EPT
StarCraft II: DeepMind Demonstration: Jan 24
| Forum Index > SC2 General |
|
keaneu
Korea (South)65 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Deleted User 3420
24492 Posts
| ||
|
mierin
United States4943 Posts
| ||
|
Deleted User 3420
24492 Posts
| ||
|
Xamo
Spain877 Posts
| ||
|
Charoisaur
Germany15942 Posts
| ||
|
paralleluniverse
4065 Posts
And I'd bet it will use Terran, because its skill with the other 2 races is bad. | ||
Nakajin
Canada8989 Posts
| ||
|
ZigguratOfUr
Iraq16955 Posts
| ||
|
Dave4
494 Posts
| ||
|
stardog
556 Posts
| ||
|
shabby
Norway6402 Posts
| ||
|
Ziggy
South Korea2105 Posts
humans get region locked out, only AI competitors | ||
|
BaneRiders
Sweden3630 Posts
| ||
|
Tayewo
Germany28 Posts
| ||
|
scotch4789
United States42 Posts
On January 23 2019 05:22 Dave4 wrote: I for one welcome our new cyborg overlords. Hahaha! a most excellent post! | ||
|
srj
Canada134 Posts
| ||
|
Ronski
Finland266 Posts
On January 23 2019 05:14 ZigguratOfUr wrote: I hope the deepmind team is more open about what they produce. Show-matches are all very well, but giving players the opportunity to out-mindgame the AI afterwards would be interesting. AlphaZero was somewhat disappointing in the sense that no one really has a good sense of exactly how good it is at Shogi or Chess. Didn't they make it pretty clear that its the best chess engine there is atm? Beating the strongest engine at chess means that no human player could ever hope to beat it so at least when it comes to chess I would say its clear that AlphaZero is the best there is. | ||
|
jy_9876543210
265 Posts
| ||
|
ZergX
France436 Posts
| ||
|
vyzion
308 Posts
| ||
|
KaiserCommander
Mexico290 Posts
On January 23 2019 06:45 vyzion wrote: eventually deepmind and similar AI will be unbeatable by humans. Assuming they'll keep improving her neural network yes. In it's actual form thera are limitations of what and how much can it learn. | ||
|
figq
12519 Posts
| ||
|
alexanderzero
United States659 Posts
| ||
|
ZigguratOfUr
Iraq16955 Posts
On January 23 2019 06:17 Ronski wrote: Didn't they make it pretty clear that its the best chess engine there is atm? Beating the strongest engine at chess means that no human player could ever hope to beat it so at least when it comes to chess I would say its clear that AlphaZero is the best there is. I mean probably? But even when their paper was eventually released, it's still just a bunch of games against an old version of Stockfish in circumstances completely controlled, set up, and chosen to be favourable by the Deepmind team. The newest version of Stockfish can also beat the older version of Stockfish by about the same margin. But arguing who is the best and stuff like that isn't too meaningful in the first place (it isn't of any importance if AlphaZero is the best or the second best)--the important thing is the machine learning research. And with Deepmind controlling everything about their research there's no room for other people to investigate things like whether AlphaZero with the current training would also be able to play Chess960 or adapt to starting with a piece handicap and so on and so forth. It would be very disappointing if AlphaStarcraft came out and crushed Serral, Maru and Stats in showmatches and got shelved never to see the light again, leaving people to wonder about how AlphaStarcraft would react to (for example) playing on an island map, or how it would defend a cannon rush. | ||
|
mierin
United States4943 Posts
On January 23 2019 10:14 alexanderzero wrote: I'm pretty skeptical like everyone else, especially considering the state of their AI when it was shown at Blizzcon just three months ago. Still, it's weird for them to bring on professional commentators and livestream something if they don't have something impressive to show. Right. this is what I think. Like...who cares if deepmind can micro units to grab mineral or gas deposits faster than humans or other AIs? If deepmind really thinks people will be impressed by an AI "solving" a limited minigame...I mean, we can already do that. If the AI can compete from the barebones beginnings that regular human players do, that'd be something. I'm pessimistic but will watch regardless. | ||
|
neutralrobot
Australia1025 Posts
On January 23 2019 10:53 ZigguratOfUr wrote: I mean probably? But even when their paper was eventually released, it's still just a bunch of games against an old version of Stockfish in circumstances completely controlled, set up, and chosen to be favourable by the Deepmind team. The newest version of Stockfish can also beat the older version of Stockfish by about the same margin. But arguing who is the best and stuff like that isn't too meaningful in the first place (it isn't of any importance if AlphaZero is the best or the second best)--the important thing is the machine learning research. And with Deepmind controlling everything about their research there's no room for other people to investigate things like whether AlphaZero with the current training would also be able to play Chess960 or adapt to starting with a piece handicap and so on and so forth. It would be very disappointing if AlphaStarcraft came out and crushed Serral, Maru and Stats in showmatches and got shelved never to see the light again, leaving people to wonder about how AlphaStarcraft would react to (for example) playing on an island map, or how it would defend a cannon rush. Well, actually... They recently played more games vs Stockfish in better conditions and AlphaZero comprehensively destroyed Stockfish. Also, they released the algorithm, which might not be as open as releasing the code or the trained network, but it did mean that the algorithm was implemented in a more open manner in the Leela Chess Zero project, which is now pretty competitive with Stockfish and playing interesting games against it in the TCEC. (https://www.youtube.com/watch?v=UPkcAS2B60s) This is the generalized Alpha Zero algorithm -- can be applied to a variety of games. So if they follow that pattern, maybe with Starcraft they'll shelve their code but release the research, which means it can be replicated. Guess we'll see! Keen to see what they've come up with. You'd think it must be a big leap. Bear in mind that once they had the right algorithm, they could train AlphaZero in a matter of hours and get it to a point where it's the best in the world by a mile. They have an incredible ability to test and implement learning algorithms quickly. Part of what gives them such an edge is their TPU hardware. So once there's been a breakthrough it could go from "how do we do this?" to "HOLY SHIT!" in a very short timeframe. | ||
|
Deleted User 3420
24492 Posts
https://github.com/deepmind/pysc2 | ||
|
KalWarkov
Germany4126 Posts
On January 23 2019 11:16 neutralrobot wrote: Well, actually... They recently played more games vs Stockfish in better conditions and AlphaZero comprehensively destroyed Stockfish. Also, they released the algorithm, which might not be as open as releasing the code or the trained network, but it did mean that the algorithm was implemented in a more open manner in the Leela Chess Zero project, which is now pretty competitive with Stockfish and playing interesting games against it in the TCEC. (https://www.youtube.com/watch?v=UPkcAS2B60s) This is the generalized Alpha Zero algorithm -- can be applied to a variety of games. So if they follow that pattern, maybe with Starcraft they'll shelve their code but release the research, which means it can be replicated. Guess we'll see! Keen to see what they've come up with. You'd think it must be a big leap. Bear in mind that once they had the right algorithm, they could train AlphaZero in a matter of hours and get it to a point where it's the best in the world by a mile. They have an incredible ability to test and implement learning algorithms quickly. Part of what gives them such an edge is their TPU hardware. So once there's been a breakthrough it could go from "how do we do this?" to "HOLY SHIT!" in a very short timeframe. until alpha zero beats stockfish in TCEC finals, i will never call alpha zero the strongest engine. everything is controlled by google. no table base, no opening books - which sf isnt trained for. and still, it isn't live games vs sf11dev. and who knows if they released all games or are just cherry picking? | ||
|
ZigguratOfUr
Iraq16955 Posts
On January 23 2019 11:16 neutralrobot wrote: Well, actually... They recently played more games vs Stockfish in better conditions and AlphaZero comprehensively destroyed Stockfish. Also, they released the algorithm, which might not be as open as releasing the code or the trained network, but it did mean that the algorithm was implemented in a more open manner in the Leela Chess Zero project, which is now pretty competitive with Stockfish and playing interesting games against it in the TCEC. (https://www.youtube.com/watch?v=UPkcAS2B60s) This is the generalized Alpha Zero algorithm -- can be applied to a variety of games. So if they follow that pattern, maybe with Starcraft they'll shelve their code but release the research, which means it can be replicated. Guess we'll see! Keen to see what they've come up with. You'd think it must be a big leap. Bear in mind that once they had the right algorithm, they could train AlphaZero in a matter of hours and get it to a point where it's the best in the world by a mile. They have an incredible ability to test and implement learning algorithms quickly. Part of what gives them such an edge is their TPU hardware. So once there's been a breakthrough it could go from "how do we do this?" to "HOLY SHIT!" in a very short timeframe. They played a newer version of Stockfish in somewhat better conditions, and released some implementation details (far from releasing the entire algorithm). I certainly hope they'll be much more open this time. | ||
|
imCHIEN
14 Posts
vs Maru to see how AI deals with his creative vs Serral to see how AI deals with a strong late game opponent. | ||
Excalibur_Z
United States12235 Posts
On January 23 2019 11:45 travis wrote: The event could be a way of issuing an open challenge to independent programmers and other professional groups, deepmind did release their API after all. https://github.com/deepmind/pysc2 That's more or less what I'm expecting as well. They'll probably have some showcases like how the AI handles different types of tasks like micro, adapting build order based on scouting information, which parts of the map it prefers to hold, how it learns different types of maps, and so on. The rest will probably be left up to third parties to develop further. I'm certainly not expecting a full exhibition series, especially not against top-level competition. If that is what they present, though, I'll be absolutely floored. That said, if it is a surprise exhibition, then it means the AI has already been developing skills against real players on the ladder ahead of this event, in which case it would be amusing to speculate who exactly that is. | ||
|
blunderfulguy
United States1415 Posts
| ||
|
MrMischelito
347 Posts
| ||
|
MrMischelito
347 Posts
On January 23 2019 13:12 Excalibur_Z wrote: That's more or less what I'm expecting as well. They'll probably have some showcases like how the AI handles different types of tasks like micro, adapting build order based on scouting information, which parts of the map it prefers to hold, how it learns different types of maps, and so on. The rest will probably be left up to third parties to develop further. I'm certainly not expecting a full exhibition series, especially not against top-level competition. If that is what they present, though, I'll be absolutely floored. That said, if it is a surprise exhibition, then it means the AI has already been developing skills against real players on the ladder ahead of this event, in which case it would be amusing to speculate who exactly that is. Isn't botting something that is completely against Blizzards Terms and Conditions? I wonder how many accounts they got banned from battle.net already... | ||
|
alexanderzero
United States659 Posts
On January 23 2019 15:29 MrMischelito wrote: Isn't botting something that is completely against Blizzards Terms and Conditions? I wonder how many accounts they got banned from battle.net already... Blizzard can always make an exception if they want to. | ||
|
Akio
Finland1838 Posts
| ||
|
Akio
Finland1838 Posts
Also I wonder what race it would play, maybe Terran? The production mechanics are the "easiest" so that's my guess. | ||
|
FFW_Rude
France10201 Posts
On January 23 2019 03:22 Charoisaur wrote: Probably Deepmind vs Serral showmatch FlaSh quit SC2 a long time ago man... | ||
|
Poopi
France12836 Posts
| ||
|
Loccstana
United States833 Posts
| ||
|
MockHamill
Sweden1798 Posts
| ||
|
deacon.frost
Czech Republic12129 Posts
On January 23 2019 16:47 Loccstana wrote: I hope we will get a Bo31 showmatch between deepmind and avilo! Nah, if the Deepmind is really good let it play Avilo so Avilo doesn't know he's playing it. And let us bet how many cheater calls will be made. Then we can give those money to some charity | ||
|
Lazzarus
Faroe Islands114 Posts
| ||
|
Grumbels
Netherlands7031 Posts
There are some interesting quirks with Leela. For instance, it's not capable of playing endgames efficiently, it seemingly aimlessly moves around, making moves that don't lose the advantage. It doesn't "get to the point". If an SC2 AI is built on the same concept, expect it to not be able to finish off games quickly and take an hour to mine out the entire map and build a fleet of random units to randomly move around the map. Another quirk of the project is that the algorithm uses not just the current move as input, but also the history of moves. This gives it some measure of what part of the board to pay "attention" to. It also means that if you give it a random position as input, without history, that it can't function. As far as I know Leela is useless in solving tactical puzzles and in handicap games without training it first. Leela also typically doesn't understand theory of endgames. It doesn't just play them weirdly, but it also doesn't grasp some almost mathematical ideas such as identifying a class of endgames that are drawn despite material imbalances (opposite color bishops, wrong color bishop). It's apparently also not better at fortress positions, where you have material disadvantages, but your position can't be cracked. There are some known positions like these, and it was hoped that neural networks would be better at them, and would be capable of reasoning that these are a special class of positions that require a different approach. But it doesn't really seem like it. Leela is also probably already better than Stockfish if you have bad hardware and no opening book. You can imagine that if there was a market for SC2 bots, that they could have opening books updated for every patch and which would have a team of people dedicated to keeping track of the meta and adding knowledge of it to the bot. But Deepmind's AI would use self-learning, i.e. only playing itself and developing its own meta. I don't know if that would make it easier or harder to beat as a human. I think the tree-search method for chess is bound to scale better with hardware than a neural network approach, given that chess is theoretically solvable with tree search. But this method would be useless for SC2, unless the AI uses some sort of abstraction of strategy and tries to think ahead. But I don't think you really need to think ahead in SC2 to get decent results. If you just react to your opponent and have perfect, bot-like control, you will win. | ||
|
Grumbels
Netherlands7031 Posts
Also, I heard a pro player say that an engine such as Leela would be less useful than Stockfish in preparing, because the latter is tactically superior, while the former is strategically superior. But humans are already good at strategy, they just need to make use of their tactical ability of engines to check their ideas and openings for tactical flaws. Leela's is unreliable because it doesn't have concrete reasons for preferring one move over the other, just a strategical intuition. Whereas Stockfish can instantly tell you if there are tactical problems with a move and produce a refutation. It might be the case that AlphaZero will remain a novelty for computer chess enthusiasts. Especially since Leela and AlphaZero run on TPU/GPU's, not CPU's, afaik, so if you want to use both locally, you have to invest in both a good graphic card and a good processor. | ||
|
alexanderzero
United States659 Posts
I think the tree-search method for chess is bound to scale better with hardware than a neural network approach, given that chess is theoretically solvable with tree search. This would suggest otherwise: ![[image loading]](https://storage.googleapis.com/deepmind-live-cms/images/AZ-Blog-Fig2-Search-Per-Decision.width-1500.png) Isn't go also theoretically solvable with a search tree? | ||
|
xongnox
540 Posts
Out-microing and out-multitasking everyone by playing 30.000 APMs and 100 screens/second is surely automaton-2000 impressive to watch one or two time, but is not very conclusive for the intelligence part. I guess they done it right and have set limiting factors as parameters (like 250/apms max, max actions per second, max screens per second, human-like time mouse movements, etc, etc. ) | ||
|
Grumbels
Netherlands7031 Posts
On January 23 2019 18:07 alexanderzero wrote: This would suggest otherwise: Isn't go also theoretically solvable with a search tree? AlphaZero claimed that their approach scaled well, iirc they had extremely good hardware for the recent rematch. On the other hand, there is some reason to doubt their work, since they might be more familiar setting up their own engine versus setting up Stockfish. Leela seems to do worse than Stockfish on good hardware / longer time controls. But I'm not sure, since e.g. people complain about hardware set-ups for computer chess tournaments all the time, since now it's the case that you have a prominent engine that requires a different set-up. There are also different ways of comparing hardware, e.g. price or energy consumption. Go is comparable to chess, but is has significantly more possibilities per move than chess. There existed engines using the chess-like tree search for Go, but they were pretty bad because they get lost in all the variations. The neural network approach works much better there. Chess is interesting since both approaches seem fairly equal, so you can investigate scaling more meaningfully. And there's no real point in comparing humans to AlphaZero, since humans are much worse. edit: just a point about terminology, it's misleading to say that Stockfish uses tree search while AlphaZero uses neural networks. Because AlphaZero also uses (MC) tree search and Stockfish uses an evaluation function with weights tuned with machine learning tools. Given the obvious weaknesses that Leela possesses (and presumably AlphaZero too), the future best chess engine is probably somewhere in the middle between current SF and AZ. | ||
|
gpanda.sc2
20 Posts
