. the point is to show that bots are more intelligent then humans not that they have better mechanics. This was never the point? Certainly as far as chess engines go, they are superior simply because they can brute force calculate in the way that humans can't. Humans have to "teach" engines strategy by assigning values to various strategic aspects. The brute force calculation power of machines in chess/go I would say is roughly the equivalent to mechanics in SC2. It's part of the deal. For the love of me, READ ABOUT THAT THING ANYTHING. It haven't brute forced the game. Bruteforcing GO is technically near impossible, taht's why the net lost one game(IIRC the score is now 3-1)
i'm perfectly aware of the limitations of brute forcing the game. Back when chess engines only went on brute force they couldn't beat top humans either because they could outplay the engine strategically.
Nonetheless, *tactically*, in Go and Chess, engines/AI are perfect or near-perfect. And it's the tactics that are like SC2 mechanics.
|
On March 13 2016 08:09 Liquid`Snute wrote:Naive. Of course AIs will be able to beat humans, even with APM/micro limitations (no mineral hax etc). It will take a lot of work to get the AI to such a stage, but a computer's game-sense and execution will be absolute next level, far beyond that of any human. Perfect memory, perfect theory. Obviously the awkward 'mindless machine' quirks will be dealt with in the development of the AI. If the computer is fast enough to process well in "broodwar real-time" with several strategic layers working together (like AlphaGo), humans won't stand a chance. It will take crazy strong computers to do this, but progress is always there. Would be very cool to watch and I hope they undertake the project
I can assure you human brain is very good at fast situational incomplete exercises.The only way to create an ai with same intelligence would be to create a human brain and we are far away from that. How many neurons can we simulate today? That of a bee or two bees?
Dont forget, the ai needs to use the same interface as humans with the difference of image recognition but you need to simulate that it can only see one screen and the minimap. Otherwise the whole experiment misses the point, it would be like playing against a player who is connected to the computer and just has to imagine things.
|
Russian Federation93 Posts
On March 17 2016 22:56 todespolka wrote:
I can assure you human brain is very good at fast situational incomplete exercises.The only way to create an ai with same intelligence would be to create a human brain and we are far away from that. How many neurons can we simulate today? That of a bee or two bees?
More than we need for a computer to drive a car on a real life streets. How many bee brains combined can do this?
On March 17 2016 22:56 todespolka wrote:
Dont forget, the ai needs to use the same interface as humans with the difference of image recognition but you need to simulate that it can only see one screen and the minimap.
Why the hell do you want to translate computer output to a human interface to be than translated back from a human interface to a computer input? Than we should encrypt it 5 times to make it even harder for AI!
You can even limit it's "vision" to 1 square centimeter. He will move it million times per second "scanning" every possible "pixel" on the map for every rendered frame. That incredible mechanical advantage will most likely lead to a flawless victory for a computer. Because, as I was saying like 3 times before:
On March 16 2016 20:11 sh1RoKen wrote:
Starcraft units wasn't designed to be microed by computer.
On March 17 2016 22:56 todespolka wrote:
Otherwise the whole experiment misses the point, it would be like playing against a player who is connected to the computer and just has to imagine things.
That is the whole fucking point of an experiment. To create a program that can imagine things by itself. The program than can learn things and become better than human at things you can't completely calculate through. And that was already done. They just want to test it in different environments. Which you should have been aware of if you've read anything about that AlphaGO before posting arguments based only on your complete ignorance about the subject.
|
France12886 Posts
It's hard not to ad hominem on such people :o ^
|
On March 17 2016 21:50 sh1RoKen wrote:
FYI
Google self-driving cars has driven 1,011,338 in autonomous mode on life streets. It has been in an accident for 12 times and only once by it's fault. Considering that car has not to only decide how to drive, it has to scan the space around it, determine the objects (cars, humans, animals, garbage, marking, road signs, pits, traffic lights) and predict object's behavior in real-time. Starcraft isn't more complex that the real life. And the car AI isn't even 1% as smart as the AlplaGo.
i dont know how complex that self driving car is but AI being able to play starcraft would be like doing everything you can to disrupt the self driving cars; crash a paint truck that covers the whole road, a flood, burning fuel depot, tornado miles ahead, have it determine if it is safe to drive on ice, threat assessment from weird behaving cars and people, etc.
AI being to understand starcraft and act accordingly is beyond my imagination so i would love to see this happen. making an AI to beat a professional bw player sounds like impossible technology at the moment. its mind boggling if google is able to pull it off within a few years. perhaps im greatly underestimating current AI technology but i'd estimate at least 5 and probably closer to double digits in years to accomplish this.
a simple way of if player does X, AI must respond with Y will not work. it is truly being able to asses threat, determine the information to be fake or not, intentional or not, and after all that decide what to do.
|
On March 17 2016 21:50 sh1RoKen wrote:
FYI
Google self-driving cars has driven 1,011,338 in autonomous mode on life streets. It has been in an accident for 12 times and only once by it's fault. Considering that car has not to only decide how to drive, it has to scan the space around it, determine the objects (cars, humans, animals, garbage, marking, road signs, pits, traffic lights) and predict object's behavior in real-time. Starcraft isn't more complex that the real life. And the car AI isn't even 1% as smart as the AlplaGo.
the user above me already said a bit about this, but i want to add something none the less:
"it has been in an accident for 12 times" - and we are assuming, that even those people causing these accident, didn't want to make them themselfs.
a game is a total different situation: everyone wants to hit you. the car failed 12 times, even thou noone tried to hit it. that is a really bad number, not a good one.
