|
On March 13 2016 08:58 DuckloadBlackra wrote:Show nested quote +On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information.
You're overlooking human deceptiveness.
Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive?
If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff?
|
"Even if it has studied all of the many strategies I've used, I'll go at it with an unstoppable strategy I've prepared."
Am I the only one who read this in Goku's voice?
|
On March 13 2016 09:15 iFU.pauline wrote:
Most of people are saying that an AI would beat a human at starcraft based of the assumption that in the future it will. That's stupid... How do you want to debate with that type of arguments :/ The thread is about AlphaGo present day. And it is nowhere near that. And even if it manages to do it in 50 years when no one is left playing Starcraft. What's the point...
Anyway, the thing that would definitely settle this debate is to know if all problems can be solved by calculations...
If yes, then eventually AI would be capable just by itself to run a campaign and become president. Or eradicate say, violence in the world.
Now talk about taking it to the next level...
Then it is obvious that humans can beat the AI because a human with any skill at all can easily defeat the native AI in SC.
What makes the question intriguing is how to break down the problem to create an AI that is clearly better than any human (by beating the world champion).
Someone has to ask the ultimate balance questions: which race does the AI master first? And if you play the perfect AI from each race against each other, would one be clearly better than the others?
|
France12902 Posts
On March 13 2016 09:28 coverpunch wrote:Show nested quote +On March 13 2016 09:15 iFU.pauline wrote:
Most of people are saying that an AI would beat a human at starcraft based of the assumption that in the future it will. That's stupid... How do you want to debate with that type of arguments :/ The thread is about AlphaGo present day. And it is nowhere near that. And even if it manages to do it in 50 years when no one is left playing Starcraft. What's the point...
Anyway, the thing that would definitely settle this debate is to know if all problems can be solved by calculations...
If yes, then eventually AI would be capable just by itself to run a campaign and become president. Or eradicate say, violence in the world.
Now talk about taking it to the next level... Then it is obvious that humans can beat the AI because a human with any skill at all can easily defeat the native AI in SC. What makes the question intriguing is how to break down the problem to create an AI that is clearly better than any human (by beating the world champion). Someone has to ask the ultimate balance questions: which race does the AI master first? And if you play the perfect AI from each race against each other, would one be clearly better than the others?
There is no transitivity in e-sports so beating the world champion doesn't necessarily mean you can beat everyone tho xd.
But in the mind of the majority it means exactly that, which is what they care about
|
On March 13 2016 09:17 necrosexy wrote:Show nested quote +On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff?
What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units
and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures.
An AI has no center of attention like we humans do, it does not need so called awareness like us humans. It sees what is and what is not in an instant and does not question itself or its decisions.
|
On March 13 2016 09:12 Poopi wrote:Show nested quote +On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You can prevent scouting from happening, you can cancel buildings or fly CC, you can kill your own units... or many other random things. The incomplete information is a really tricky issue, as well as the notion of "perfect" micro. The fun thing is that even if you make two AI play against each other with unlimited APM and complete information, maybe (speaking of sc2) you can't solve the game or one race is indeed overpowered or whatever, thus in the "real" game you can't ensure win either. So being too confident in either side is probably a bad idea.
I actually can't speak for BW since I don't know a lot about it, I was thinking in SC2 terms and forgetting the context is BW. In SC2 there's no way you would be able to deny the AI gathering sufficient information to at least stay on equal footing if it was good enough at it. It would take into account the possibility of canceling buildings/flying CC (of course this delves into very complicated territory but that isn't the point) and killing your own units is very rarely if ever a useful idea. I agree this is a very tricky issue, but doable.
|
France12902 Posts
On March 13 2016 09:35 thePunGun wrote:Show nested quote +On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures.
Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's
|
I think at the end it all comes down to this :
Limit the human mind to games based on calculation tasks, and eventually machine will win.
Add other variables in a game where feelings has a deep involvement in winning, then the machine will be easily outclassed.