On January 23 2019 12:50 imCHIEN wrote: vs $O$ to see how AI deals with cheese vs Maru to see how AI deals with his creative vs Serral to see how AI deals with a strong late game opponent. vs TY to see all the above at one time. | ||
|
DreamOen
Spain1400 Posts
But making it look like human problem solving and winning due to strategy and not insane sharp micro/multitask would be a really different thing. | ||
|
neutralrobot
Australia1025 Posts
On January 23 2019 11:53 KalWarkov wrote: until alpha zero beats stockfish in TCEC finals, i will never call alpha zero the strongest engine. everything is controlled by google. no table base, no opening books - which sf isnt trained for. and still, it isn't live games vs sf11dev. and who knows if they released all games or are just cherry picking? Well, I mean, it's always possible that they're presenting some kind of falsehood about the 100 game match vs Stockfish recently where Alpha Zero took no losses, but... why? Why would they flatly lie about the results of that match? Honestly I don't think they even care much about proving themselves in the domain of chess -- it was just part of a proof of concept about generalizing the AlphaGo algorithm to be applicable to other games. What do they gain by lying about this? Like, if you want to say that there should be a public tournament with different conditions before it's definitive, I can respect that, but the cherry picking idea seems pretty far-fetched to me, particularly considering the growth of Leela this year. On January 23 2019 17:40 Grumbels wrote: AlphaZero becoming the strongest engine in a matter of hours is a bit deceiving, given that it still required fifty million games of practice and computing a new version of the network every 25k games. It was estimated to take months for the Leela project (open source imitation of AZ), which is distributed on hundreds of computers. Google just has really powerful hardware. There are some interesting quirks with Leela. For instance, it's not capable of playing endgames efficiently, it seemingly aimlessly moves around, making moves that don't lose the advantage. It doesn't "get to the point". If an SC2 AI is built on the same concept, expect it to not be able to finish off games quickly and take an hour to mine out the entire map and build a fleet of random units to randomly move around the map. Another quirk of the project is that the algorithm uses not just the current move as input, but also the history of moves. This gives it some measure of what part of the board to pay "attention" to. It also means that if you give it a random position as input, without history, that it can't function. As far as I know Leela is useless in solving tactical puzzles and in handicap games without training it first. Leela also typically doesn't understand theory of endgames. It doesn't just play them weirdly, but it also doesn't grasp some almost mathematical ideas such as identifying a class of endgames that are drawn despite material imbalances (opposite color bishops, wrong color bishop). It's apparently also not better at fortress positions, where you have material disadvantages, but your position can't be cracked. There are some known positions like these, and it was hoped that neural networks would be better at them, and would be capable of reasoning that these are a special class of positions that require a different approach. But it doesn't really seem like it. Leela is also probably already better than Stockfish if you have bad hardware and no opening book. You can imagine that if there was a market for SC2 bots, that they could have opening books updated for every patch and which would have a team of people dedicated to keeping track of the meta and adding knowledge of it to the bot. But Deepmind's AI would use self-learning, i.e. only playing itself and developing its own meta. I don't know if that would make it easier or harder to beat as a human. I think the tree-search method for chess is bound to scale better with hardware than a neural network approach, given that chess is theoretically solvable with tree search. But this method would be useless for SC2, unless the AI uses some sort of abstraction of strategy and tries to think ahead. But I don't think you really need to think ahead in SC2 to get decent results. If you just react to your opponent and have perfect, bot-like control, you will win. Yeah, there are some quirks about Leela's play like the ones you mentioned. It's kinda hilarious watching Leela take forever to mate with Queen and King vs King, for example. But in most contexts, when both engines agree that the game is completely decided, they call it. Maybe Fantasy would make a new AI play for 2+ hours under totally lost conditions, but hopefully there would be a gg called before then in most cases. The talk of openings and the translation to SC2 is interesting to think about. AlphaZero seemed to keep going back to a relatively small handful of openings (I seem to remember it kept using the Berlin defense?) when left to its own devices as opposed to starting from a book position. But SC2 openings seem like they have to account for a lot more variables. Would a deep RL algorithm for SC2 play differently when optimizing for series vs single maps? Would it develop opening strategies that are more or less water-tight no matter what the context? Also, Would it show some of AlphaZero/Leela's brilliance for understanding positional compensation and imbalanced material? I guess we might find out about all this stuff soon. | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
ZigguratOfUr
Iraq16955 Posts
On January 23 2019 17:17 Lazzarus wrote: So this is another AI playing SCII? https://twitter.com/ENCE_Serral/status/1087742590357774336 Yes, but those are 'regular' AIs coded up by someone (and with 100k APM for crazy micro tricks). | ||
|
Ronski
Finland266 Posts
On January 24 2019 00:57 neutralrobot wrote: Well, I mean, it's always possible that they're presenting some kind of falsehood about the 100 game match vs Stockfish recently where Alpha Zero took no losses, but... why? Why would they flatly lie about the results of that match? Honestly I don't think they even care much about proving themselves in the domain of chess -- it was just part of a proof of concept about generalizing the AlphaGo algorithm to be applicable to other games. What do they gain by lying about this? Like, if you want to say that there should be a public tournament with different conditions before it's definitive, I can respect that, but the cherry picking idea seems pretty far-fetched to me, particularly considering the growth of Leela this year. Yeah, there are some quirks about Leela's play like the ones you mentioned. It's kinda hilarious watching Leela take forever to mate with Queen and King vs King, for example. But in most contexts, when both engines agree that the game is completely decided, they call it. Maybe Fantasy would make a new AI play for 2+ hours under totally lost conditions, but hopefully there would be a gg called before then in most cases. The talk of openings and the translation to SC2 is interesting to think about. AlphaZero seemed to keep going back to a relatively small handful of openings (I seem to remember it kept using the Berlin defense?) when left to its own devices as opposed to starting from a book position. But SC2 openings seem like they have to account for a lot more variables. Would a deep RL algorithm for SC2 play differently when optimizing for series vs single maps? Would it develop opening strategies that are more or less water-tight no matter what the context? Also, Would it show some of AlphaZero/Leela's brilliance for understanding positional compensation and imbalanced material? I guess we might find out about all this stuff soon. The latest match where Stockfish and AlphaZero played 1000 games Stockfish was using its opening books and did manage to win a decent amount of games with white pieces. Alphazero still won the match overall but Stockfish did take games on a somewhat consistent rate. | ||
|
waiting2Bbanned
United States154 Posts
On January 24 2019 00:31 gpanda.sc2 wrote: vs TY to see all the above at one time. TY's cheese is repetitive and boring. sOs' is not bad, but neither can hold a candle to Has' dairy farm. | ||
|
Zreg
9 Posts
I wouldnt want to play against it! | ||
|
Rodya
546 Posts
| ||
|
Cyro
United Kingdom20285 Posts
On January 24 2019 02:18 Rodya wrote: Is there something about neural nets that make this interesting? I mean wont we just see insane tank dropship abuse? Even with deep learning computers are really bad at some stuff and really good at other stuff, it would be amazing to see one able to take on a pro in a variety of situations | ||
|
zealotstim
United States455 Posts
| ||
|
Grumbels
Netherlands7031 Posts
On January 24 2019 02:18 Rodya wrote: afaik there is no information available on what type of type of progress they have been making and what the tendencies of the AI will be. I scanned their website last year to see if it had any papers or releases which obviously involved SC2 (I don’t know that much about AI though) and couldn’t find it. We will have to wait and see, but probably it won’t be able to play a game competently. Recall that in the demo they revealed a year ago their AI couldn’t even build units or move them around the map, it would click on the minimap to an empty location and then get stuck trying to get back.Is there something about neural nets that make this interesting? I mean wont we just see insane tank dropship abuse? edit: apparently more information has come out, supposedly it can beat the SC2 insane AI 50% of the time https://www.pcgamer.com/blizzard-will-show-off-googles-deepmind-ai-in-starcraft-2-later-this-week/ | ||
|
Grumbels
Netherlands7031 Posts
On January 24 2019 02:31 zealotstim wrote: This should be interesting. I hope if it gets to the point of playing against a person, it has limited apm. Everyone knows a computer can micro better than a person; the question is whether it can outsmart one with a clever strategy or smarter execution. They released a paper which explains they have limited the APM to 180 APM. Note this is still super human because an AI can potentially be very efficient in its actions. | ||
|
Acrofales
Spain17982 Posts
On January 23 2019 02:21 travis wrote: What they have done is isolate "mini games" from within sc2 - tasks like mineral mining or base organization. So it may be the case that they are now performing well on such minigames and want to show that off and talk about how they've done it. Was about to PM you to tell you you might want to check that out, but I see you're way ahead of me! | ||
|
ScarPe
Germany392 Posts
but remember, once they got the environment the AI will learn by playing 1000 or more games at once. it will overcome human with ease then. even if it learns only very little with every game. the pure mass will do it. sadly most people think this game would be too hard to learn for an ai, but it will destroy every progamer in the end. | ||
|
PanS3rnik
18 Posts
![[image loading]](https://goo.gl/images/RDsYFm) | ||
|
[F_]aths
Germany3947 Posts
Announcement that a Deepmind version gets implemented in the SC2 client so one can train versus that AI, and then let differently trained AIs play versus each other? Or replay analysis of human 1v1 by the AI? Perhaps not today. Still, the possibilities are endless ... if they have an AI able to beat a human. What if Blizzard creates a GSL seed for the AI? | ||
|
_fool
Netherlands677 Posts
| ||
|
outscar
2832 Posts
FlaSh Inno EffOrt | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
JimmyJRaynor
Canada16699 Posts
On January 24 2019 05:18 [F_]aths wrote:Still, the possibilities are endless ... if they have an AI able to beat a human. What if Blizzard creates a GSL seed for the AI? Fire Pro Wrestling had an entire scene devoted to AI versus AI action. AIs handcrafted by humans fighting each other. If Blizzard can create a competitive scene out of various computer science teams building their own AI-bots.. that'd be incredible. | ||
|
neutralrobot
Australia1025 Posts
On January 24 2019 01:26 Ronski wrote: The latest match where Stockfish and AlphaZero played 1000 games Stockfish was using its opening books and did manage to win a decent amount of games with white pieces. Alphazero still won the match overall but Stockfish did take games on a somewhat consistent rate. Yeah, true. It seems there were 12 matches of 100 games each = 1200 games total. AlphaZero won 290 and lost 24. Of those losses, I remember watching a few games that were started with book openings that seem to have been disadvantageous. | ||
|
UncleVinny
United States35 Posts
| ||
|
starithm
United States118 Posts
On January 24 2019 12:03 UncleVinny wrote: I am hunnert percent “working from home” tomorrow. Twitch.tv on personal laptop, conference webinar on work laptop.  | ||
|
Parrek
United States893 Posts
On January 24 2019 03:22 ScarPe wrote: the problems they had in the beginning was, that it was hard to set up a proper learnming environment. biggest problem was building placement imho. but remember, once they got the environment the AI will learn by playing 1000 or more games at once. it will overcome human with ease then. even if it learns only very little with every game. the pure mass will do it. sadly most people think this game would be too hard to learn for an ai, but it will destroy every progamer in the end. Yes, but there are a lot of intricacies in high level starcraft. Brute force and micro won't win. AI can't do cleverness or trickery. Even Chess which has many orders of magnitude less possible moves than SC2 isn't really viable to brute force. I think you're giving too much credit to current neural nets | ||
|
Popkiller
3415 Posts
On January 24 2019 14:57 Parrek wrote: Yes, but there are a lot of intricacies in high level starcraft. Brute force and micro won't win. AI can't do cleverness or trickery. Even Chess which has many orders of magnitude less possible moves than SC2 isn't really viable to brute force. I think you're giving too much credit to current neural nets playing thousands of games really quickly and learning isn't the same as winning a game with brute force. AlphaGo didn't just brute force calculate every possible Go move, it used strategies. But, it learned by playing numerous games against itself and other versions of itself. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 24 2019 14:57 Parrek wrote: Yes, but there are a lot of intricacies in high level starcraft. Brute force and micro won't win. AI can't do cleverness or trickery. Even Chess which has many orders of magnitude less possible moves than SC2 isn't really viable to brute force. I think you're giving too much credit to current neural nets Neural networks don't brute force things. The only question is how good the learning was. | ||
|
Acrofales
Spain17982 Posts
On January 24 2019 16:50 deacon.frost wrote: Neural networks don't brute force things. The only question is how good the learning was. Mind tricks work in StarCraft, because the game is only partially observable. That is a very important difference and it remains to be seen how well deepmind works in partially observable games. | ||
|
Grumbels
Netherlands7031 Posts
| ||
|
-Archangel-
Croatia7457 Posts
On January 24 2019 02:34 Grumbels wrote: afaik there is no information available on what type of type of progress they have been making and what the tendencies of the AI will be. I scanned their website last year to see if it had any papers or releases which obviously involved SC2 (I don’t know that much about AI though) and couldn’t find it. We will have to wait and see, but probably it won’t be able to play a game competently. Recall that in the demo they revealed a year ago their AI couldn’t even build units or move them around the map, it would click on the minimap to an empty location and then get stuck trying to get back. edit: apparently more information has come out, supposedly it can beat the SC2 insane AI 50% of the time https://www.pcgamer.com/blizzard-will-show-off-googles-deepmind-ai-in-starcraft-2-later-this-week/ From the comment section of that article: Tetsuo Wow nice to finally see some results lets see if any A.I can beat Serral ^^ Prrredictable Negative. The Koreans deployed eight humanoid AIs to Blizzcon last year who were all met with a Serral victory. ggwp | ||
|
True_Spike
Poland3422 Posts
| ||
|
Hider
Denmark9376 Posts
On January 24 2019 02:36 Grumbels wrote: They released a paper which explains they have limited the APM to 180 APM. Note this is still super human because an AI can potentially be very efficient in its actions. link to paper? | ||
|
Grumbels
Netherlands7031 Posts
https://deepmind.com/documents/110/sc2le.pdf On January 24 2019 18:08 -Archangel- wrote: From the comment section of that article: these jokes are kinda racist imo, I don't think it's that funny. in general you shouldn't read the comments on these sites, since either people feel the need to make inside jokes, or they will start to talk about skynet. I think that seeing similarities between a SC2 AI and military technology is a bit far-fetched, since SC2 is not too different from dota, it just has military themes for the visuals. it reminds me of when I researched mental imaging at university and people kept bringing up mind reading and sci-fi plots. people process new technology not through thinking about what it does and is capable of, but via comparing it to pop culture... | ||
|
Loccstana
United States833 Posts
| ||
|
LoneYoShi
France1348 Posts
On January 24 2019 20:21 Loccstana wrote: Anyone excited that with good AI, blizzard can finally test their balance changes properly before pushing them out? I'm more excited about the potential effects this can have on the metagame ! I expect good progress from Deepmind's team compared to what has been shown at Blizzcon, but no pro level yet. I'm banking on an AI that's diamond/low-master level, but the comparison with human players will probably not be easy since it won't be playing like a human player (probably better macro but questionnable/surprising decision making ?) Let's see how far from the truth I'll be ! Damn, this whole thing is really hyping me up ! | ||
|
Grumbels
Netherlands7031 Posts