/edit: oh and i just saw "the car has driven [number] in autonomous [...]" [number] what? seconds? hours? miles? km? lightyears? O_o
|
On March 17 2016 21:50 sh1RoKen wrote:
FYI
Google self-driving cars has driven 1,011,338 in autonomous mode on life streets. It has been in an accident for 12 times and only once by it's fault. Considering that car has not to only decide how to drive, it has to scan the space around it, determine the objects (cars, humans, animals, garbage, marking, road signs, pits, traffic lights) and predict object's behavior in real-time. Starcraft isn't more complex that the real life. And the car AI isn't even 1% as smart as the AlplaGo.
not sure being able to drive a car has any correlation with being able to play Starcraft at a high level. Also, they've been working on that since 2009 and its still not ready, in fact, other companies are beating them to the chase (Honda, Tesla, rumors about Uber as well).
|
An interesting thought I've had recently is...is Starcraft truly in real time? It doesn't seem right to assume that your brain thinks completely continuously since space itself can't truly be considered continuous. Wouldn't vectorizing the game with a really small deltaT be kind of computationally equivalent to the Go game? Maybe not equivalent, it may even be like at each point in time a sort of Go level game has to be evaluated, but at our rate of technology I imagine we'll be seeing a successful Starcraft AI in our lifetimes.
EDIT: I think a reasonable deltaT can even be chosen at this point. It'd likely be (the time at which the average human brain cannot register additional information delivered at a certain framerate) + (the time the "average" mouse move physically takes). I'm assuming the time for a thought to be transmitted to the hand is negligible.
|
On March 18 2016 06:52 mierin wrote:
An interesting thought I've had recently is...is Starcraft truly in real time? It doesn't seem right to assume that your brain thinks completely continuously since space itself can't truly be considered continuous. Wouldn't vectorizing the game with a really small deltaT be kind of computationally equivalent to the Go game? Maybe not equivalent, it may even be like at each point in time a sort of Go level game has to be evaluated, but at our rate of technology I imagine we'll be seeing a successful Starcraft AI in our lifetimes.
EDIT: I think a reasonable deltaT can even be chosen at this point. It'd likely be (the time at which the average human brain cannot register additional information delivered at a certain framerate) + (the time the "average" mouse move physically takes). I'm assuming the time for a thought to be transmitted to the hand is negligible.
Technically StarCraft is based on frames. At each frame, the actions of all players are handled and the game logic is run. So you could theoretically analyze it this way. The main problem is the very large search space. Much larger than that of GO.
|
Starcraft is a very hard game and probably a good choice for them, but they will have more technical difficulties with a video game verses board games. With the resources and knowledge they have im sure their bot will be on a level and scale far beyond any starcraft so far. Its going to be a tall order though especially if you consider all 3 races, the huge variety of maps, the slight luck involved with scout timings and things. Not at all easy and to be honest its probably one of the hardest games left.
They are running out of hard games GO was one of the hardest that exists. They been trying to win at that for like 30 years and I am pretty amazed they were able to do it. Their AI is truly learning in almost every sense of the word its just at one specific domain still
|
I have a feeling that most people are either underestimate the complexity of Go (due to it's not being well-known in the west), overestimate the complexity of Starcraft (due to not understanding heuristic or bias), or not understand how the new AI technology work (due to have not read the Nature's paper yet).
Out of curiosity, how many people here have read or skim through the Nature's paper that describe how Alphago works?
|
On March 18 2016 09:38 Veldril wrote:
I have a feeling that most people are either underestimate the complexity of Go (due to it's not being well-known in the west), overestimate the complexity of Starcraft (due to not understanding heuristic or bias), or not understand how the new AI technology work (due to have not read the Nature's paper yet).
Out of curiosity, how many people here have read or skim through the Nature's paper that describe how Alphago works?
http://www.teamliquid.net/forum/viewpost.php?post_id=25502046
Read page 2 section A of the pdf
|
On March 18 2016 14:16 necrosexy wrote:Show nested quote +On March 18 2016 09:38 Veldril wrote:
I have a feeling that most people are either underestimate the complexity of Go (due to it's not being well-known in the west), overestimate the complexity of Starcraft (due to not understanding heuristic or bias), or not understand how the new AI technology work (due to have not read the Nature's paper yet).
Out of curiosity, how many people here have read or skim through the Nature's paper that describe how Alphago works? http://www.teamliquid.net/forum/viewpost.php?post_id=25502046Read page 2 section A of the pdf
You are confusing state space with complexity. What's the state space for throwing a basket ball in real life? That would be utterly impossible to do for an AI, right?
|
Hey guys, why are you even arguing? You totally misread Boxer (and Flash). What they simply said was :
- "hey there is plenty of money in these show matches, please pick me!"
|
On March 18 2016 15:07 Mendelfist wrote:Show nested quote +On March 18 2016 14:16 necrosexy wrote:On March 18 2016 09:38 Veldril wrote:
I have a feeling that most people are either underestimate the complexity of Go (due to it's not being well-known in the west), overestimate the complexity of Starcraft (due to not understanding heuristic or bias), or not understand how the new AI technology work (due to have not read the Nature's paper yet).
Out of curiosity, how many people here have read or skim through the Nature's paper that describe how Alphago works? http://www.teamliquid.net/forum/viewpost.php?post_id=25502046Read page 2 section A of the pdf You are confusing state space with complexity. What's the state space for throwing a basket ball in real life? That would be utterly impossible to do for an AI, right?
Didn't realize there was an AI that can beat NBA players!
|
Why is this thread going on ? I'm sure it was stated on every page that no AI will make a dragoon go up a ramp. Nor a goliath (ultralisk can apply)
|
|
|
|
|
|
EPT
