JulyZerg once said: "If one could perform perfect micro, Zerg would be the strongest race"
Proof to July's words?
I'd like to see it's Muta micro. And how it copes with unusual strategies.
On January 20 2011 02:23 Torte de Lini wrote: Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
'Game Over: Kasparov and the Machine' was one, though probably not the only one:
On January 20 2011 02:23 Torte de Lini wrote: Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
'Game Over: Kasparov and the Machine' was one, though probably not the only one:
Is that Sucker? He was really godd at his time, but i don't think he played much the last years. Can't really think of other good terrans from spain right now. So i kinda doubt the top16 europe^^
On January 20 2011 02:32 IPS.ZeRo wrote: Is that Sucker? He was really godd at his time, but i don't think he played much the last years. Can't really think of other good terrans from spain right now. So i kinda doubt the top16 europe^^
it could also been another player like Zelotito offracing
On January 20 2011 02:36 Jibba wrote: An impressive bot, no doubt, but hugely misleading title. Especially when we're talking about BW and the thresh hold of "pro gamer" is a lot higher.
The arstechnica article is pretty awesome though.
Hmmm... well hes pro for the foreigner scene.... so its not mis-leading at all.. or are you saying no foriegner's are not pro enough? thanks.
On January 20 2011 02:36 Jibba wrote: An impressive bot, no doubt, but hugely misleading title. Especially when we're talking about BW and the thresh hold of "pro gamer" is a lot higher.
The arstechnica article is pretty awesome though.
Hmmm... well hes pro for the foreigner scene.... so its not mis-leading at all.. or are you saying no foriegner's are not pro enough? thanks.
Progamer has a specific meaning, the only progamers in the foreign scene were Draco, Ret and Nony briefly and Idra.
On January 20 2011 02:37 KwarK wrote: I really doubt it beat anyone notable. Title would be more accurate if it said "computer successfully executes a basic muta rush build order".
But it's not just a muta rush build order. If you read the article it's a lot more sophisticated than that. It adapts itself to early rushes and does not follow a fixed build at all. More importantly, the muta micro is far more advanced than any human player could execute. The whole muta flock either clumps or spreads in response to different threats.
This AI uses a technique that is out of reach of humans. It is interesting to note that the AI that came second also uses a technique that humans would never be able to execute: SCVs are used throughout the battlefield to repair tanks and goliaths.
I could easily see AI one day beating 'real' progamers because they are not constrained by speed.
On January 20 2011 02:36 Jibba wrote: An impressive bot, no doubt, but hugely misleading title. Especially when we're talking about BW and the thresh hold of "pro gamer" is a lot higher.
The arstechnica article is pretty awesome though.
Hmmm... well hes pro for the foreigner scene.... so its not mis-leading at all.. or are you saying no foriegner's are not pro enough? thanks.
Progamer has a fairly specific meaning in the BW community--that is, a player that has a Professional license issued by Kespa.
On January 20 2011 02:36 Jibba wrote: An impressive bot, no doubt, but hugely misleading title. Especially when we're talking about BW and the thresh hold of "pro gamer" is a lot higher.
The arstechnica article is pretty awesome though.
Hmmm... well hes pro for the foreigner scene.... so its not mis-leading at all.. or are you saying no foriegner's are not pro enough? thanks.
Getting to WCG doesn't make you a progamer. As Jaedong said of one of his WCG opponents, "it was like playing the computer."
The muta micro is pretty incredible though. I wonder if they could program it to probe rush with perfect micro to win games.
AI's always have holes in them, so it actually IS suprising that an AI can compete quite well in starcraft, though, I would like to see how it fares against more well-known progamers.
That's actually an amazing feat of programming, to make a computer that calculates cost effectiveness of kill options in real time, in reaction to what an opponent is doing. Most people would just program a threat priority list, and call it a day, but that can be abused hard. This computer actually goes "Hey, he might have five goliaths, but since he can't kill as many mutas as I can workers, it's cost effective to go in here and kill seven workers to lose two mutas." There's a reason that it takes a team of PhD's to program this. It's hard to do! I think they should bot an account on iCCup with this AI, and see how far it can go, to find more consistent results.
We have to move our hands to give input to starcraft. The computer only has to give those commands, which are obviously billions of times quicker than the fastest mouse hand.
Want to know something funny? It doesn't need to hotkey groups. It controls all the mutas individually.
It's an absolutely awesome and interesting article if you read it and the muta micro is amazing.
Read the article naysayers.
On January 20 2011 02:36 Jibba wrote: Does that mean the adjustments weren't made by the programmers, but the AI adapted on its own? o.o
Yup -- "Using StarCraft’s map editor, we built Valhalla for the Overmind, where it could repeatedly and automatically run through different combat scenarios. By running repeated trials in Valhalla and varying the potential field strengths, the agent learned the best combination of parameters for each kind of engagement." -- perfect micro through infinite trial and error.