On January 24 2019 21:56 LoneYoShi wrote: I'm more excited about the potential effects this can have on the metagame ! I expect good progress from Deepmind's team compared to what has been shown at Blizzcon, but no pro level yet. I'm banking on an AI that's diamond/low-master level, but the comparison with human players will probably not be easy since it won't be playing like a human player (probably better macro but questionnable/surprising decision making ?) Let's see how far from the truth I'll be ! Damn, this whole thing is really hyping me up ! If it's possible to develop an AI that can approximate the human approach to playing Starcraft, mimicking human physical ability, but at an accelerated level such that it can reach a very sophisticated level of strategical understanding in hours, then what would have happened if the community would have had access to this AI in 2010? I think it would turn the competitive scene into something of a joke, because these players would so obviously be inferior to the AI, and because the most successful players would just be copying the AI's strategy. We would have just been waiting for this whole phase of the game's life cycle to end, until players would finally be capable enough to achieve some degree of parity with the AI. And suppose that only Blizzard can use the AI. It's not so easy to generate useful data with it, because any such thing as development of the meta game is bound to be arbitrary and dependent on whatever breakthroughs made by pro gamers and such, that will then trickle down to the rest of the player base. You can not be certain that the AI will take the same path towards advancing the meta, and you can't be certain that the end point for the AI's development is anywhere close to what humans would be capable of reaching on their own. So even if you aim for balancing the game in such a way that it is balanced at this end point, this can still end up as a failure. Unless you basically force the players to play in a certain way by giving them access to the AI and constantly showcasing certain strategies. But then you remove the human element from the game, kind of how chess nowadays involves memorization of computer lines. I do actually like the latter idea though. All such questions of simulations feel very similar to time travel concepts. It's as if Blizzard had access to games from 2012 in 2011. Surely they would have been more cautious with buffing zerg around that time, and then we wouldn't have the BL/infestor era. And even if the community had demanded improvements to the queen and infestor, Blizzard then would have abstained as they had access to an oracle telling them zerg players could learn to defend better. And potentially it would allow Blizzard to bypass the sort of problems discussed in this thread, where balance approaches in BW and SC2 are discussed. | ||
|
Grumbels
Netherlands7031 Posts
The name of the AI is AlphaStar, so either AS or A*. By the way, A* is a well-known AI algorithm | ||
|
Musicus
Germany23576 Posts
No way in hell it can be on that level already though right? | ||
|
MaryJoana
Germany156 Posts
And to add on to this. Saying that it's impossible that AlphaStar has improved in 3 months time to be able to at least give a pro gamer something to work for shows that there is a lack of understanding of the speed at which these things can improve. OpenAI went through an absurd increase in competence in a matter of a 3 week span. This is a 3 month span and maybe even more, because the ability of the AI at BlizzCon might have been a week or two old. | ||
|
MaryJoana
Germany156 Posts
On January 25 2019 00:36 Grumbels wrote: If it's possible to develop an AI that can approximate the human approach to playing Starcraft, mimicking human physical ability, but at an accelerated level such that it can reach a very sophisticated level of strategical understanding in hours, then what would have happened if the community would have had access to this AI in 2010? I think it would turn the competitive scene into something of a joke, because these players would so obviously be inferior to the AI, and because the most successful players would just be copying the AI's strategy. We would have just been waiting for this whole phase of the game's life cycle to end, until players would finally be capable enough to achieve some degree of parity with the AI. Look at what happened in Dota 2. Human players got much better at playing 1v1 mid as a direct result of playing versus the AI made by OpenAI that beat all of them handidly. Having an opponent that is better than you increases the playing field. If you take Serral and put him back in time in 2010 and replay how the sc2 competitive scene plays out, you would see the derivative of what Maru or Byun perform better, because there was better things to learn from. This is not detrimental to the scene. It's good for the scene. We watch humans play the game not because they perform better than a machine, but because they perform better than any other human. Chess is not any less competitive since introduction of computers but it has increased the level of competition. | ||
|
Grumbels
Netherlands7031 Posts
On January 25 2019 02:47 MaryJoana wrote: Look at what happened in Dota 2. Human players got much better at playing 1v1 mid as a direct result of playing versus the AI made by OpenAI that beat all of them handidly. Having an opponent that is better than you increases the playing field. If you take Serral and put him back in time in 2010 and replay how the sc2 competitive scene plays out, you would see the derivative of what Maru or Byun perform better, because there was better things to learn from. This is not detrimental to the scene. It's good for the scene. We watch humans play the game not because they perform better than a machine, but because they perform better than any other human. Chess is not any less competitive since introduction of computers but it has increased the level of competition. Well, the idea is that it should be humans who figure out how to play. They shouldn't have to be taught how to play by an AI. I wouldn't mind if an AI was introduced when the meta was becoming stale, but if it already showed perfect gameplay in 2010 that would be a joke. I kinda don't think we should want to travel back in time and see Maru or Serral destroy a 2010 field and laugh at how bad Mvp is. I think it would nullify their accomplishments. | ||
|
ASoo
2864 Posts
On January 25 2019 02:06 Musicus wrote: Seems like AlphaStar will face TLO? No way in hell it can be on that level already though right? This feels a lot like OpenDota against Veggies Esports... | ||
|
MaryJoana
Germany156 Posts
On January 25 2019 02:52 Grumbels wrote: Well, the idea is that it should be humans who figure out how to play. They shouldn't have to be taught how to play by an AI. I wouldn't mind if an AI was introduced when the meta was becoming stale, but if it already showed perfect gameplay in 2010 that would be a joke. I kinda don't think we should want to travel back in time and see Maru or Serral destroy a 2010 field and laugh at how bad Mvp is. I think it would nullify their accomplishments. You're misunderstanding me. Of course the focus is not on watching Mvp being destroyed by Serral. It's about the process that evolves from it. It's essentially the same thing as when someone like Serral just came around in 2018. Nobody has figured out how to beat him. But everyone who plays him gets to improve. This is the exact same. | ||
|
MaryJoana
Germany156 Posts
On January 25 2019 02:54 ASoo wrote: This feels a lot like OpenDota against Veggies Esports... Exactly the reason why I think people shouldn't be so dismissive and unexcited. It's been shown that an AI can play at levels that were unthinkable only a few years ago. | ||
|
GreasedUpDeafGuy
United States398 Posts
On January 24 2019 18:35 Grumbels wrote: https://deepmind.com/documents/110/sc2le.pdf these jokes are kinda racist imo, I don't think it's that funny. in general you shouldn't read the comments on these sites, since either people feel the need to make inside jokes, or they will start to talk about skynet. I think that seeing similarities between a SC2 AI and military technology is a bit far-fetched, since SC2 is not too different from dota, it just has military themes for the visuals. it reminds me of when I researched mental imaging at university and people kept bringing up mind reading and sci-fi plots. people process new technology not through thinking about what it does and is capable of, but via comparing it to pop culture... Is it difficult living a life where jokes trigger you? | ||
|
Grumbels
Netherlands7031 Posts
On January 25 2019 02:55 MaryJoana wrote: You're misunderstanding me. Of course the focus is not on watching Mvp being destroyed by Serral. It's about the process that evolves from it. It's essentially the same thing as when someone like Serral just came around in 2018. Nobody has figured out how to beat him. But everyone who plays him gets to improve. This is the exact same. I think there's a difference between generating scientific and cultural data. It's great to have an AI to come up with new cancer treatments, and nobody will complain that this should be left to humans to explore. But SC2 is a game, shouldn't it be up to humans to discover how it can be played? Chess engines are very controversial, of course, but they only came out once the game was hundreds of years old. and they could show new strategies and ways of playing. But Starcraft in 2010 was still in its infancy and there was plenty left for humans to explore. | ||
|
Grumbels
Netherlands7031 Posts
On January 25 2019 03:00 GreasedUpDeafGuy wrote: Is it difficult living a life where jokes trigger you? pls go live in twitch chat | ||
|
PanS3rnik
18 Posts
| ||
|
HsDLTitich
Italy830 Posts
| ||
|
Loccstana
United States833 Posts
Also according to Aligulac, if Serral plays TLO in Bo7, Serral will win 99.41% of the time. | ||
|
Aeromi
France14459 Posts
| ||
|
imre
France9263 Posts
| ||
|
Patton3D
United States65 Posts
| ||
|
MajuGarzett
Canada635 Posts
On January 25 2019 03:13 imre wrote: Did they announced any restrictions regarding the IA mechanical skills? It's limited to 180 apm, they said they'll talk about it more later. | ||
|
RandomPlayer
Russian Federation390 Posts
| ||
|
Vanillatoss
76 Posts
On January 25 2019 03:13 imre wrote: Did they announced any restrictions regarding the IA mechanical skills? yes... apm <160 or smt like that | ||
|
Loccstana
United States833 Posts
On January 25 2019 03:15 RandomPlayer wrote: Why did they pick a player that doesn't win games against other pros... I mean tournaments. Still, he's GM so should be exciting. Give some respect to TLO, he has been playing since 2010. Also, people's reaction time gets slower as they age. | ||
|
Brutaxilos
United States2628 Posts
| ||
|
Majick
416 Posts
| ||
|
Shathe
Hungary422 Posts
| ||
|
Nerchio
Poland2633 Posts
| ||
|
Aeromi
France14459 Posts
On January 25 2019 03:21 Nerchio wrote: I'd say it's around low master level at best Which is quite impressive I think | ||
Pandemona
Charlie Sheens House51484 Posts
| ||
|
ASoo
2864 Posts
| ||
|
darklycid
3485 Posts
| ||
|
Kazi25
Philippines236 Posts
| ||
|
raff100
498 Posts
| ||
|
Loccstana
United States833 Posts
On January 25 2019 03:21 Nerchio wrote: I'd say it's around low master level at best I would say its Avilo level without all the salt and accusations.  | ||
|
Majick
416 Posts
Exactly. Especially if we take into account this is a machine so it can be 100% consistent all the time and is not affected by emotion. | ||
|
Shathe
Hungary422 Posts
| ||
|
Vanillatoss
76 Posts
 | ||
|
Pandemona
Charlie Sheens House51484 Posts
| ||
|
Garrl
Scotland1972 Posts
so it's not fair? | ||
|
Brutaxilos
United States2628 Posts
| ||
|
imre
France9263 Posts
On January 25 2019 03:21 Nerchio wrote: I'd say it's around low master level at best Still pretty impressive and let's be real, if it doesn't build more than 35 probes it'll be ranked higher | ||
|
Shathe
Hungary422 Posts
| ||
|
Haukinger
Germany131 Posts
| ||
|
Poopi
France12836 Posts
| ||
|
Aeromi
France14459 Posts
On January 25 2019 03:28 Shathe wrote: Why only 2 maps? :'( Other pros are going to be in the panel I think | ||
|
ASoo
2864 Posts
On January 25 2019 03:29 Poopi wrote: What did I miss? 1-base stalkers > 2-base stalkers. | ||
|
Loccstana
United States833 Posts
| ||
|
Poopi
France12836 Posts
No I mean what is the AI allowed to do, stuff like that | ||
|
Grumbels
Netherlands7031 Posts
| ||
|
imre
France9263 Posts
On January 25 2019 03:30 Poopi wrote: No I mean what is the AI allowed to do, stuff like that it plays the full game in PvP, limited APM (160 or 180) and has a ful map view. | ||
|
Tchado
Jordan1831 Posts
What happened so far ? and whats going on ? | ||
|
Zephyp
238 Posts
| ||
|
Poopi
France12836 Posts
On January 25 2019 03:31 Zephyp wrote: Isn't it a bit odd using a non-Protoss to play as Protoss in this event? Or are they worried a good player would win? Yeah it's staged | ||
|
Musicus
Germany23576 Posts
| ||
|
ASoo
2864 Posts
On January 25 2019 03:31 Tchado wrote: Can someone tell me what happened so far ? I just tuned in , Alpha star vs tlo , 2-0. What happened so far ? and whats going on ? AI won a stalker vs. stalker mirror match with some decent micro and some questionable warp-ins from TLO. We didn't see the second map. | ||
|
Nakajin
Canada8989 Posts
| ||
|
Kantuva
Uruguay206 Posts
| ||
|
Brutaxilos
United States2628 Posts
| ||
|
Poopi
France12836 Posts
Is it capped at 180apm on average, but can go higher? Plus full map view of non fog of war map, so basically can defend harass pretty well? | ||
|
darklycid
3485 Posts
| ||
|
stilt
France2749 Posts
Alphastar is impressive for sure but it's not very interesting to see. | ||
|
ASoo
2864 Posts
Why don't they just play actual pros in the first place? Is it that hard to schedule something? | ||
|
Doink
75 Posts
They know people who don't know SC2 will believe it. | ||
|
Grumbels
Netherlands7031 Posts
On January 25 2019 03:37 Poopi wrote: Why was it at 500apm? Is it capped at 180apm on average, but can go higher? Plus full map view of non fog of war map, so basically can defend harass pretty well? I think the 180 apm cap is just from an old version, and that's not true anymore. Probably they'll release another paper this year and we can see the details of the current apm set-up. | ||
|
Poopi
France12836 Posts
This and playing against an offrace pro... I guess DeepMind is all about PR nowadays | ||
|
Nerchio
Poland2633 Posts
On January 25 2019 03:40 Poopi wrote: It's so boring, why is it PvP and not at least a more interesting mirror with clear indications of how the AI works? This and playing against an offrace pro... I guess DeepMind is all about PR nowadays News about DeepMind winning will be all over the internet so I guess that's good but for us it's disappointing | ||
|
darklycid
3485 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
darklycid
3485 Posts
On January 25 2019 03:44 Nakajin wrote: Do we know what TLO protoss rank is? It look like mid-master honestly It looks like low master to diamond i think. | ||
|
mechzdeus
88 Posts
| ||
|
TBO
Germany1350 Posts
On January 25 2019 03:44 Nakajin wrote: Do we know what TLO protoss rank is? It look like mid-master honestly 5500ish on NA. But playing vs something you don't know and never played vs before makes you easily look silly. | ||
|
darklycid
3485 Posts
He played way worse if that is true. | ||
|
imre
France9263 Posts
On January 25 2019 03:38 ASoo wrote: I kinda wish they'd get good pros for these events, instead of TLO offracing or that caster team that obviously didn't practice for OpenDota. Like, you know the AI is going to win, and then you know everybody is going to say it doesn't count because they were up against mediocre semi-pros. Why don't they just play actual pros in the first place? Is it that hard to schedule something? OpenAI on dota played real team at TI, it was competitive for 10 to 15minutes then unable to deal with splitpush and finally humiliated by a more farmed human team. Pretty cool to watch. | ||
|
Kazi25
Philippines236 Posts
| ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 03:45 darklycid wrote: It looks like low master to diamond i think. dont indulge yourselves, tlo would murder low masters | ||
|
mierin
United States4943 Posts
| ||
|
FFW_Rude
France10201 Posts
| ||
|
Poopi
France12836 Posts
Well at least we'll have interesting papers from an AI point of view | ||
|
Vanillatoss
76 Posts
On January 25 2019 03:37 Poopi wrote: Why was it at 500apm? Is it capped at 180apm on average, but can go higher? Plus full map view of non fog of war map, so basically can defend harass pretty well? The fog of war is active... its just Alfa Star view is zoom out but at the same time it is limited to 30 screens per minute.. | ||
|
Paljas
Germany6926 Posts
On January 25 2019 03:48 Poopi wrote: Why the heck is it blizzcon patch btw? Why did they wait that much time before the demo? Well at least we'll have interesting papers from an AI point of view Trained the agent on that patch would be my guess. | ||
|
Aeromi
France14459 Posts
On January 25 2019 03:47 mierin wrote: Doesn't alphago play sub-optimally when faced with suboptimal opponents? It may be that it's not as bad as it looks, it's just this particular playstyle is "enough" for it to confidently win? No, AlphaGo was playing to win and it did not care on how it win, in the documentary it shows that it won by 2 points by doing quite a weird move in move 37 or 73 | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 03:46 TBO wrote: 5500ish on NA. But playing vs something you don't know and never played vs before makes you easily look silly. I guess but isn't pheonix the super standard hard counter to disruptor, they came super late. And many attack seemed odd and he miss a LOT of forcefield on the ramp in front of the natural. | ||
|
Xamo
Spain877 Posts
The full map vision is a huge difference though. | ||
|
Acrofales
Spain17982 Posts
On January 25 2019 03:26 Pandemona wrote: Well thats a bit of a cheat isnt it? Being able to see whole map all time (non fog of war part) Well, that removes the "partial observable" part... guess they couldn't deal with that | ||
|
IshinShishi
Japan6156 Posts
| ||
|
Brutaxilos
United States2628 Posts
| ||
|
Haukinger
Germany131 Posts
| ||
|
Nakajin
Canada8989 Posts
We need Nestea back guys it's our only hope | ||
|
imre
France9263 Posts
On January 25 2019 03:51 IshinShishi wrote: Im pretty sure that it's playing at GM level not at all | ||
|
cha0
Canada504 Posts
| ||
|
Tchado
Jordan1831 Posts
Mankind is doomed  | ||
|
Paljas
Germany6926 Posts
On January 25 2019 03:49 Aeromi wrote: No, AlphaGo was playing to win and it did not care on how it win, in the documentary it shows that it won by 2 points by doing quite a weird move in move 37 or 73 AlphaGo plays to maximize winning percantage, not point difference. But this shows normally only rather late in the game (the move 37 in the second game is famous for being very creative and good, not because it is suboptimal). | ||
|
Musicus
Germany23576 Posts
| ||
|
Garrl
Scotland1972 Posts
On January 25 2019 03:51 IshinShishi wrote: Im pretty sure that it's playing at GM level they would've put a protoss player against it if it could beat them. | ||
|
I wasbanned fromthis
113 Posts
On January 23 2019 03:55 paralleluniverse wrote: I don't expect its ability to be more than a high diamond player, or at best a high masters player. And I'd bet it will use Terran, because its skill with the other 2 races is bad. Still betting? | ||
|
FFW_Rude
France10201 Posts
| ||
|
Poopi
France12836 Posts
That APM thing is questionable as well | ||
|
Acrofales
Spain17982 Posts
On January 25 2019 03:50 Acrofales wrote: Well, that removes the "partial observable" part... guess they couldn't deal with that Hm. Not sure now. TLO just said that proxy could have worked because the AI doesn't scout well... so they *don't* see the whole map? | ||
|
