EPTBoxeR: "AlphaGo won't beat humans in StarCraft" - Page 2
| Forum Index > SC2 General |
|
duke91
Germany1458 Posts
| ||
stuchiu
Fiddler's Green42661 Posts
On March 13 2016 02:52 Heyoka wrote: I mean your duty as a SC player is to thump your chest and say "no machine will beat me!" right? It's not like anyone is going to have a reasonable discussion about this in interviews. Should have asked Innovation. As a robot with a human face I feel he'd have a more objective view of the matter. | ||
|
Charoisaur
Germany15900 Posts
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. | ||
|
Scarlett`
Canada2381 Posts
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 wrong game fam | ||
Nakajin
Canada8988 Posts
On March 13 2016 03:18 brickrd wrote: lunar colonies have nothing to do with adaptive AI winning at complex competitive games... the fact that an AI beat the go champion earlier than expected has everything to do with it... just because we don't know for certain that doesn't mean that we shouldn't use relevant reference points to make predictions... Just an example that science don't always progress as fast as we might predict, maybe in a year deepmind will beat Flash, the point is I don't think everyone here is a AI programming specialist, so saying it happen will in six month or in 15 years by comparing it to go is not realy fair. (At least for me if you have inside information in the process of AI development, and the challenge that pose playing Starcraft then of course it is different) | ||
|
Musicus
Germany23576 Posts
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" sorry for finding science interesting! I find the science extremely interesting and love following DeepMind vs Lee Se Do. I'm not talking about the people discussing the possibilities here. I just find the interviews with StarCraft pros pretty boring. They won't say "I think I will lose" and none of them has been challenged yet and we don't know if they ever will be. On March 13 2016 03:01 Garrl wrote: people thought Go AI beating professional players was a long way off. AFAIK deepmind's project is a generalized solution that takes only pixel data as an input. could be far, far closer than you might think. I think the number "5-10 years" came from people on the Deepmind team. Can't give you a source though, read it somewhere :/. | ||
|
AsAr
Germany52 Posts
| ||
|
Empirimancer
Canada1024 Posts
IMO the only interesting question here is how long it will take for AlphaStarcraft to beat pro-gamers consistently starting from now. I would say probably less than 2 years, definitely less than 3 years. Perhaps less than 1 year. | ||
|
Salteador Neo
Andorra5591 Posts
 | ||
|
JimmyJRaynor
Canada16648 Posts
On March 13 2016 03:30 Empirimancer wrote: So... Boxer admits that he knows nothing about AI or AlphaGo, but still says it will never exceed human capabilities? When literally a few months ago, some Go players were saying the exact same thing? Talk about overconfidence. IMO the only interesting question here is how long it will take for AlphaStarcraft to beat pro-gamers consistently starting from now. I would say probably less than 2 years, definitely less than 3 years. Perhaps less than 1 year. what i really want to see is AI versus AI competitions like already exists today .. but with giant prize pools. in FirePro Wrestling there is an entire hardcore wagering community built around AI v AI matches | ||
|
AdrianHealeyy
114 Posts
It's probably not that hard to come up with an AI that can have perfect micro. The trick is: can we design an AI with 'human' micro that can still consistently beat humans, based on insight, analysis, response, etc.? That would be the ultimate challenge. I still think they can do it, but it'll take longer. | ||
|
nohole
United States56 Posts
| ||
|
Oshuy
Netherlands529 Posts
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" 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. | ||
|
[PkF] Wire
France24192 Posts
On March 13 2016 03:47 AdrianHealeyy wrote: I think we need to differentiate two things here. It's probably not that hard to come up with an AI that can have perfect micro. The trick is: can we design an AI with 'human' micro that can still consistently beat humans, based on insight, analysis, response, etc.? That would be the ultimate challenge. I still think they can do it, but it'll take longer. The problem is how do you define human micro (and even human multitask). A simple limit on the APM wouldn't even be enough I think, since the computer doesn't spam and -more importantly- sees all screens at once. | ||
|
MyLovelyLurker
France756 Posts
On March 13 2016 03:53 Oshuy wrote: 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 | ||
|
Aron Times
United States312 Posts
That's what we're up against, and I doubt the Dominion will win. Hey, Dustin Browder, i just had an idea for Starcraft 3: Bio vs. Mech. Terran, Zerg, and Protoss vs. an unstoppable race of intelligent machines. | ||
|
OtherWorld
France17333 Posts
| ||
|
Shield
Bulgaria4824 Posts
| ||
Waxangel
United States33280 Posts
On March 13 2016 03:19 duke91 wrote: Why is this again posted in SC2 forum instead of Broodwar where it is obvious it is about Broodwar, talking about Boxer who majorly played BW, as well as Flash who is back at BW talking about BW AI implemented in BW. At least put the same post in the BW forum as well. I don't see any relevance of this topic being in the SC2 forum because I care about visibility more than adherence to TL's outdated forum structure | ||
|
Grumbels
Netherlands7031 Posts
On March 13 2016 04:01 Eternal Dalek wrote: I would be impressed by an AI that can win in DotA, where decision-making matters far more than mechanics. In Starcraft, it would be no contest. We've seen all the macro and micro hacks and bots over the years, and they're unstoppable by most players. Bear in mind that these hacks weren't made by Google, who has pretty much unlimited money to spend on development. Imagine a brilliant hacker who doesn't have to worry about paying his/her bills, free to devote everything into AI development. That's what we're up against, and I doubt the Dominion will win. Hey, Dustin Browder, i just had an idea for Starcraft 3: Bio vs. Mech. Terran, Zerg, and Protoss vs. an unstoppable race of intelligent machines. There are some disadvantages to DotA though: there are still patch changes, so that you can't easily train a bot on one specific patch; and it's not so easy for an AI to mass practice games vs itself since you might be stuck with Steam or with 45 mins per game and those sort of things. And there's the question of whether you should have an AI control all five heroes at once, since that might be seen as cheating since it'll have perfect coordination. | ||
| ||