So there's supposedly some AI out there that can beat some n00b nobody on really old maps using bad BO and simply superior micro... why was this published at all? Every incarnation of these articles praises it so highly like it just beat a progamer. How insulting.
Until this thing is released to the public so we can test it, there's really no way of calling it "good" or "intelligent" in any way. Sounds just like pre-programmed responses to stimuli. Not really a "thinking" process going on.
Effort often did similar splits. He often had less Mutas and did damage with Lings during the Muta battles and then saved most of his Mutas with these splits. They were pretty insane.
Until this thing is released to the public so we can test it, there's really no way of calling it "good" or "intelligent" in any way. Sounds just like pre-programmed responses to stimuli. Not really a "thinking" process going on.
There's two situations here. One: you read the article, but it all flew over your head. Two: you didn't read the article. It has nothing to do with being good. I'm sure any kespa sanctioned pro could stomp it into the ground. That's not the point. It's an AI for a game that teaches itself how to deal with situations. It's also not just going for mutas ASAP. It gauges the opponents BO, decides wether or not there's going to be an early push, and deals with the situation accordingly. While this is basic for a player, it's something that has ALWAYS plagued AI's, in any game. It's an intresting step forward.
I disagree with the article, hyping that it beat a progamer, though. That's more of a "Hey, read me, I'm intresting!" headline.
On January 20 2011 03:25 Diminotoor wrote: So there's supposedly some AI out there that can beat some n00b nobody on really old maps using bad BO and simply superior micro... why was this published at all? Every incarnation of these articles praises it so highly like it just beat a progamer. How insulting.
Until this thing is released to the public so we can test it, there's really no way of calling it "good" or "intelligent" in any way. Sounds just like pre-programmed responses to stimuli. Not really a "thinking" process going on.
You might want to read the article, it's pretty interesting how they figured out gas steal, ovie scouting, rush defense etc. The AI might only be able to do mass muta, but it has sickkkkk muta micro.
On January 20 2011 03:25 Diminotoor wrote: Sounds just like pre-programmed responses to stimuli. Not really a "thinking" process going on.
Nope, it doesn't sound like that. Read the article again please.
And btw I don't think that a lot of thinking process is going on while a hardcore player plays. You don't really have that much time to think while playing BW. You mostly play based on your experience and use learnt counters to strategies, use a bit of gambling there and there (calculated risks, lacking scouting information). This AI does the same while making its play based around mutalisks.
We have to move our hands to give input to starcraft. The computer only has to give those commands, which are obviously billions of times quicker than the fastest mouse hand.
Want to know something funny? It doesn't need to hotkey groups. It controls all the mutas individually.
Yeah I know.
That is funny btw lolololol. No hotkeys, so ridiculous.
=_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
Until this thing is released to the public so we can test it, there's really no way of calling it "good" or "intelligent" in any way. Sounds just like pre-programmed responses to stimuli. Not really a "thinking" process going on.
There's two situations here. One: you read the article, but it all flew over your head. Two: you didn't read the article. It has nothing to do with being good. I'm sure any kespa sanctioned pro could stomp it into the ground. That's not the point. It's an AI for a game that teaches itself how to deal with situations. It's also not just going for mutas ASAP. It gauges the opponents BO, decides wether or not there's going to be an early push, and deals with the situation accordingly. While this is basic for a player, it's something that has ALWAYS plagued AI's, in any game. It's an intresting step forward..
There's 2 situations here: one, you read my response and it all flew over your head (more likely), or two, you didn't read my response or just don't understand what any of the concepts I'm talking about were. It has everything to do with being "good" because they're claiming it provides challenge for good Starcraft players. A simple algorithm mixed with maphack that would provide input to whether or not a certain amount of units are being made or whether a certain amount of production structures have been made before a certain time. I majored in Computer Sciences, I can see exactly how this thing works on a level the vast majority of people can't; apparently including you. Bottom line, don't speak on things you know nothing about.
You might want to read the article, it's pretty interesting how they figured out gas steal, ovie scouting, rush defense etc. The AI might only be able to do mass muta, but it has sickkkkk muta micro.
I did. I can see the big picture rather than get caught up in all the details.
And btw I don't think that a lot of thinking process is going on while a hardcore player plays. You don't really have that much time to think while playing BW.
Considering I know KaL, Jaedong, and FlaSh personally as well as a large amount of B-team semi-pros, I'm pretty sure they'd be insulted by this blatantly misinformed undercutting of their strategic abilities.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
I gathered that he's no active pro anymore, they just found somebody who knows how to play RTS. I have no doubt that a korean A-teamer would run the overmind bot over ten out of ten times.
if u check the computer's apm , i can't recall where i saw it it's lie3,4 k , so........... u only need to implement some rushe's and chees
p.s as far as i know kasparow beat before that day a few times the computer , and they always the computer smarter and better, and after he lost , he comed back and beat the computer and they made alot of ties
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
I still don't get it ? the last time I check (from the 2010 AI competition thread) the bot cannot even beat a high D / D+ on iCCup, how the hell can it beat a WCG competitor now ? And who is that WCG competitor guy ?