StasisField
United States1086 Posts
| ||
Waxangel
United States33371 Posts
| ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 03:53 Garrl wrote: they would've put a protoss player against it if it could beat them. you guys seriously overestimate the level of play of GM players, I dont think it could beat a pro, but low-mid GM? For sure. | ||
|
imre
France9263 Posts
On January 25 2019 03:55 Poopi wrote: Still if this is the result we get when the AI can view full map (beating an offracing TLO), it'll be a long time before they can beat pros in fair conditions. That APM thing is questionable as well From what we saw the APM looked a bit below the standard pro player (especially game 1 a Korean Protoss would have had a way better prism micro) so this point isn't an issue. | ||
|
Waxangel
United States33371 Posts
On January 25 2019 03:55 Acrofales wrote: Hm. Not sure now. TLO just said that proxy could have worked because the AI doesn't scout well... so they *don't* see the whole map? the guy worded it really poorly; it doesn't have maphack, more like a "zoom-out-hack" (not EXACTLY this either) | ||
|
Xamo
Spain877 Posts
| ||
|
Dav1oN
Ukraine3164 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
Pandain
United States12989 Posts
| ||
|
Gandie
158 Posts
| ||
|
Kazi25
Philippines236 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Howard_Kao
China261 Posts
| ||
|
I wasbanned fromthis
113 Posts
| ||
|
Serimek
France2274 Posts
| ||
|
TheDougler
Canada8304 Posts
| ||
|
imre
France9263 Posts
| ||
|
Aeromi
France14459 Posts
| ||
|
JustPassingBy
10776 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 03:57 Serimek wrote: LiquidMana top 10 Protoss? The guy is good, but it's PR bullshit. Hum seems fair honestly, I would take top 15 toss | ||
|
Poopi
France12836 Posts
On January 25 2019 03:56 imre wrote: From what we saw the APM looked a bit below the standard pro player (especially game 1 a Korean Protoss would have had a way better prism micro) so this point isn't an issue. It's not an issue now because they couldn't yet optimize this part as well, but as they progress it'll become a problem | ||
|
gophersnake
48 Posts
| ||
|
Cyro
United Kingdom20285 Posts
On January 25 2019 03:56 Dav1oN wrote: APM thing is questionable, but even more questionable is a "maphack", it needs to have the same conditions and vision as a human to make it close to fair It doesn't have maphack, it can just technically see everywhere that it has vision | ||
|
Acrofales
Spain17982 Posts
On January 25 2019 03:56 Waxangel wrote: the guy worded it really poorly; it doesn't have maphack, more like a "zoom-out-hack" (not EXACTLY this either) Seeing as clicking around the map costs 1 APM, I'm not sure a "zoom-out-hack" is actually a hack at all. So... am I understanding this right? Did it also play Mana? This should be good :O | ||
|
Loccstana
United States833 Posts
| ||
|
Xamo
Spain877 Posts
| ||
|
TheDougler
Canada8304 Posts
On January 25 2019 03:57 Serimek wrote: LiquidMana top 10 Protoss? The guy is good, but it's PR bullshit. Top 10 foreigner protoss maybe. I mean, it's basically Neeb, Showtime, and then everyone else as far as non-Korean protoss goes. | ||
|
Majick
416 Posts
| ||
|
Loccstana
United States833 Posts
On January 25 2019 03:58 Nakajin wrote: Hum seems fair honestly, I would take top 15 toss Check Liquid Mana's ELO on aligulac | ||
|
Serimek
France2274 Posts
On January 25 2019 03:58 imre wrote: being as fast as an human with full vision basically means you can't harass it or i'm totally wrong? you wouldn't be able to play faster than it and there is no way you're going to attention starved him since it's basically always looking at the minimap. I'm not sure he can "focus" on the all map at the same time if I understood correctly. | ||
|
Musicus
Germany23576 Posts
| ||
|
imre
France9263 Posts
On January 25 2019 03:57 Serimek wrote: LiquidMana top 10 Protoss? The guy is good, but it's PR bullshit. Lying is the best PR sadly | ||
|
Musicus
Germany23576 Posts
| ||
|
FFW_Rude
France10201 Posts
On January 25 2019 03:58 Loccstana wrote: Why not play Stats or Classic? Maybe after. That would be fun | ||
|
Paljas
Germany6926 Posts
He's Top 11 Toss there | ||
|
Zheryn
Sweden3653 Posts
On January 25 2019 03:57 Serimek wrote: LiquidMana top 10 Protoss? The guy is good, but it's PR bullshit. He's rated rank #11 in the world and rank #3 of non-koreans on aligulac. | ||
|
Garrl
Scotland1972 Posts
On January 25 2019 03:56 IshinShishi wrote: you guys seriously overestimate the level of play of GM players, I dont think it could beat a pro, but low-mid GM? For sure. you're underestimating the difference in playing your mainrace is; of course it looks stronger playing someone's offrace. | ||
|
Serimek
France2274 Posts
11th ^_^ | ||
|
yht9657
1810 Posts
| ||
|
Nakajin
Canada8989 Posts
There was 9 korean P in the last GSL, can you give me 6 foreign P ahead of Mana? | ||
|
Howard_Kao
China261 Posts
| ||
|
Paljas
Germany6926 Posts
to weak, to slow | ||
|
Xamo
Spain877 Posts
| ||
|
Musicus
Germany23576 Posts
Let it beat TLO's offrace, then bring Mana, so good! | ||
|
Tyrhanius
France947 Posts
On January 25 2019 03:58 imre wrote: being as fast as an human with full vision basically means you can't harass it or i'm totally wrong? you wouldn't be able to play faster than it and there is no way you're going to attention starved him since it's basically always looking at the minimap. Actually last game, TLO does zealot runbies and Deepmind handle some quite poorly (no units on place, no probes pull, he just spends a lot of time running behind TLO's units while he could have finished the game way sooner). | ||
|
Paljas
Germany6926 Posts
| ||
|
Serimek
France2274 Posts
| ||
|
yht9657
1810 Posts
| ||
|
Dav1oN
Ukraine3164 Posts
On January 25 2019 03:58 Cyro wrote: It doesn't have maphack, it can just technically see everywhere that it has vision So it means AI has a higher resolution with it's "eyes"? Still kinda not even, players are not allowed to do so, and even if they could... :D | ||
|
Waxangel
United States33371 Posts
On January 25 2019 03:59 TheDougler wrote: Top 10 foreigner protoss maybe. I mean, it's basically Neeb, Showtime, and then everyone else as far as non-Korean protoss goes. #11 according to Aligulac, a fellow machine  http://aligulac.com/periods/233/?page=1&sort=&race=p&nats=all | ||
|
IshinShishi
Japan6156 Posts
| ||
|
Chris_Havoc
United States600 Posts
Playing vs TLO, MaNa, Stats or Classic doesn't matter to Deepmind. It doesn't know who they are most importantly it doesn't care. It can't be intimidated. The agents will run & adapt their strats regardless of the opponent. | ||
|
Kazi25
Philippines236 Posts
| ||
|
Majick
416 Posts
| ||
|
stilt
France2749 Posts
Poor mana | ||
|
Nakajin
Canada8989 Posts
| ||
|
Paljas
Germany6926 Posts
| ||
|
Xamo
Spain877 Posts
| ||
|
Durnuu
13320 Posts
| ||
|
Aeromi
France14459 Posts
On January 25 2019 03:58 Aeromi wrote: Now if MaNa drops game we start freaking out and running to London to stop Deepmind : > We are dead boys | ||
|
Firkraag8
Sweden1006 Posts
| ||
|
Paljas
Germany6926 Posts
On January 25 2019 04:09 Firkraag8 wrote: You think it overmakes probes cause it predicts losing some to Adepts so it can continue 1 base strong? No | ||
|
Dav1oN
Ukraine3164 Posts
| ||
|
TheDougler
Canada8304 Posts
| ||
|
Firkraag8
Sweden1006 Posts
How so? It didn't replace the probes after losing them. | ||
|
Gandie
158 Posts
| ||
|
Velitey
Canada13 Posts
Sorry, but losing like this to a 4 Gate is legit bad Scouted 4 gate too | ||
|
Thezzy
Netherlands2117 Posts
| ||
|
FFW_Rude
France10201 Posts
| ||
|
IshinShishi
Japan6156 Posts
| ||
|
stilt
France2749 Posts
| ||
|
Paljas
Germany6926 Posts
On January 25 2019 04:11 FFW_Rude wrote: Did they just spoiled 2nd game by showing the scoreboard ? They wont show game 2 anyway and Mana just explained how it went | ||
|
Shathe
Hungary422 Posts
| ||
|
mierin
United States4943 Posts
On January 25 2019 04:11 FFW_Rude wrote: Did they just spoiled 2nd game by showing the scoreboard ? They said we weren't going to watch it--I'm assuming they just show 3 games out of each series. | ||
|
Xamo
Spain877 Posts
| ||
|
Zheryn
Sweden3653 Posts
On January 25 2019 04:11 FFW_Rude wrote: Did they just spoiled 2nd game by showing the scoreboard ? They are not showing all games, only game 1, 3 and 4. | ||
|
Loccstana
United States833 Posts
| ||
|
yht9657
1810 Posts
On January 25 2019 04:11 Shathe wrote: Looks like we need the Koreans to defend humanity. Call Stats or Zest or something... I feel like sOs is our best hope, machines can't handle his crazy mindgames. | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:11 Paljas wrote: They wont show game 2 anyway and Mana just explained how it went Didn't hear it. Having a new born and listening to stream is not really compatible | ||
|
Kazi25
Philippines236 Posts
| ||
|
stilt
France2749 Posts
On January 25 2019 04:11 Shathe wrote: Looks like we need the Koreans to defend humanity. Call Stats or Zest or something... Release Life | ||
|
Xain0n
Italy3963 Posts
On January 25 2019 04:11 IshinShishi wrote: "alphastar is diamond "tlo is a fortnite player" "is this a joke" "im not impressed by this low masters from the african server" The one that TLO faced was way worse than this one. It trained and evolved for a weak, which means 200 years! | ||
|
Aeromi
France14459 Posts
| ||
|
TheDougler
Canada8304 Posts
| ||
|
imre
France9263 Posts
| ||
|
Poopi
France12836 Posts
Their render of view is confusing w/ regards to what was said earlier :o | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:15 TheDougler wrote: In a sense that "outcome prediction" graph is terrifying. Impressive stuff. yeah thats' awesome | ||
|
Gandie
158 Posts
On January 25 2019 04:14 Aeromi wrote: AlphaStar has no control group It doesn't need to click or control anything directly. Control groups are not necessary. | ||
|
Serimek
France2274 Posts
On January 25 2019 04:14 Aeromi wrote: AlphaStar has no control group Yeah. I'm confused about how fair it is micro-wise. | ||
|
Cyro
United Kingdom20285 Posts
On January 25 2019 04:16 Poopi wrote: So does it see the whole map or not? Their render of view is confusing w/ regards to what was said earlier :o It can see everything with vision the view is just what it's focusing attention on | ||
|
Tyrhanius
France947 Posts
| ||
|
Poopi
France12836 Posts
On January 25 2019 04:17 Cyro wrote: It can see everything with vision the view is just what it's focusing attention on Okay so it's a PR stunt, not a breakthrough in reinforcement learning. Quite disappointing | ||
|
Shathe
Hungary422 Posts
| ||
|
TheDougler
Canada8304 Posts
There is also for sure some epic meme potential there. Player 1: gl hf Player 2: gg I win easy Player 1: (picture of outcome prediction graph spiking) something like that. | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:16 Serimek wrote: Yeah. I'm confuse about how fair it is micro-wise. Meh the micro seemed pretty fair, it's way less insane than a standard Blizz brutal IA. | ||
|
IshinShishi
Japan6156 Posts
| ||
|
TheDougler
Canada8304 Posts
On January 25 2019 04:18 Poopi wrote: Okay so it's a PR stunt, not a breakthrough in reinforcement learning. Quite disappointing It doesn't have vision of everything, yes it can "see" everything it has vision of however. | ||
|
Chris_Havoc
United States600 Posts
On January 25 2019 04:18 Tyrhanius wrote: Will we see other races played today ? Or only PvP ? They're only going to be doing PvP on Catalyst today using Patch 4.6.2. | ||
|
Majick
416 Posts
On January 25 2019 04:18 Poopi wrote: Okay so it's a PR stunt, not a breakthrough in reinforcement learning. Quite disappointing Yeah it also doesn't have to scratch it's balls, while real players sometimes have to sacrifice some APM for that. Quite disappointing. | ||
|
Serimek
France2274 Posts
On January 25 2019 04:18 Nakajin wrote: Meh the micro seemed pretty fair, it's way less insane than a standard Blizz brutal IA. The footage of game 2 made me doubtful. I need to see replays. | ||
|
IshinShishi
Japan6156 Posts
| ||
|
Poopi
France12836 Posts
On January 25 2019 04:19 TheDougler wrote: It doesn't have vision of everything, yes it can "see" everything it has vision of however. Yes I know, still makes it pretty easier | ||
|
JyB
France466 Posts
| ||
|
TheDougler
Canada8304 Posts
| ||
|
sneakyfox
8216 Posts
e: and interesting that it lifted sentries instead of immortals | ||
|
yht9657
1810 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:23 JyB wrote: Disappointing. Zoom hack and basically just outmicroing players, which everyone knew is unbeatable if done perfectly. The build order and mouvement army are great on this engine. | ||
|
Tyrhanius
France947 Posts
| ||
|
MarthTV
Germany387 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
IshinShishi
Japan6156 Posts
| ||
|
Howard_Kao
China261 Posts
| ||
|
Alethios
New Zealand2765 Posts
| ||
|
peanuts
United States1225 Posts
| ||
|
Xamo
Spain877 Posts
On January 25 2019 04:25 Nakajin wrote: The build order and mouvement army are great on this engine. Micro and decision making is also very good, and it is very very decisive. Its evaluation of its winning position in the game and in each battle is at the top. Notice how it pulled a few probes when Mana attacked in game 3, to assure it would win the battle decisively. So far I do not see any weakness in this second AlphaStar. Fear. | ||
|
Musicus
Germany23576 Posts
On January 25 2019 04:28 Alethios wrote: Is AlphaStar apm capped like the DotA2 agent? Yeah it's capped and on average uses less APM than pros, it also has an artificial reaction of 350ms which is also on the slow end of progames. | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:27 IshinShishi wrote: they need to tune the micro ability down for sure, some moves look very inhuman like immediately targeting the immortal as it was coming out without missing a beat I don't see how you tune that off, you click on the immortal as quick as you can when it pop off, you can put the bot at 60 APM and he is still gonna click on it at the first second. | ||
|
TheDougler
Canada8304 Posts
On January 25 2019 04:28 peanuts wrote: Just got home, what's the verdict thus far? Not much to say about the TLO series, like, it was okay, but TLO played really bad to be honest. This second series is the real deal though. This is awesome! | ||
|
pvsnp
7676 Posts
On January 25 2019 04:28 peanuts wrote: Just got home, what's the verdict thus far? It can beat at least some progamers, so far TLO and Mana in PvP. | ||
|
Acrofales
Spain17982 Posts
On January 25 2019 04:27 Howard_Kao wrote: The micro is just too impossible for human being, if Serral forgets his mouse pad again there'll be no way that human can beat it. Google should add restriction on things like mouse movement/friction This is a fair point. The next step should definitely be a robot arm moving a mouse or something to introduce this lag and imprecision, which is a very real limitation in the game. | ||
|
Alethios
New Zealand2765 Posts
On January 25 2019 04:29 Musicus wrote: Yeah it's capped and on average uses less APM than pros, it also has an artificial reaction of 350ms which is also on the slow end of progames. Ah right. Neat. | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:30 TheDougler wrote: Not much to say about the TLO series, like, it was okay, but TLO played really bad to be honest. This second series is the real deal though. This is awesome! And the 2 series were played a week of each others and it's much much better already. | ||
|
fLyiNgDroNe
Belgium4005 Posts
| ||
|
peanuts
United States1225 Posts
On January 25 2019 04:31 pvsnp wrote: It can beat at least some progamers, so far TLO and Mana in PvP. Can it only play PvP at present? Or did they choose that matchup to showcase its micro potential? | ||
|
Serimek
France2274 Posts
| ||
|
Xamo
Spain877 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:33 fLyiNgDroNe wrote: This is shit, AI's are going to figure out optimal strategies and tricks too fast and we won't see learning curve anymore He's doing the most standard build right now... a bit more probe I guess | ||
|
Alethios
New Zealand2765 Posts
On January 25 2019 04:34 Xamo wrote: AlphaStar teaching us that it is better to go to 24 probes? 2/3 on gas. | ||
|
Zaros
United Kingdom3692 Posts
On January 25 2019 04:32 Nakajin wrote: And the 2 series were played a week of each others and it's much much better already. It plays at an increased speed in practice too, 200 years of starcraft already played. So one week is like the hyperbolic time chamber for the AI. | ||
|
Majick
416 Posts
15 AlphaStar Agents Serral | ||
|
FFW_Rude
France10201 Posts
| ||
|
Tyrhanius
France947 Posts
| ||
|
Puosu
6985 Posts
So sick. | ||
|
TBO
Germany1350 Posts
On January 25 2019 04:38 FFW_Rude wrote: How did it know hallucination will go that route. it moved before seeing it right ? maybe AI figured out how to download maphacks :D | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:38 TBO wrote: maybe AI figured out how to download maphacks :D That would be funny tbh | ||
|
Kazi25
Philippines236 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
aringadingding
474 Posts
| ||
|
fLyiNgDroNe
Belgium4005 Posts
| ||
|
Acrofales
Spain17982 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:41 Nakajin wrote: Multitask seems like it's flaw right now? NVM lol | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:42 fLyiNgDroNe wrote: at which point do we call Mana "John Connor"? Did John Connor died ? | ||
|
Garrl
Scotland1972 Posts
| ||
|
stilt
France2749 Posts
| ||
|
Gandie
158 Posts
| ||
|
Jacenoob
299 Posts
| ||
|
TheDougler
Canada8304 Posts
| ||
|
Majick
416 Posts
| ||
|
Xamo
Spain877 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Zetter
Germany629 Posts
| ||
|
chipmonklord17
United States11944 Posts
| ||
|
phodacbiet
United States1740 Posts
| ||
|
FFW_Rude
France10201 Posts
| ||
|
Garrl
Scotland1972 Posts
On January 25 2019 04:43 Zetter wrote: Yea Artosis, 1.5k APM is an acceptable pro level. lol a 200 apm limit would even probably be higher than most proes considering wasted actions and spam | ||
|
Brutaxilos
United States2628 Posts
On January 25 2019 04:37 Majick wrote: Blizzcon 2019 Ro16: 15 AlphaStar Agents Serral Honestly though, if AlphaStar can beat Mana 5-0 in 2 weeks of training, Serral doesn't stand a chance. | ||
|
IshinShishi
Japan6156 Posts
| ||
|
Zaros
United Kingdom3692 Posts
On January 25 2019 04:44 Brutaxilos wrote: Honestly though, if AlphaStar can beat Mana 5-0 in 2 weeks of training, Serral doesn't stand a chance. It hasnt had 2 weeks it has had 200 years | ||
|
pvsnp
7676 Posts
| ||
|
gophersnake
48 Posts
| ||
|
Musicus
Germany23576 Posts
On January 25 2019 04:45 IshinShishi wrote: if its 310 pure effective apm then its still inhuman actually that is Serral level | ||
|
Garrl
Scotland1972 Posts
| ||
|
Serimek
France2274 Posts
On January 25 2019 04:46 Musicus wrote: actually that is Serral level https://twitter.com/retjah/status/959047573930696704 But Serral does not have a zoomhack and he has hands. | ||
|
Aeromi
France14459 Posts
| ||
|
Acrofales
Spain17982 Posts