Present day, I don't think AI can match a human in a game like Brood war because of the big feeling variable, and this is exactly what Boxer meant by giving the "scouting" example.
|
On March 13 2016 09:36 Poopi wrote:Show nested quote +On March 13 2016 09:35 thePunGun wrote:On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's
which is another plus for the AI, it won't need a keyboard or fingers to execute commands. It will be faster than any human with a keyboard.
|
France12902 Posts
On March 13 2016 09:41 thePunGun wrote:Show nested quote +On March 13 2016 09:36 Poopi wrote:On March 13 2016 09:35 thePunGun wrote:On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's which is another plus for the AI, it won't need a keyboard or fingers to execute commands. It will be faster than any human with a keyboard.
Then it doesn't count :o
|
On March 13 2016 09:17 necrosexy wrote:Show nested quote +On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff?
Hence why I said "sufficiently advanced." This isn't a matter of how it will do it, but rather a matter of being capable of dealing with the possibility of deception. I can't tell you exactly how AlphaGo deals with all the problems it faces in the game of Go, but it does. Not being able to explain how doesn't mean it can't be done and I'm convinced it will be possible.
|
On March 13 2016 09:42 Poopi wrote:Show nested quote +On March 13 2016 09:41 thePunGun wrote:On March 13 2016 09:36 Poopi wrote:On March 13 2016 09:35 thePunGun wrote:On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's which is another plus for the AI, it won't need a keyboard or fingers to execute commands. It will be faster than any human with a keyboard. Then it doesn't count :o
When you play vs the SC AI, do you think it's using a keyboard? xD
|
France12902 Posts
On March 13 2016 09:45 thePunGun wrote:Show nested quote +On March 13 2016 09:42 Poopi wrote:On March 13 2016 09:41 thePunGun wrote:On March 13 2016 09:36 Poopi wrote:On March 13 2016 09:35 thePunGun wrote:On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's which is another plus for the AI, it won't need a keyboard or fingers to execute commands. It will be faster than any human with a keyboard. Then it doesn't count :o When you play vs the SC AI, do you think it's using a keyboard? xD
SC ingame AI is irrelevant since they artificially rise the difficulty for the player by cheating with different ways.
What Google wants is showing mustles and beat humans with superior decision making, if it becomes easy to win by cheating there is no interest for them.
|
The SC ingame "AI" isn't even really an AI, it's a bot.
|
On March 13 2016 09:53 Poopi wrote:Show nested quote +On March 13 2016 09:45 thePunGun wrote:On March 13 2016 09:42 Poopi wrote:On March 13 2016 09:41 thePunGun wrote:On March 13 2016 09:36 Poopi wrote:On March 13 2016 09:35 thePunGun wrote:On March 13 2016 09:17 necrosexy wrote:On March 13 2016 08:58 DuckloadBlackra wrote:On March 13 2016 08:35 writer22816 wrote:
Since SC isn't a perfect information game, it stands to reason that a good human player should always have the ability to at least take occasional games off of an AI. Nevertheless, even though I love Boxer and Flash, they're kidding themselves if they think that there will never be an AI that can reliably take games off of them. Most people in the gaming community think AIs are a joke because bots in video games are always easy to beat. If a company like Google or IBM threw significant resources into making a video game AI, these people would very quickly be eating their words lol. There is nothing in either Starcraft game that remotely approaches the intractability of Go, and mechanics-wise a good AI would be able to completely shit on any human player. A sufficiently advanced AI would be able to do all the scouting it needs to gain enough information to win every time. It would remember everything perfectly and calculate the implications of what it learns with extreme precision. The biggest challenge is programming the decisions it will need to make based off this information. You're overlooking human deceptiveness. Computer scouts, e.g., a reactored factory and a starport. So it's a drop (or is it?), but when and where will it arrive? If the AI detects an enemy scan, how does he interpret this? Does the scan mean dropping at the scan location or is it a bluff? What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. Problem is the correct counter measures vary depending of your own mechanical skills as well as your opponent's which is another plus for the AI, it won't need a keyboard or fingers to execute commands. It will be faster than any human with a keyboard. Then it doesn't count :o When you play vs the SC AI, do you think it's using a keyboard? xD SC ingame AI is irrelevant since they artificially rise the difficulty for the player by cheating with different ways. What Google wants is showing mustles and beat humans with superior decision making, if it becomes easy to win by cheating there is no interest for them.
Well an AI using a robotic arm will probably still be faster (some day). However, current robotics are not on the same level as a human arm and are even worse when it comes to hands...
|
well it depends. how good is this Ai? ive seen some of those BW AIs that are supposedly good but in reality they suck even with their awesome mechanics.