Another superhuman play you saw was the timing on making the observer: a human almost certainly wouldn't see the blurred dark templar crossing the map until they are very near his base. Whereas alphastar has 100% map awareness. | ||
|
peanuts
United States1225 Posts
| ||
|
Poopi
France12836 Posts
On January 25 2019 04:46 Musicus wrote: actually that is Serral level https://twitter.com/retjah/status/959047573930696704 Fun thing is that Lambo had the same ePM in his games against Maru. Shows how much they hyped Serral on even silly things :D. | ||
|
Zaros
United Kingdom3692 Posts
On January 25 2019 04:46 Musicus wrote: actually that is Serral level https://twitter.com/retjah/status/959047573930696704 No it isnt, are everyone of serral's 310 actions perfectly calculated and precise. no. | ||
|
Cyro
United Kingdom20285 Posts
put it back in the hyperbolic time chamber | ||
|
Dav1oN
Ukraine3164 Posts
On January 25 2019 04:46 Garrl wrote: this is a place where that zoomhack also comes into play which I see, of course, none of them are mentioning. exactly, it needs to see the same way as a player does, only one screen | ||
|
rednusa
651 Posts
| ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 04:48 peanuts wrote: The micro is incredibly interesting, the overdroning to immediately saturate/handle any workers killed from harassment was intriguing, and the army movement was brilliant. But overall, it seems like it's just exploiting its inherent ability to control units better/have persistent awareness of its FOV, and leveraging that to get into micro intensive situations and win. I feel like Artosis and Rotterdam are kind of over-hyping it to the point where they're overlooking the ridiculous inhuman micro. | ||
|
Brutaxilos
United States2628 Posts
Yea, but 2 weeks of real life time. If we do a showmatch at Blizzcon, thats like another 40 or so weeks? So the equivalent of 4 millennia of training? | ||
|
JyB
France466 Posts
| ||
BisuDagger
Bisutopia19234 Posts
| ||
|
Musicus
Germany23576 Posts
On January 25 2019 04:48 Zaros wrote: No it isnt, are everyone of serral's 310 actions perfectly calculated and precise. no. We need that Kappa emote allowed on TL already, does not cut in anymore. | ||
|
fLyiNgDroNe
Belgium4005 Posts
| ||
|
renaissanceMAN
United States1840 Posts
Can't recall the last time I saw Serral's APM peak above 700 even in intense fights. | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:49 JyB wrote: Again, just outmicroing with zoom hack. Boring. You forget everything else though... The learning, the positionning etc... But i'm more puzzled by the obs without seeing the darkshrine or pre positionning with the hallu | ||
|
Zaros
United Kingdom3692 Posts
On January 25 2019 04:49 Brutaxilos wrote: Yea, but 2 weeks of real life time. If we do a showmatch at Blizzcon, thats like another 40 or so weeks? So the equivalent of 4 millennia of training? I think they said 200 years in total since development over the past 2 years so it might be another 100 years training if it came again at blizzcon. | ||
|
CobaltBlu
United States919 Posts
| ||
|
siakam
Korea (South)11 Posts
But I think the alphastar has far more advantage in micro than even top pro player from what I see | ||
|
Zheryn
Sweden3653 Posts
Blizzcon 2019 is not in 4000 years tho. | ||
|
Garrl
Scotland1972 Posts
On January 25 2019 04:50 BisuDagger wrote: When you talk about years played, SC2 players are really in their infancy. You are never a year away from a patch. This is coupled with the fact that we don't have truly dominant players in SC2 yet. Winning for a year or two is as good as it has gotten. I would have found this more interesting had SC2 had 5 more years between its last patch or if this were trained in Brood War. In Brood War their is at least a definitive player with the most well rounded skill set in Macro, Micro, and Strategy. *Note: I'm not comparing the games at all in terms of content. now I'm opening myself up to flame here but I do think that BW micro is much less about the sheer APM and more tactical so I don't think it would be as good as it is in SC2. | ||
|
mierin
United States4943 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
yht9657
1810 Posts
| ||
|
peanuts
United States1225 Posts
| ||
|
JyB
France466 Posts
On January 25 2019 04:51 siakam wrote: Well it is impressive But I think the alphastar has far more advantage in micro than even top pro player from what I see Of course it has near perfect micro => On January 25 2019 04:43 Jacenoob wrote: wow mono blink stalkers with high APM and perfect clicking what high intelligence | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:52 mierin wrote: Blink stalkers seem to be one of the easiest things to abuse from an AI. I wish they'd chosen any other race than P honestly. Dunno thing like medivack pick up and speed lings have insane micro potential | ||
|
renaissanceMAN
United States1840 Posts
| ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 04:51 Garrl wrote: now I'm opening myself up to flame here but I do think that BW micro is much less about the sheer APM and more tactical so I don't think it would be as good as it is in SC2. it would beat flash 101-0 with vulture micro alone | ||
|
FFW_Rude
France10201 Posts
On January 25 2019 04:55 renaissanceMAN wrote: Is there a reason they're only having it play PVP? I guess it's only programmed for this as of now | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 04:53 peanuts wrote: Wait... they're hyping up its ability to generate new strats and win through multiple means, but didn't they say that each match involved a different agent? How do we know that it could actually learn and react to a human player over the course of a series? It wouldn't, I don't think it gives more importance to the last game than the 10 000 previous one. | ||
|
geokilla
Canada8231 Posts
| ||
|
StackerTwo
United States41 Posts
| ||
|
pvsnp
7676 Posts
| ||
|
Majick
416 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
Gandie
158 Posts
On January 25 2019 04:59 Musicus wrote: zoom out map hack removed for this new agent, no complains after this? Why would anyone complain. It's not a battle of humans vs. machines. | ||
|
Brutaxilos
United States2628 Posts
AlphaStar trains on a faster version of the game. They said 1 week of training was the equivalent of 200 years in real life. So if it trains 24/7 until Blizzcon, it'd be a few thousand years equivalency of training in real life time. | ||
|
gophersnake
48 Posts
On January 25 2019 04:59 Musicus wrote: zoom out map hack removed for this new agent, no complains after this? No, they'll find something to complain about. | ||
|
Duckman
United States158 Posts
| ||
|
SaintTieum
South Africa22 Posts
On January 25 2019 04:59 Nakajin wrote: Deepmind is calculating and responding to the hack accusation instanly, even avilo is doomed. Hahaha | ||
|
Poopi
France12836 Posts
Live game with AI less unfair Sounds fun | ||
|
Musicus
Germany23576 Posts
On January 25 2019 04:59 Nakajin wrote: Deepmind is calculating and responding to the hack accusation instanly, even avilo is doomed. haha, it's really next level | ||
|
cha0
Canada504 Posts
On January 25 2019 05:00 Gandie wrote: Why would anyone complain. It's not a battle of humans vs. machines. Because of the PR angle they keep putting on it. Like in this behind the scenes video they just showed the lead guy says "AlphaStar is limited to having the reaction time and control speed of an average human" when having just seen the games we can all see it clearly has inhuman micro. | ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 04:59 Majick wrote: You can clearly see what starcraft is about. If you have significantly better mechanics, you don't need sophisticated strategy, you just make more stuff and go kill em. Yes, this is something that not many people are willing to admit or even consider in sc2, but some players are just dumb with very good mechanics, not the master minds that many would like to think. | ||
|
StackerTwo
United States41 Posts
| ||
|
ZlyKiss
Poland697 Posts
| ||
|
JyB
France466 Posts
On January 25 2019 04:59 Musicus wrote: zoom out map hack removed for this new agent, no complains after this? No 1.5K apm spike? | ||
|
Puosu
6985 Posts
On January 25 2019 05:02 IshinShishi wrote: Yes, this is something that not many people are willing to admit or even consider in sc2, but some players are just dumb with very good mechanics, not the master minds that many would like to think. Can you not? Calling players dumb is just fucking stupid no matter how much you think the game's got or hasn't got strategy or mechanics or whatever. | ||
|
Gandie
158 Posts
On January 25 2019 05:02 cha0 wrote: Because of the PR angle they keep putting on it. Like in this behind the scenes video they just showed the lead guy says "AlphaStar is limited to having the reaction time and control speed of an average human" when having just seen the games we can all see it clearly has inhuman micro. The quote is true though. Where AS differs in its decision making. If every action is perfectly placed, regular pro reaction time and control speed are enough to dominate these games. | ||
|
ASoo
2864 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
Gandie
158 Posts
On January 25 2019 05:02 StackerTwo wrote: Also something to note AlphaStar roughly 300 "effective" apm vs human pros who has maybe 400 spam apm & 150 eapm Serral in his WCS Leipzig replays consistently has 300+ EPM. 344 EPM in a game vs major's bio. The 3 other semi-finalists are around 200 EPM. Top Koreans I've looked are between 200-240. Serral is 50% faster than his opponents on average. Scary! https://twitter.com/retjah/status/959047573930696704?lang=de | ||
|
Majick
416 Posts
| ||
|
cha0
Canada504 Posts
On January 25 2019 05:03 Gandie wrote: The quote is true though. Where AS differs in its decision making. If every action is perfectly placed, regular pro reaction time and control speed are enough to dominate these games. Have you been watching the games? Specifically the part where it's apm spiked to 1500 while microing stalkers vs Mana?? They limited it's 'average' apm, and the AI just learned to game the system, purposefully lowering it's apm out of battle and then spiking it super high during battle. | ||
|
IshinShishi
Japan6156 Posts
That's why I emphasized the "pure" aspect of it, I can get to 200+ eapm sometimes or at least the engine says I did, but we all know that wasn't really 200 precise meaningful actions at all, that's not the case with this A.I. | ||
|
Musicus
Germany23576 Posts
don't think they changed that! | ||
|
Musicus
Germany23576 Posts
| ||
|
Shathe
Hungary422 Posts
| ||
|
yht9657
1810 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
Zaros
United Kingdom3692 Posts
| ||
|
JyB
France466 Posts
so it's still just BS micro :o) The AI just goes around the avg limitations. | ||
|
renaissanceMAN
United States1840 Posts
| ||
|
Paljas
Germany6926 Posts
| ||
|
renaissanceMAN
United States1840 Posts
| ||
|
Garrl
Scotland1972 Posts
also apparently doesnt know how to build any other units than stalkers and oracles | ||
|
Nakajin
Canada8989 Posts
| ||
|
Majick
416 Posts
On January 25 2019 05:10 renaissanceMAN wrote: fucking bot LOVES stalkers holy cow lol | ||
|
MarthTV
Germany387 Posts
| ||
|
Jasper_Ty
101 Posts
| ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:12 Garrl wrote: AI actually getting out multitasked lmao CAN'T BREATH | ||
|
RandomPlayer
Russian Federation390 Posts
| ||
|
Majick
416 Posts
| ||
|
Shathe
Hungary422 Posts
| ||
|
Zaros
United Kingdom3692 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Brutaxilos
United States2628 Posts
| ||
|
cha0
Canada504 Posts
On January 25 2019 05:13 RandomPlayer wrote: People who find excuses and complain about EPM/POV difference - you are so many! Just shut up! You are not correct. Enlighten us. | ||
|
yht9657
1810 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Nakajin
Canada8989 Posts
| ||
|
pvsnp
7676 Posts
| ||
|
Tyrhanius
France947 Posts
| ||
|
romson87
Poland487 Posts
| ||
|
FFW_Rude
France10201 Posts
| ||
|
ramask2
Thailand1024 Posts
| ||
|
Duckman
United States158 Posts
| ||
|
Jono7272
United Kingdom6330 Posts
| ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:15 pvsnp wrote: FantasyStar crying I'm laughing so hard | ||
|
Poopi
France12836 Posts
On January 25 2019 05:13 RandomPlayer wrote: People who find excuses and complain about EPM/POV difference - you are so many! Just shut up! You are not correct. So why is alphastar suddenly so bad now that it has only one screen vision? | ||
|
Doko
Argentina1737 Posts
| ||
|
Musicus
Germany23576 Posts
| ||
|
Ace Frehley
2030 Posts
| ||
|
pvsnp
7676 Posts
| ||
|
Xamo
Spain877 Posts
Mana our hero | ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:16 Ace Frehley wrote: Walking back whole army for warp prism was a bad move, alphastar would win basetrade easily would it though? | ||
|
Brutaxilos
United States2628 Posts
| ||
|
peanuts
United States1225 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 05:15 Poopi wrote: So why is alphastar suddenly so bad now that it has only one screen vision? You think it looked bad? | ||
|
Paljas
Germany6926 Posts
On January 25 2019 05:16 Ace Frehley wrote: Walking back whole army for warp prism was a bad move, alphastar would win basetrade easily It wouldn't be a base trade, it would be Mana defending and AlphaStar losing his main | ||
|
Ace Frehley
2030 Posts
Mana had 2 bases and tiny army, alphastar had 30+ stalkers plus godly micro | ||
|
IshinShishi
Japan6156 Posts
| ||
|
siakam
Korea (South)11 Posts
Alphastar was impressive, it look liked have some limitations but will soon be much better Hope it will be good affect for sc2 | ||
|
pvsnp
7676 Posts
Agree with this, AlphaStar is not so stupid. Remember that in the other blink game it relied on poking and skirmishing instead of committing against a bunch of Immortals. Not sure going for the basetrade would have been optimal. | ||
|
Garrl
Scotland1972 Posts
as such it's a micro bot and not really intelligent in terms of builds or improvisation. | ||
|
gophersnake
48 Posts
| ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:18 IshinShishi wrote: it really doesnt deal well with harass, seems like you can beat it with dropship shenanigans all day Someone hook it up to BW and call Boxer. | ||
|
RandomPlayer
Russian Federation390 Posts
On January 25 2019 05:15 Poopi wrote: So why is alphastar suddenly so bad now that it has only one screen vision? because Mana has adjusted and tricked it. | ||
|
Poopi
France12836 Posts
Yeah going back and forth to defend the prism, not making a phoenix and stuff was really bad. | ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:18 RandomPlayer wrote: because Mana has adjusted and tricked it. So if you give it an inhuman advantage it wins, but if you make it more human it's just mana tricking it?  | ||
|
ZigguratOfUr
Iraq16955 Posts
| ||
|
CobaltBlu
United States919 Posts
| ||
|
FFW_Rude
France10201 Posts
On January 25 2019 05:19 renaissanceMAN wrote: So if you give it an inhuman advantage it wins, but if you make it more human it's just mana tricking it? That's called CounterStrike mentality. You are a noob or a cheater. No middleground | ||
|
[F_]aths
Germany3947 Posts
| ||
|
siakam
Korea (South)11 Posts
No gg no skill | ||
|
renaissanceMAN
United States1840 Posts
On January 25 2019 05:21 FFW_Rude wrote: That's called CounterStrike mentality. You are a noob or a cheater. No middleground good thing I'm fucking garbage at cs:go | ||
|
The_Red_Viper
19533 Posts
On January 25 2019 05:19 ZigguratOfUr wrote: Alphastar leaving the base with his army repeatedly and Mana's warp prism just going back in immediately doesn't speak too highly of its decision making ability. But I'm sure it'll get better. That's a very specific case though, at some point it will build a phoenix when something like this happens and then win the game and thus be reinforced to build a phoenix more often when something like this happens and thus learn. | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 05:19 Poopi wrote: Yeah going back and forth to defend the prism, not making a phoenix and stuff was really bad. Sure but the oracle play was great, that 2 oracle in the main with a front pressure was nasty, would have won against 99% of the players. | ||
|
Zaros
United Kingdom3692 Posts
| ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 05:22 The_Red_Viper wrote: That's a very specific case though, at some point it will build a phoenix when something like this happens and then win the game and thus be reinforced to build a phoenix more often when something like this happens and thus learn. yea, I think the other version faced this in it's 200 years of training, this one is just a lil pup | ||
|
[F_]aths
Germany3947 Posts
| ||
|
Poopi
France12836 Posts
On January 25 2019 05:22 Nakajin wrote: Sure but the oracle play was great, that 2 oracle in the main with a front pressure was nasty, would have won against 99% of the players. I don't think you should judge how an AI like that really performs based on its strong showings (often due to inherent advantages), but on how robust it is to trickeries. | ||
|
_fool
Netherlands677 Posts
On January 24 2019 05:20 _fool wrote: I think we will be amazed. I was amazed | ||
|
Kazi25
Philippines236 Posts
| ||
|
Xamo
Spain877 Posts
On January 25 2019 05:25 Poopi wrote: I don't think you should judge how an AI like that really performs based on its strong showings (often due to inherent advantages), but on how robust it is to trickeries. The DeepMind team seems to agree, they specifically told us that they picked the "least exploitable agents" from their league. | ||
|
Xamo
Spain877 Posts
| ||
|
Lanthdoral
Germany7 Posts
| ||
|
Deleted User 3420
24492 Posts
| ||
|
RandomPlayer
Russian Federation390 Posts
| ||
|
Aiingel
243 Posts
| ||
|
FFW_Rude
France10201 Posts
"AlphaStarAi says : GIMME SERRAL OR I'M GONNA KILL CORDANA" made me laugh | ||
|
SPcrusader
Norway99 Posts
| ||
|
Nakajin
Canada8989 Posts
On January 25 2019 05:25 Poopi wrote: I don't think you should judge how an AI like that really performs based on its strong showings (often due to inherent advantages), but on how robust it is to trickeries. Well the strategt the IA came up with was both very good vs what Mana was doing and didn't depend on any inherent advantages and it dealt well enough with the adepts harras, sure clearly it didn't understand how to respond to immortal-WP but it still look like an IA who would be unbeatable in a few week, at least in pvp on this map. | ||
|
Zrana1
Netherlands45 Posts
Would love to see an all races all maps version in the GSL edit: ..or in GM ladder | ||
|
Aiingel
243 Posts
| ||
|
Akio
Finland1838 Posts
Maybe the meta for 2019 PvP will be to go for 18 disruptors now | ||
|
PanS3rnik
18 Posts
INoVation.exe LOADING ████░░░░░░░░ | ||
|
swissman777
1106 Posts
| ||
|
Charoisaur
Germany15942 Posts
The thing that interests me more how it will perform in the other matchups since those are I think more complex and harder to figure out (especially the non-mirrors). I also think they should limit the max apm since it's not human-like to reach 800 peaks unless its rapidfire/holding down the drone key. When it was microing stalkers in the 4th game against Mana it got 1000 apm peaks microing completely inhuman with them | ||
|
Brutaxilos
United States2628 Posts
On January 25 2019 05:48 Charoisaur wrote: Incredibly impressive. If it can beat one pro it can beat every pro with a bit more time. The thing that interests me more how it will perform in the other matchups since those are I think more complex and harder to figure out (especially the non-mirrors). I also think they should limit the max apm since it's not human-like to reach 800 peaks unless its rapidfire/holding down the drone key. Honestly, even if other matchups are like 10-100 times more complex, that's pretty easy for a computer to scale up to. So I doubt it'll have any issues with that. | ||
|
Nakajin
Canada8989 Posts
On January 25 2019 05:50 Brutaxilos wrote: Honestly, even if other matchups are like 10-100 times more complex, that's pretty easy for a computer to scale up to. So I doubt it'll have any issues with that. Maybe varieties are gonna be his weakness? Dunno if it need to start up from scratch when you change the maps/races, they probably talk about it but I don't recall. A single engine may have a hard time learning to play all of the match up and maps. | ||
|
mishimaBeef
Canada2259 Posts
contrast to if you spent on a nexus, now you are underminining with 2 nexuses (sure you can produce probes two at a time, but your mining rate drops considerably for some time) | ||
|
Poopi
France12836 Posts
https://deepmind.com/research/alphastar-resources/ I'm talking about the 24 january 2019 game (exhibition game). Can someone open it or has the same issue? edit: the error message is "impossible d'ouvrir la carte" aka "impossible to open the map" in english | ||
|
Achamian
82 Posts
Overall so cool to see the program use different play styles and organically react. I was worried it would simply blink stalker rush every game. I think Mana winning after preparing shows how fucking awesome StarCraft is. EDIT: I also think to truly beat a human player it needs to be one Agent in a BO7 on multiple maps. | ||
|
aringadingding
474 Posts
| ||
|
aringadingding
474 Posts
| ||
|
Doko
Argentina1737 Posts
| ||
|
True_Spike
Poland3422 Posts
| ||
|
Achamian
82 Posts
On January 25 2019 06:05 aringadingding wrote: sOs would break AlphaStar Yeah I sort of agree, especially in a longer series. | ||
|
Rowrin
United States280 Posts
On January 25 2019 06:02 Poopi wrote: Probably since it's an old version of the game, my sc2 doesn't want to open up the replay of mana vs alphastar, which is weird considering in the past you could load old replays just fine (even tho it took more time than regular replays). https://deepmind.com/research/alphastar-resources/ I'm talking about the 24 january 2019 game (exhibition game). Can someone open it or has the same issue? edit: the error message is "impossible d'ouvrir la carte" aka "impossible to open the map" in english You need to download their custom CatalystLE map I believe. Instructions are at the bottom of their replays page. | ||
|
Lazzarus
Faroe Islands114 Posts
People are talking about making the AI "human-like", and I'm just amazed that it is at a stage that it needs to have handicaps to make it fair. | ||
|
Poopi
France12836 Posts
On January 25 2019 06:20 Rowrin wrote: You need to download their custom CatalystLE map I believe. Instructions are at the bottom of their replays page. Yeah, did the trick, thanks! They should have put that on top | ||
|
Anc13nt
1557 Posts
| ||
|
Brutaxilos
United States2628 Posts
On January 25 2019 05:55 Nakajin wrote: Maybe varieties are gonna be his weakness? Dunno if it need to start up from scratch when you change the maps/races, they probably talk about it but I don't recall. A single engine may have a hard time learning to play all of the match up and maps. It's not going to make a difference. There will probably be different AIs for different matchups. And in the end you can just conjoin them together in the beginning of a decision tree to make one big AI (if PvZ use AI2342423424 if TvP use AI234324323424 etc.). There's no reason an AI should need to know a PvZ metagame to play TvP for example. | ||
|
yep
Canada9 Posts
| ||
|
CoupdeBoule
73 Posts
User was warned for this post | ||
|
SnowAngel
Finland38 Posts
| ||
|
Poopi
France12836 Posts
Super cool to see it able to play vs real players, but each agent seems very narrow minded. Their article https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/ is really detailed if you wanna see the technical details of it | ||
|
Torvaltz
United States188 Posts
i dont understand the amazement but ok, to me this looks like were still far away from an actual sc2 ai | ||
|
deacon.frost
Czech Republic12129 Posts
| ||
|
_fool
Netherlands677 Posts
On January 25 2019 06:44 Torvaltz wrote: so it seems to me like they ended up with a maphacking microbot that got wrecked when they turned that stuff off? i dont understand the amazement but ok, to me this looks like were still far away from an actual sc2 ai Wait what? If this is not an actual SC2 AI, what is? | ||
|
cha0
Canada504 Posts
On January 25 2019 06:26 Brutaxilos wrote: It's not going to make a difference. There will probably be different AIs for different matchups. And in the end you can just conjoin them together in the beginning of a decision tree to make one big AI (if PvZ use AI2342423424 if TvP use AI234324323424 etc.). There's no reason an AI should need to know a PvZ metagame to play TvP for example. That is not their goal though. Their ultimate goal is to be able to train a model that can make general decisions regardless of the situation. So in the future with just minor tweaks they could take this model and apply it to different domains. If they end up just having a giant list of various AI's which each specialize in certain things then they aren't making the progress they desire. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 06:58 cha0 wrote: That is not their goal though. Their ultimate goal is to be able to train a model that can make general decisions regardless of the situation. So in the future with just minor tweaks they could take this model and apply it to different domains. If they end up just having a giant list of various AI's which each specialize in certain things then they aren't making the progress they desire. You will train specific neural networks and then you will create one that covers it all. This actually simulates the human brain(when you play PvZ you load your PvZ persona, when you play ZvP you load your ZvP persona) and it will create one big AI. Then they can actually train this whole new network. If it's feasable - I have no clue, I am way too out of touch with the technology, several years I haven't touched ML which is a shit ton of time | ||
|
Torvaltz
United States188 Posts
On January 25 2019 06:55 _fool wrote: Wait what? If this is not an actual SC2 AI, what is? The type that wins through its intelligence and decision making, not through impossible micro and perfect control through multiple screens. We already know micro bots are great, to me the decision making part is where the real AI breakthrough could be. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 07:04 Torvaltz wrote: The type that wins through its intelligence and decision making, not through impossible micro and perfect control through multiple screens. We already know micro bots are great, to me the decision making part is where the real AI breakthrough could be. It had a decision making. Not perfect, but considering this can be called first playable interation it was still impressive. | ||
|
GoloSC2
710 Posts
On January 25 2019 07:12 deacon.frost wrote: It had a decision making. Not perfect, but considering this can be called first playable interation it was still impressive. Also it had neither maphack nor was it like the microbots we know. It seems you didn't really try to understand was what going on | ||
|
Torvaltz
United States188 Posts
On January 25 2019 07:17 GoloSC2 wrote: Also it had neither maphack nor was it like the microbots we know. It seems you didn't really try to understand was what going on I understand what they are going for but I don't think they're as close to it as you guys seem to, thats pretty much my take. limit it more and then Id be impressed it works and it makes choices but they arent better than pros nor are they an apocalyptic sign to me | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 07:17 GoloSC2 wrote: Also it had neither maphack nor was it like the microbots we know. It seems you didn't really try to understand was what going on Actually it's hard to imagine/understand. If I didn't write some neural networks at university I would be on the pesimistic side as well. But since I know this I am on the exact other side. I am so fucking amazed. I hope those people can read this, because I am hyped. They did a really good job in a game I love. Well done. | ||
|
theunabletable
54 Posts
On January 25 2019 06:44 Torvaltz wrote: so it seems to me like they ended up with a maphacking microbot that got wrecked when they turned that stuff off? i dont understand the amazement but ok, to me this looks like were still far away from an actual sc2 ai I mean even in that game it played really well, except that it obviously had some kind of glitch or hole when it comes to dealing with drops. It was ahead for most of the game, up until the incident. It's absolutely a proof of concept. Although, yeah, it's probably a little unfairly good at micro lol. | ||
|
Poopi
France12836 Posts
On January 25 2019 07:20 Torvaltz wrote: I understand what they are going for but I don't think they're as close to it as you guys seem to, thats pretty much my take. limit it more and then Id be impressed It's actually extremely impressive. Basically, in the only version (camera interface) that is relatively close to fair (apm and stuff will have to be examined because it seems the AI exploits it a bit too much), they managed to have an AI not to be dumb for like 9 minutes, before it showed signs of being an AI. If it were a Starcraft Turing test, most people would have known it was an AI at the warp prism incident. Before that, it looked legit like an human. The principle of the anonymous tournament from NoRegret & Scarlett (not sure who exactly are the organizers) with AlphaStar would be interesting, with humans and AlphaStar agents anonymously mixed in | ||
|
_fool
Netherlands677 Posts
On January 25 2019 07:04 Torvaltz wrote: The type that wins through its intelligence and decision making, not through impossible micro and perfect control through multiple screens. We already know micro bots are great, to me the decision making part is where the real AI breakthrough could be. A common mistake with AI sceptics is that they opt to use a definition of intelligence that requires one or more properties that they only acknowledge in humans. In that case it's no wonder that they never find intelligence anywhere else. What you're doing is very similar, if not just that. Of course it made decisions. When and where to expand. What to spend its money on. When to attack. Remember that there are no procedural rules programmed into it. So every click that you saw was a decision. And making decisions based on observations is what I would call intelligence (definition up for refinement). You act like the AI was given 20 stalkers and then it micro'd its way out of the game. But it's far more complex than that. It found a way to build those 20 stalkers while harassing. And it found out by itself. | ||
|
Waxangel
United States33371 Posts
On January 25 2019 07:20 Torvaltz wrote: I understand what they are going for but I don't think they're as close to it as you guys seem to, thats pretty much my take. limit it more and then Id be impressed it works and it makes choices but they arent better than pros nor are they an apocalyptic sign to me did you just go "I don't totally understand the facts but this is what I WANT to believe"??????????? weighing a ban ;o | ||
|
Torvaltz
United States188 Posts
On January 25 2019 07:30 Waxangel wrote: did you just go "I don't totally understand the facts but this is what I WANT to believe"??????????? weighing a ban ;o when did i say that phrase my dude? | ||
|
Xamo
Spain877 Posts
On January 25 2019 06:30 CoupdeBoule wrote: Started watching. 30 sec into the first replay it shows that the AI isnt worker-pairing. Oh well i wasnt expecting much but im still disappointed - not gonna waste my time watching that garbage Games 1-5 are against "basic" (GM?) agents, with one week of training. Games 6-10 are against "advanced" (top-pro?) agents, with two weeks of training but no need to control the camera - their attention is at the same time on the whole map that is visible (although their actions are not). Game 11 is against a "basic" agent with camera control. Games 6-11 are the best. | ||
|
renaissanceMAN
United States1840 Posts
https://www.vox.com/future-perfect/2019/1/24/18196177/ai-artificial-intelligence-google-deepmind-starcraft-game?fbclid=IwAR1z-eJ5Sm1amrbR9WvXZO0CD4_jVQTWq64eJAOuXvwtyK70j2k63Qn-tug | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
snakeeyez
United States1231 Posts
No ai so far is even remotely close to this. Its very amazing that it learned all of this just on its own. | ||
|
IshinShishi
Japan6156 Posts
On January 25 2019 10:27 mishimaBeef wrote: anyone watch the game mana mentioned where alphastar did a proxy? yes, mana threw the game and got baited hard by shield batteries. I think Mana won the last game because the A.I refused to engage his army comp since the predicted outcome was a loss nearly always, what it didn't consider was that it was playing vs a human with flawed mechanics which it probably could outmicro even in that situation. | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
Rea-Rea
United States42 Posts
| ||
|
Dante08
Singapore4128 Posts
On January 25 2019 13:29 Rea-Rea wrote: It has an unfair advantage with several games where it had full map view rather than camera control. It's an unfair advantage. All of the games should've been camera control, flat out. Full map view as in vision of the whole map with no fog of war? | ||
|
Doko
Argentina1737 Posts
On January 25 2019 14:15 Dante08 wrote: Full map view as in vision of the whole map with no fog of war? No, its still limited by fog of war but capable of zooming out completely to where it can see the whole map but issue commands with perfect precision. | ||
|
Zzoram
Canada7115 Posts
On January 25 2019 14:15 Dante08 wrote: Full map view as in vision of the whole map with no fog of war? No, the first 10 games it had a zoomed out view of the map instead of having to use a mini map where you don’t know what showed up until you move your camera to check, but the fog of war was still there. It still had to move the camera to a location to perform an action, but it was an advantage. The 11th game, it had to use a regular mini map instead of getting a super zoom out. It slightly reduces multi prong micro potential. | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
cha0
Canada504 Posts
On January 25 2019 15:01 mishimaBeef wrote: i watched the games the micro didn't seem undoable with good hotkey use, e.g. reassigning hurt stalkers to a quick retreat hotkey, having 3 groups of stalkers on hotkeys and double tapping around Not sure we watched the same games if you think any human could do the stalker and disruptor micro the AI was pulling. It was literally microing 6+ units at a time and had its apm spiking to 1500 sometimes. | ||
|
loft
United States344 Posts
| ||
|
valas991
Hungary181 Posts
I think the games were really entertaining. I also think #11 is the way to go, in terms of fairness. It will be nice once all races will be shown, eventually. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 17:10 valas991 wrote: Is there a VOD available? A friend of mine missed it. I think the games were really entertaining. I also think #11 is the way to go, in terms of fairness. It will be nice once all races will be shown, eventually. | ||
|
Vision_
861 Posts
| ||
|
Acrofales
Spain17982 Posts
On January 25 2019 10:12 renaissanceMAN wrote: Ugh: https://www.vox.com/future-perfect/2019/1/24/18196177/ai-artificial-intelligence-google-deepmind-starcraft-game?fbclid=IwAR1z-eJ5Sm1amrbR9WvXZO0CD4_jVQTWq64eJAOuXvwtyK70j2k63Qn-tug Once you get past the first two or three paragraphs of hyperbolic drivel, it's actually a really good article. | ||
|
mishimaBeef
Canada2259 Posts
On January 25 2019 16:02 cha0 wrote: Not sure we watched the same games if you think any human could do the stalker and disruptor micro the AI was pulling. It was literally microing 6+ units at a time and had its apm spiking to 1500 sometimes. hmm, but if i recall, they said the reaction time was something over 300 ms and the apm graph shows this: https://imgur.com/a/PzNCXjs hmm but i guess currently it's true that the precision in its attempts to execute commands is perfect (i.e. no misclicks) would be cool to see a camera observing the screen and robotic "hand" (x-y movement) operating a mouse using control systems theory | ||
|
Deleted User 3420
24492 Posts
yes it's not fair, that's the point, that's why an AI can exceed humans. duh | ||
|
Ej_
47656 Posts
On January 25 2019 22:17 travis wrote: lol @ people complaining about an AI doing the things an AI can do yes it's not fair, that's the point, that's why an AI can exceed humans. duh Its not. The point is to make AI that can compete with humans execution... | ||
|
Deleted User 3420
24492 Posts
On January 25 2019 22:23 Ej_ wrote: Its not. The point is to make AI that can compete with humans execution... Where are you getting that from? You can read DeepMind's mission statement here: https://deepmind.com/about/ The APM limitation was just to make it more fair. If they wanted to make it imitate human execution they would need to use a robotic hand, (and i guess a camera, too). | ||
|
Poopi
France12836 Posts
On January 25 2019 22:17 travis wrote: lol @ people complaining about an AI doing the things an AI can do yes it's not fair, that's the point, that's why an AI can exceed humans. duh It's not a battle of intelligence if you just outplay hard your opponent mechanically. Them trying to make it a bit fair is because they understand otherwise it would not mean much. So far they have made good progress compared to previous gaming AI but it's still far from robust, as we have seen. | ||
|
mishimaBeef
Canada2259 Posts
| ||
|
Haukinger
Germany131 Posts
On January 25 2019 22:40 Poopi wrote: It's not a battle of intelligence if you just outplay hard your opponent mechanically. It's a battle of Starcraft. That a game not about intelligence, but mechanical execution. If else, there would be APM-limits in the game like minimal cooldowns on all commands. | ||
|
Acrofales
Spain17982 Posts