Baduk is so simple compared to SC. dont really see why there is an argument that this AI can handle something like SC where BOs, scouting, and meta/counters come into play.
|
On March 13 2016 05:33 Oshuy wrote:Show nested quote +On March 13 2016 03:57 MyLovelyLurker wrote:On March 13 2016 03:53 Oshuy wrote:On March 13 2016 02:56 brickrd wrote:it's not a question of "if," it's a question of when. maybe not in 5 years, maybe not in 10 years, but nothing is going to stop AI from getting better and becoming able to excel in complex tasks. they said the same thing about chess, same thing about go, same thing about lots of computerized tasks. it's cute that he thinks it's not possible, but there's no reasonable argument outside of "when will it happen" On March 13 2016 02:50 Musicus wrote:
All this talk about something that might or might not happen in 5 to 10 years. When the challenge is out and the date is set, I will get excited. sorry for finding science interesting! The "maybe not in 10 years" sounds hopeful. Deepmind was created in 2010. Alphago is 18months old (as in : the project started 18 months ago). There is a hurdle to design what to feed to the neural networks and how to represent the output in a game of starcraft : the space both of current status and potential action are huge; but once those representation are designed, the learning process will either fail or succeed in a few months. The fact that information is incomplete is almost irrelevant in case of a neural network feed. Those are the type of problems we designed networks for in the first place. Real time and information retention may make things more difficult, but it could get there fast. It's actually not irrelevant in reinforcement learning, as you need to compute a conditional expectation of the state of play with respect to the information you have - and the update of said expectation will change algorithms by quite a lot. This is being tackled almost as we speak, here is a two weeks old article on the subject - from one of the fathers of AlphaGo - with an application to poker : arxiv.org Building the dataset for supervised learning from replay databases consisting of both the incomplete information (one player view) and the complete information (spectator view) should provide a first estimate of a potential convergence for a given game representation. Self-play reinforcement would be great; agreed, I have no idea how to construct an evaluation function (and quite sure it cannot be done on individual actions that are mostly meaningless in themselves). Unsure if it would be necessary at this point (why isn't supervised all the way with a spectator AI impossible ?). Interesting part in the self-play is that the AI would get to the match with its own metagame that the human players faces for the first time during the match, while the human metagame will have been the basic dataset the AI learned from initialy.
I agree. Self-play reinforcement is what Google Deepmind is aiming for, but it might be easier to start a hybrid approach with replays first. In my opinion they will probably even have to settle for a 'Starcraft for dummies' subset first, with only workers and couple units first, mirroring the 'one unit at a time' learning curve you get from campaign play.
|
On March 13 2016 10:08 Golgotha wrote:
well it depends. how good is this Ai? ive seen some of those BW AIs that are supposedly good but in reality they suck even with their awesome mechanics.
Baduk is so simple compared to SC. dont really see why there is an argument that this AI can handle something like SC where BOs, scouting, and meta/counters come into play.
Nobody ever said this AI can handle any of that yet. They haven't even begun work on anything for this game yet, but they expressed interest in doing so in the future.
|
United States7483 Posts
A computer is very capable of executing well, but it sucks horribly at metagaming and situational reading. It has no star sense, for example, no gut instinct.
For this reason, there's never been an AI that can compete with human players at bridge. They have terrible table sense.
Computers do wonderfully in games of perfect information, but they actually are not very good at all at games of imperfect information.
|
Surely one has to understand there's pride in this, especially from someone like Boxer
A few bits of info of how AI could work:
#1 - Self-Learning AI = this is the "robotic" approach where you give input data on an AI what winning condition is, what losing condition is and leave it to itself to figure out over thousands and millions of iterations for self-learning process
#2 - Database AI = Blizz has millions of games in their database as well as perfect information from replays = this approach is probably the best for fast results but with obvious flaws cause it will just copy human potential
#3 - Stick to one race/build and develop the perfect micro to complement that. For example mass Marine/CC into Tankyvacs
#4 - I know people would hate me for this but it's the perfect approach to make an unbeatable AI = statistically safest opener with perfect micro only for mirror matchups
|
|
|
|
|
|
EPT