On January 25 2019 22:36 travis wrote: Where are you getting that from? You can read DeepMind's mission statement here: https://deepmind.com/about/ The APM limitation was just to make it more fair. If they wanted to make it imitate human execution they would need to use a robotic hand. That's kinda the point being made though. We can make a machine that can do surgery with far more precision than a human could ever hope to achieve, but we still need the medic controlling it, because the medic knows *what* to do, even if the robot is far better at actually doing it. Creating a bot that has absolutely perfect control over the units and beats humans based (mostly) on that control, is not "solving intelligence", it is using a superhuman unit control to win. Now I don't know what inputs AlphaStar receives, whether it's the videofeed and it needs to process that, or whether it's a list of everything that is happening in the game in some kind of symbolic format (e.g. an XML of all positions of all units it can see, all structures, what they are producing, how far along they are, etc. etc.). But clearly to make the competition "fair" in a real-time game, you cannot treat it in the same way as a turn-based game like chess or go: you need to factor in real-time limitations of humans, which includes things like screen lag and time to physically move an arm which moves the mouse, as well as the associated imprecision in both processes. Now if you say "computer vision is a hard problem that we don't want to deal with", and "robotics is a hard problem that we don't want to deal with", that's fine, and I agree that that is not really the point of a Starcraft bot. But then the simulation scenario should add in an approximately human level latency and add random noise to "clicks" that approximates human error. A start was made by removing the "zoom out hack", which I don't believe really should affect much, because superhuman minimap awareness + superhuman speed at clicking there and back on the minimap when a red blip appears is about the same, however in Mana's game with the DT, the superhuman map awareness almost certainly allowed AlphaStar to build an observer in time, whereas a human would *probably* only have noticed them by the time they were in his base, attacking shit. Consider the difference in reaction to DTs to the presence of an observer in the AI's base for the entire bloody game in the live showmatch without the zoom hack Now I may sound negative about AlphaStar. I am not. I think it is mindblowing what they managed to achieve, and I think it's time I start looking at LSTMs myself to see if I can apply them (in a far more limited scale) to some of my research questions. I fear I don't have the data available to train them (my main critic of deep learning approaches in general). And I also wish I had a casual 16 tensorflow units standing by for me But AlphaStar was far ahead of what I was expecting from an SC2 bot. I really thought the presentation would be some very limited scenarios, not full game play level (even if only PvP and on a limited map pool). It is truly impressive what they accomplished in at most 2 years of work on this.I also really really liked their visualizations of what the network was "thinking". Very very cool stuff. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 22:56 Acrofales wrote: That's kinda the point being made though. We can make a machine that can do surgery with far more precision than a human could ever hope to achieve, but we still need the medic controlling it, because the medic knows *what* to do, even if the robot is far better at actually doing it. Creating a bot that has absolutely perfect control over the units and beats humans based (mostly) on that control, is not "solving intelligence", it is using a superhuman unit control to win. Now I don't know what inputs AlphaStar receives, whether it's the videofeed and it needs to process that, or whether it's a list of everything that is happening in the game in some kind of symbolic format (e.g. an XML of all positions of all units it can see, all structures, what they are producing, how far along they are, etc. etc.). But clearly to make the competition "fair" in a real-time game, you cannot treat it in the same way as a turn-based game like chess or go: you need to factor in real-time limitations of humans, which includes things like screen lag and time to physically move an arm which moves the mouse, as well as the associated imprecision in both processes. Now if you say "computer vision is a hard problem that we don't want to deal with", and "robotics is a hard problem that we don't want to deal with", that's fine, and I agree that that is not really the point of a Starcraft bot. But then the simulation scenario should add in an approximately human level latency and add random noise to "clicks" that approximates human error. A start was made by removing the "zoom out hack", which I don't believe really should affect much, because superhuman minimap awareness + superhuman speed at clicking there and back on the minimap when a red blip appears is about the same, however in Mana's game with the DT, the superhuman map awareness almost certainly allowed AlphaStar to build an observer in time, whereas a human would *probably* only have noticed them by the time they were in his base, attacking shit. Consider the difference in reaction to DTs to the presence of an observer in the AI's base for the entire bloody game in the live showmatch without the zoom hack Now I may sound negative about AlphaStar. I am not. I think it is mindblowing what they managed to achieve, and I think it's time I start looking at LSTMs myself to see if I can apply them (in a far more limited scale) to some of my research questions. I fear I don't have the data available to train them (my main critic of deep learning approaches in general). And I also wish I had a casual 16 tensorflow units standing by for me But AlphaStar was far ahead of what I was expecting from an SC2 bot. I really thought the presentation would be some very limited scenarios, not full game play level (even if only PvP and on a limited map pool). It is truly impressive what they accomplished in at most 2 years of work on this.I also really really liked their visualizations of what the network was "thinking". Very very cool stuff. I take it shortly 1) Don't project human emotions and ways of work on machines. That doesn't work, that's why they're machines and not humans(duh) 2) AlphaStar knew. YOur surgeon example is wrong, machines whcih make decisions based on machine learning experience know. It's not human knowing per se, it's machine knowing, but they know. 3) Because machines don't have emotions and they have bigger experience(also faster thinking) they can tell which fights are worth it and which aren't on more precise scale. Similarly a pro can tell this while you wouldn't be able to tell it on the same scale. And because they don't have the emotions they don't fear of losing, because they don't fear of losing they are doing humanly insane things(e.g. the ramp things). (just imagine what some pros would be able to do without the fear of losing the units which is always there, even if it's on the background with low priority(to use machine terms ))To me all the threads about the games are full of big misunderstanding how such machine operates and projections of humanism on machines. | ||
|
Haukinger
Germany131 Posts
On January 25 2019 22:56 Acrofales wrote: Creating a bot that has absolutely perfect control over the units and beats humans based (mostly) on that control, is not "solving intelligence", it is using a superhuman unit control to win. That's like asking it to win with only Zealots... if it's in the game, the AI may use it. Only winning matters, if it wins by controlling stalkers individually with 10000apm, it's a win nevertheless. | ||
|
counting
11 Posts
On January 25 2019 22:56 Acrofales wrote: Now I don't know what inputs AlphaStar receives, whether it's the videofeed and it needs to process that, or whether it's a list of everything that is happening in the game in some kind of symbolic format (e.g. an XML of all positions of all units it can see, all structures, what they are producing, how far along they are, etc. etc.). But clearly to make the competition "fair" in a real-time game, you cannot treat it in the same way as a turn-based game like chess or go: you need to factor in real-time limitations of humans, which includes things like screen lag and time to physically move an arm which moves the mouse, as well as the associated imprecision in both processes. From their blog post and video showing the behind the scene videos, AlphaStar certainly used a modified version of PySC2 API as the input and output interface. deepmind.com There is a AMA thread on reddit MachineLearning subreddit If we really want to know the details, maybe we can get some details soon. On January 25 2019 22:56 Acrofales wrote: Now I may sound negative about AlphaStar. I am not. I think it is mindblowing what they managed to achieve, and I think it's time I start looking at LSTMs myself to see if I can apply them (in a far more limited scale) to some of my research questions. I fear I don't have the data available to train them (my main critic of deep learning approaches in general). And I also wish I had a casual 16 tensorflow units standing by for me But AlphaStar was far ahead of what I was expecting from an SC2 bot. I really thought the presentation would be some very limited scenarios, not full game play level (even if only PvP and on a limited map pool). It is truly impressive what they accomplished in at most 2 years of work on this.I also really really liked their visualizations of what the network was "thinking". Very very cool stuff. LSTM is interesting and powerful, but not a silver bullet, and it is famously hard to train (most RNNs have the same problem). It is essentially just a time series pattern recognition mechanism (can be used as a sequence generator). But it is just one component (albeit a crucial one), You will need many more components to solve complex problems. Like if you use an Action-Critic system, the LSTM will most likely be the action/sequence generator, and you still need a critic/evaluation network/system for it to function. | ||
|
Acrofales
Spain17982 Posts
On January 26 2019 00:26 Haukinger wrote: That's like asking it to win with only Zealots... if it's in the game, the AI may use it. Only winning matters, if it wins by controlling stalkers individually with 10000apm, it's a win nevertheless. Not really, I'm saying it should play SC2, not some heavily modified version that you can play by "plugging electrodes into your brain and thinking about exactly where you want each individual stalker to blink to and they do that instantly". SC2 is still a computer game for humans, which means you have to look at the screen, process the screen and then move the mouse and keyboard to perform actions. | ||
|
Acrofales
Spain17982 Posts
On January 26 2019 00:35 counting wrote: LSTM is interesting and powerful, but not a silver bullet, and it is famously hard to train (most RNNs have the same problem). It is essentially just a time series pattern recognition mechanism (can be used as a sequence generator). But it is just one component (albeit a crucial one), You will need many more components to solve complex problems. Like if you use an Action-Critic system, the LSTM will most likely be the action/sequence generator, and you still need a critic/evaluation network/system for it to function. I know. I just happen to be bumping into a rather hard time series pattern recognition problem. Unfortunately, my first problem is the dataset itself. It's small, and unlabelled at small scales: I have labels for hour-long sequences, and then I *definitely* don't have enough samples, and never will, so I need to look at smaller windows, and then they're unlabelled. So that'd mean labelling them manually, and I don't have time nor money at the moment to do so, but if they accept my newest grant proposal, I'll definitely consider this. We'll see For now, I'll stick with manual feature engineering from the whole sequence and using random forests, which is giving decent results at a more macro level | ||
|
Charoisaur
Germany15942 Posts
On January 26 2019 00:26 Haukinger wrote: That's like asking it to win with only Zealots... if it's in the game, the AI may use it. Only winning matters, if it wins by controlling stalkers individually with 10000apm, it's a win nevertheless. Of course if it does that it's technically better than humans at sc2. But that misses the point of this competition, the point of this is to show that the AI is smarter than humans, we already know that it's physically far superior. And comparing the strategic thinking/decision-making of humans vs AI can only be done if the AI's physical capabilities are limited. | ||
|
niteReloaded
Croatia5281 Posts
On January 25 2019 05:10 renaissanceMAN wrote: fucking bot LOVES stalkers holy cow haha :'D | ||
|
Polypoetes
20 Posts
On January 25 2019 22:40 Poopi wrote: It's not a battle of intelligence if you just outplay hard your opponent mechanically. Them trying to make it a bit fair is because they understand otherwise it would not mean much. So far they have made good progress compared to previous gaming AI but it's still far from robust, as we have seen. Weren't you the guy that claimed just a year ago that no AI/Bot/NN would be able to play Starcraft? Playing Starcraft isn't a test of intelligence. It is a test of playing and winning. I don't get how people expect a neural network to optimize winning at Starcraft to also have an eerie ability to (seemingly) read the minds of people. On January 26 2019 03:32 Charoisaur wrote: Of course if it does that it's technically better than humans at sc2. But that misses the point of this competition, the point of this is to show that the AI is smarter than humans, we already know that it's physically far superior. And comparing the strategic thinking/decision-making of humans vs AI can only be done if the AI's physical capabilities are limited. No. If you think an AI that plays Starcraft isn't 'interesting' to you because it turns out the best way to play SC2 is to mass stalkers and blink micro like crazy, then that means that you don't really like SC2 the way it is meant to be played. So you try a different more interesting game. Getting a finely tuned unit composition and moving across the map trying to outflank your opponent is objectively a stupid way to play if you can just mass stalkers, a1a2a3, micro like an AI, and win. But it seems people have some alternate axis of stupid-smart completely independent of winning-losing. Which is interesting, and very confusing, in itself. So is a bot that plays 'like a human' but plays weaker, smarter than a bot that plays 1-dimentionally, 'like an AI', but players very strong? BTW, we already know humans are smarter than AI because humans create and use AIs. AIs don't create humans and use them for their purposes. This is a silly line to even do down. | ||
|
Poopi
France12836 Posts
On January 27 2019 05:41 Polypoetes wrote: Weren't you the guy that claimed just a year ago that no AI/Bot/NN would be able to play Starcraft? Playing Starcraft isn't a test of intelligence. It is a test of playing and winning. I don't get how people expect a neural network to optimize winning at Starcraft to also have an eerie ability to (seemingly) read the minds of people. No. If you think an AI that plays Starcraft isn't 'interesting' to you because it turns out the best way to play SC2 is to mass stalkers and blink micro like crazy, then that means that you don't really like SC2 the way it is meant to be played. So you try a different more interesting game. Getting a finely tuned unit composition and moving across the map trying to outflank your opponent is objectively a stupid way to play if you can just mass stalkers, a1a2a3, micro like an AI, and win. But it seems people have some alternate axis of stupid-smart completely independent of winning-losing. Which is interesting, and very confusing, in itself. So is a bot that plays 'like a human' but plays weaker, smarter than a bot that plays 1-dimentionally, 'like an AI', but players very strong? I'm the guy that said it's basically impossible to have a completely fair fight between an AI and a human, because of the heavy emphasis on mechanics in starcraft 2, and that's hard to make things fair in that department. And the goal of deepmind is general AI so it's a small step towards that but far from enough, that's why fair things regarding mechanics is important if you want a robust and adaptable AI. | ||
|
Polypoetes
20 Posts
I don't know what to say to you that can convince you now that there is no such a thing, by definition, of a 'fair' competition between an AI and a human. What would that even entail? You simulate a human brain inside a computer so that you know that genetically/biochemically, such a human as the one you simulate could theoretically exist? An AI is an AI and a human is a human. The question is if we humans can create AI to do tasks. And games are nice benchmarks because they are well-defined problems. There is plenty of test data and it is easy to come up with a cost/objective function. What is next? People here arguing that an AI should miss-click like a human would? Rage and get 'emotional' after being cheesed? Get 'nervous' for important matches? Actually, those may be interesting AI challenges down the road for AI's that should be able to engage socially with humans better than humans are. But right now when the question is if AIs can beat humans in RTS games, that seems silly. I guess people who were convinced that AIs wouldn't be able to play RTS have to move the goalpost somewhere to keep their peace of mind. You want an SC2-based Turing test? As for SC2. It turns out mechanics is at the core of the game. Who would have thought! That is why many of us knew that AI would have good chances of taking games off humans. Exactly because many of these very human soft and subtle skills aren't that important. | ||
|
Charoisaur
Germany15942 Posts
On January 27 2019 05:41 Polypoetes wrote: Weren't you the guy that claimed just a year ago that no AI/Bot/NN would be able to play Starcraft? Playing Starcraft isn't a test of intelligence. It is a test of playing and winning. I don't get how people expect a neural network to optimize winning at Starcraft to also have an eerie ability to (seemingly) read the minds of people. No. If you think an AI that plays Starcraft isn't 'interesting' to you because it turns out the best way to play SC2 is to mass stalkers and blink micro like crazy, then that means that you don't really like SC2 the way it is meant to be played. So you try a different more interesting game. Getting a finely tuned unit composition and moving across the map trying to outflank your opponent is objectively a stupid way to play if you can just mass stalkers, a1a2a3, micro like an AI, and win. But it seems people have some alternate axis of stupid-smart completely independent of winning-losing. Which is interesting, and very confusing, in itself. So is a bot that plays 'like a human' but plays weaker, smarter than a bot that plays 1-dimentionally, 'like an AI', but players very strong? BTW, we already know humans are smarter than AI because humans create and use AIs. AIs don't create humans and use them for their purposes. This is a silly line to even do down. that only means massing blinkstalkers is the smartest way to play if you have the physical capabilities of an AI. It doesn't mean at all that the AI is better at strategizing because a human might do the same if it had the physical capabilities of an AI. If it would only be about beating humans no matter how there wouldn't be the need for the Deepmind team to work on it. Just a standard micro bot executing a basic strategy would be enough. Also a new 1 post user that memorizes what Poopi said 1 year ago................ Hmmmmmm | ||
|
The_Red_Viper
19533 Posts
On January 27 2019 06:22 Charoisaur wrote: that only means massing blinkstalkers is the smartest way to play if you have the physical capabilities of an AI. It doesn't mean at all that the AI is better at strategizing because a human might do the same if it had the physical capabilities of an AI. Also a new 1 post user that memorizes what Poopi said 1 year ago................ Hmmmmmm That is like saying AI's should only be able to calculate X moves in whatever timeframe for chess or go. People do not realize that the point isn't to have an AI which simulates human abilities in everything but Y, the point is to have an AI which is able to learn a task and be more efficient/better at it than the human counter part. Now it is somewhat interesting to bring down the AI's abilities in apm and whatever else you think is important mostly because of the PR and it being more challenging for the developers (that means they learn more about possible other uses), not because it has to be "fair". The interesting part is that you have an AI which learns new tasks without any hardcoded rules, that is what's fascinating. | ||
|
Polypoetes
20 Posts
But how is that not strategizing? The AI has AI mechanics. The proper strategizing therefore is to use blink stlaker with superior AI mechanics. Apparently, it is better at strategizing than you are, because you don't seem to get that. You act as if AIs could beat humans at RTS for years using just some build-in micro. Have you ever considered what kind of code is needed for an AI to outplay a human in a micro battle? And have you considered how hard it is to make an AI that doesn't get stuck or exploited easily by a human? Their strong AI has a 100% winrate vs these players. I don't know how people can be so stubborn and say that the AI doesn't 'think', doesn't 'known', isn't 'intelligent', doesn't 'strategize'. Those are all anthropomorphization of what machine learning is. You optimize it to win games. That's what they did. They did very well. And they have shown the potential to do it even better. In principle, you could also train a neural net to play indistinguishable from a human, passing a SC2 Turing test. But that requires a human agent. You cannot have 200 years of human lifetimes of humans objectively judging if an AI is more human-like than some other iteration. | ||
|
lolfail9001
Russian Federation40190 Posts
| ||
|
ETisME
12387 Posts
On January 27 2019 06:38 lolfail9001 wrote: I mean, i don't get the argument going on. The ultimate point is that AI did figure out a proper decision making in AI terms, but both this decision making was not very satisfying from human's PoV.... nor could be it be in conditions AI was granted. you summed it up perfectly. it's never about winning or losing, the goal isn't to win the game but to learn and solve complex situations with AI. The wins were with too much of mechanics advantage (raw input) and the loss was with player limited vision. | ||
|
figq
12519 Posts
The key here is that the AI trained with itself - AI vs AI in ~200 human years of training. That doesn't give it realistic perspective on what works best vs humans. If the AI was let to compete on ladder vs real humans, I bet it would focus exclusively on things like blink stalkers all the time - not just in some random instances - that give it inherent mechanical advantage no human can match. | ||
|
TaKeTV
Germany1199 Posts
On January 27 2019 21:25 figq wrote: To remind you that we've seen various different agents who trained together, not one entity. Some of them focused on blink stalkers and won by inhumanly good micro. Those seemed like easy victories. But really we've seen a great variety of approaches - aggressive, defensive, tactical and strategic. The key here is that the AI trained with itself - AI vs AI in ~200 human years of training. That doesn't give it realistic perspective on what works best vs humans. If the AI was let to compete on ladder vs real humans, I bet it would focus exclusively on things like blink stalkers all the time - not just in some random instances - that give it inherent mechanical advantage no human can match. On the one hand it obviously would be smart and the logical thing to do: exploit your own strength and abilities - aka unmatched mechanics. On the other hand that really wouldn't give a realistic approach of an Ai actually playing against a human. Humans need to physically move the mouse against friction, the distance etc. Also observation and reaction play a role as well. I find it interesting but what we saw is far away from actually an Ai playing / outplaying humans but rather a mechanically superior agent. And the biggest task will be learning to play the game with no knowledge of the map, actually having to fight for information and telling the difference between a fake, a human mistake etc. | ||
|
Polypoetes
20 Posts
On January 27 2019 21:25 figq wrote: If the AI was let to compete on ladder vs real humans, I bet it would focus exclusively on things like blink stalkers all the time - not just in some random instances - that give it inherent mechanical advantage no human can match. But it is not really possible to train the AI on the ladder. How do you get 200 years of game time worth of your bot playing vs humans on the ladder? Adjusting the weights of the NN based on how well it does vs humans is just really tricky because you will have a much smaller dataset. And the dataset could have all kinds of biases. And that even ignores the fact that people could be trolling. We have seen chat NNs say racist things because of people trolling. Secondly, even if you can train a NN to exploit mistakes humans make, maybe that is actually an inferior way to play. The NN will make deliberate mistakes 'knowing' that humans cannot exploit it. But another bot will. So you will just be muddying the way the NN plays with weakness. And third, things the NN does that seem like a mistake, once it gets so strong it beats humans, it is not clear we humans can recognize a mistake. Our criticism of what it does is only justified if we can show that we can exploit it as a weakness. Humans cut corners too. Maybe it knows it can cut corners. If the AI builds 7 observers vs Mana, maybe it does that because it is minimizing the risk to lose. And it judged the risk of losing by being outmacroed many magnitudes lower than being killed by DTs. So it just overproduced observers and then puts them all near it's biggest army. So while it seems like a mistake, and it probably is, humans stepping in and manually adjusting weights doesn't work because you have a very poor idea of what you are doing. | ||
|
Haukinger
Germany131 Posts
| ||
|
snakeeyez
United States1231 Posts
On January 27 2019 21:46 TaKeTV wrote: On the one hand it obviously would be smart and the logical thing to do: exploit your own strength and abilities - aka unmatched mechanics. On the other hand that really wouldn't give a realistic approach of an Ai actually playing against a human. Humans need to physically move the mouse against friction, the distance etc. Also observation and reaction play a role as well. I find it interesting but what we saw is far away from actually an Ai playing / outplaying humans but rather a mechanically superior agent. And the biggest task will be learning to play the game with no knowledge of the map, actually having to fight for information and telling the difference between a fake, a human mistake etc. I feel like it already showed some of this. It could see the entire map unlike a human, but it never had any different vision. It had to fight for and scout all of its vision just a person in all the games. It just didnt need the camera which is a big advantage. I think they still have a long ways to go though they need to play more matches with a limited camera, and there is a large variety of strats in this compared to chess. When it plays the best players its going to have a tough time if its limited in its micro to be more like a human. That is the point of this though overcoming these challenges. If it won with this demonstration easily then it was not much of a challenge to begin with. The fact it still has a ways to go is why they choose to try to win at starcraft 2. its probably one of the hardest games left to beat. The other was GO and they beat that which was amazing. | ||
|
Jasper_Ty
101 Posts
| ||
|
niteReloaded
Croatia5281 Posts
On January 27 2019 23:14 Haukinger wrote: It's funny how people deem a korean terran exploiting marines with superior mechanical skill the highest level of play but if an AI uses blink stalkers it's just stupid with high apm. It's not funny. Super, human micro requires very rare talent and huge amounts of practice. Super, AI micro requires coding it without APM limits. Strategy, metagame, balance are all built on the capacity and potential of individual units. Unit's utility is based on what a player can do with it. What a human player can do with a unit is vastly different from what a AI can do with it. Based on that, AI uses different tactics, different strategies, almost a different game. A human can't click 1500 in a minute, AlphaStar did. StarCraft is a realtime strategy game where speed is very important. They made a bot which has, at crucial times, played at least 2X faster than any human can. It's like building a sports playing bot which can run 2x faster. Suddenly, all tactics and strategies are affected and the game is almost not the same. --- DeepMind's only mistake is that they didn't put a proper peak-APM limit. Frankly, I would love to see games where AlphaStar is limited to 50 peak APM, then more games where the limit is 100 peak apm, then 150 etc. Would love to see the strategies and tactics used. | ||
|
Charoisaur
Germany15942 Posts
On January 27 2019 23:14 Haukinger wrote: It's funny how people deem a korean terran exploiting marines with superior mechanical skill the highest level of play but if an AI uses blink stalkers it's just stupid with high apm. it's not stupid, it's just worthless if we want to know how good the AI is strategically if it has unlimited apm because then it can win with anything. If the goal would be to just beat a human no matter how there would be no need for the Deepmind team to take this on, a regular micro bot would be enough. | ||
|
Xamo
Spain877 Posts
| ||
|
Polypoetes
20 Posts
I understand it is not interesting for you personally to improve your game. But we have had this discussion in the community ever since AlphaGo how long it would take for an AI to play properly and to beat top humans. I remember discussing with programmers who talked about how difficult of a problem it is to simulate combat before they make a move. Because that is what they tried to put into the BWAPI AIs. You see a game state, you see your units, you see the opponent units, you predict how the units will move and what they will attack, you calculate how much each side will lose, and then you know if this engagement is favorable. So when people say that they consider the AI microing like crazy 'trivial', I just have to laugh. Personally, I would like to see an unlimited APM AI. And I would love to see an AI play SC BW, because it seems to me that SC2 is way more straightforward and SC BW is way more on a knife edge and subtle. Maybe it is me being a former SC BW player, but I have this feeling that any play by an AI is way more exploitable in SC BW than in SC2. Honestly, this reminds me of the debate we had back when SC2 was announced and when we had this influx of AoE, Civ, and C&C players arguing that slowing down the game, or making the interface easier, would give the game a richer game of strategy. The game is not what you think it is. Imagine an AI figuring out how to play Civ3. It is just calculating which builds to build first under which circumstances. It is boring. There is one solution. You calculate what it is, and you just carry it out. Let's not forget that unlike Go or Chess, RTS games are convergent towards the end. | ||
|
Kafka777
361 Posts
| ||
|
fededevi
Italy45 Posts
A simple APM limit is not gonna do the trick, sometimes a player can have huge "useful" apm spikes so you will need a model that account for that and many other things. If you limit the AI too much, the AI will have to find workarounds to it's sub optimal micro by finding strategies that are sub-optimal for a player. If you do not limit the AI enough, the AI will come up with strategies that players will never be able to execute. Not to mention that each human player is different. Still it would be very interesting to see if the AI can come up with good/decent "very-low-micro" strategies. | ||
|
mierin
United States4943 Posts
| ||
|
fededevi
Italy45 Posts
But we do not know if that strategy is actually good for human players too because the AI is playing with different limitations. If over-probing was done by all agents on different APM limits then it would be an indication that it is a good idea to always build more probes. With "strategy" I mean the whole game strategy, you can't take out a single thing (like over-probing) and discard the rest, because everything is connected in a single game. For example the agent could be so good at winning with stalkers micro, that the only way to beat it is somehow to kill a lot of its probes. In this condition the agent would improve its chance of winning having a less than ideal number of probes. It would still have enough stalkers to win the engagements thanks to its god-like micro, but he would not lose to other kind of attacks to the mineral line. But if he was not so good at microing then it would have better chances by expanding its economy faster. This is just an example, I don't know if this is the case. | ||
|
Grumbels
Netherlands7031 Posts
This is a common “weakness” to AI’s: a complete inability to take the psychology of the opponent into consideration. Agents tend to converge to the same cautious, conservative style which shows maximum respect to the opponent. E.g. if you had to play a match versus an AI, it might be the case that if you just made blink stalkers it would vastly overestimate the strength of your army, because it doesn’t know you can’t actually micro blink stalkers at 1500 APM. And as a result it wouldn’t take any engagements. This is why chess engines have a “contempt” setting programmed into them, which forces them to make sub-optimal moves which nevertheless increase winning chances against weaker opponents. This is a must in any sort of tournament or league play in chess, since it avoids draws. I don’t know how you would program “contempt” into AlphaStar, other than training it versus agents with handicapped APM, such that it would develop more confidence in its micro, in order to better approach match conditions vs humans. But of course this doesn’t have any scientific value, only competitive utility. | ||
|
niteReloaded
Croatia5281 Posts
On January 29 2019 05:38 Grumbels wrote: An agent isn’t confident in its stalker micro, as it’s training vs bots, not humans. It won’t think it can win a four vs five stalker fight vs a human player because it is used to losing to other agents in such situations. This is a common “weakness” to AI’s: a complete inability to take the psychology of the opponent into consideration. Agents tend to converge to the same cautious, conservative style which shows maximum respect to the opponent. E.g. if you had to play a match versus an AI, it might be the case that if you just made blink stalkers it would vastly overestimate the strength of your army, because it doesn’t know you can’t actually micro blink stalkers at 1500 APM. And as a result it wouldn’t take any engagements. This is why chess engines have a “contempt” setting programmed into them, which forces them to make sub-optimal moves which nevertheless increase winning chances against weaker opponents. This is a must in any sort of tournament or league play in chess, since it avoids draws. I don’t know how you would program “contempt” into AlphaStar, other than training it versus agents with handicapped APM, such that it would develop more confidence in its micro, in order to better approach match conditions vs humans. But of course this doesn’t have any scientific value, only competitive utility. You release it on the ladder, with 1000s of instances playing vs humans 24/7. Machine learning and genetic algorithms (or whatnot) would cause bots to start taking some smaller sub optimal moves, just to gauge the reactions of the opponent. As the bots have played just other bots with same 'mechanical' capabilities, the ability to gauge opponent's mechanical skill wasn't necessary. | ||
|
Cyro
United Kingdom20285 Posts
You release it on the ladder, with 1000s of instances playing vs humans 24/7. There are not even that many humans on the ladder | ||
|
Deimos
Mexico134 Posts
| ||
|
Acrofales
Spain17982 Posts
On January 30 2019 09:11 Cyro wrote: There are not even that many humans on the ladder Especially not at GM level. Pretty sure it's already good enough that even knowing its weaknesses it'll still beat diamond players on mechanics alone. | ||
|
Plopus
Switzerland112 Posts
| ||
|
sneakyfox
8216 Posts
https://starcraft2.com/en-us/profile/3/1/6155907/ladders?ladderId=71017 | ||
|
GreenHorizons
United States23221 Posts
Seriously though, I think the confidence TLO had going in is totally humanity. The outcome prediction seems like it could be a useful/interesting tool for casters/analysts, especially for replays. | ||
|
MadMod
Norway4 Posts
On February 01 2019 20:15 sneakyfox wrote: It seems AlphaStar is now 2nd on the Korean ladder: https://starcraft2.com/en-us/profile/3/1/6155907/ladders?ladderId=71017 This is really awesome, if this is real and they keep it up. Kudos to DeepMind! I hope that those that run into it post their replays. | ||
|
Excalibur_Z
United States12235 Posts
On February 01 2019 20:15 sneakyfox wrote: It seems AlphaStar is now 2nd on the Korean ladder: https://starcraft2.com/en-us/profile/3/1/6155907/ladders?ladderId=71017 Not real. Check the timestamps and types of achievements that account has earned. | ||
|
Charoisaur
Germany15942 Posts
On February 01 2019 20:15 sneakyfox wrote: It seems AlphaStar is now 2nd on the Korean ladder: https://starcraft2.com/en-us/profile/3/1/6155907/ladders?ladderId=71017 I've played already against 3 "Alphastars" on the ladder. It just seems to be the new Barcode. | ||
|
Deleted User 3420
24492 Posts
| ||
| ||