In fact if those new bots are really good I will try to install the broodwarAPI and try them, since I'm really bored with the AIs from BWAI.
On January 20 2011 03:40 Diminotoor wrote: A simple algorithm mixed with maphack that would provide input to whether or not a certain amount of units are being made or whether a certain amount of production structures have been made before a certain time.
it doesn't have maphack.. it has to scout. and it does. but never mind..
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
irridate? lol
....whats wrong with irridate for mutas??? I see you don't play terran...
Irradiate and valks are useless coz AI doesn't stack muta. That is why this AI will be terribad as practice partner. It does things in its special way (mostly APM abuse), so you will never learn how to fight against real players. To beat this AI you need to utilize new strats and BOs which will be good against AI, not humans. Personally I don't see any points in making AI which doesn't resemble human play.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
irridate? lol
....whats wrong with irridate for mutas??? I see you don't play terran...
I think he means the computer can likely query which mutalisk has been irradiated and separate it way faster than a human. Also this AI doesn't group its mutalisks in a stack of 11 like a human does.
Anyways like I posted in the last topic about this AI, the video of the human WCG player showed he didn't know any modern builds and was flat out bad. I wouldn't be surprised if the regular AI beat him once in a while.
On January 20 2011 03:45 seRapH wrote: Would overmind respond to valkonic correctly? Because I could see them getting absolutely wrecked by that
This is a smart response, I pose the same question.
irridate? lol
If that man's questions didn't spark anything for your debate column other than this pointless retort, you don't belong in this conversation.
it doesn't have maphack.. it has to scout. and it does. but never mind..
Ok so make it scout and then what it sees, it adds to its databank and takes things like production structures and... you know what nevermind, you obviously have no idea what you're talking about and have no place in intelligent conversation.
...whats wrong with irridate for mutas??? I see you don't play terran...
On January 20 2011 03:49 kamikami wrote: I still don't get it ? the last time I check (from the 2010 AI competition thread) the bot cannot even beat a high D / D+ on iCCup, how the hell can it beat a WCG competitor now ? And who is that WCG competitor guy ?
In fact if those new bots are really good I will try to install the broodwarAPI and try them, since I'm really bored with the AIs from BWAI.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
irridate? lol
....whats wrong with irridate for mutas??? I see you don't play terran...
It insta splits the muta. Irradiate was good because it would do alot of damage to muta as you worked to remove the damaged one. This AI just takes out the damage one immediately and isn't even that clumped usually in the first place. Spending money on vessels to kill one muta every 75 energy just wouldn't be worth it I don't think.
Ok so make it scout and then what it sees, it adds to its databank and takes things like production structures and... you know what nevermind, you obviously have no idea what you're talking about and have no place in intelligent conversation.
This isn't how you argue, no need to personally attack the guy. This has the potential to be a pretty interesting discussion but people almost invariably respond to flames like this de-railing the whole topic.
Also, what Chef said, I kinda want to see a full game, as well as the entire game vs Oriol.
Exactly how do you want to counter mass mutalisks with irridate if the opponent never stacks them? It's clearly worth teching for it to put down 1 irridate on 1 mutalisk. Wow, you just killed one!
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
irridate? lol
....whats wrong with irridate for mutas??? I see you don't play terran...
It insta splits the muta. Irradiate was good because it would do alot of damage to muta as you worked to remove the damaged one. This AI just takes out the damage one immediately and isn't even that clumped usually in the first place. Spending money on vessels to kill one muta every 75 energy just wouldn't be worth it I don't think.
For the bajillionth time "What about its response to Valkonic"? Valkyries and MM and vessels. Stop trying to act like the killing components aren't there and its just irradiate vs mutas.
Exactly how do you want to counter mass mutalisks with irridate if the opponent never stacks them?
Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
I think the best counter would be something like marine medic where the movement of the marines would be similar to the AI controlled mutalisks' and the medics did smart healing calculating which marine needs more healing at the moment.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
Goliath negate Muta micro with their range. Search for a video that showcased the power of AI muta micro.
It could micro 36 mutalisks indidually to kill off unbelievable amount of archon, turning aside at the correct moment. Basicly, AI mutas are invinsible if you cannot outrange them.
Also, Valks die to scourge pretty easily. And the AI can split the mutas anc control them seperately perfectly.
About irradiate, look at above
On January 20 2011 04:01 Diminotoor wrote: Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
The irradiated muta would be separated immediately and sent to his doom.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
irridate? lol
....whats wrong with irridate for mutas??? I see you don't play terran...
It insta splits the muta. Irradiate was good because it would do alot of damage to muta as you worked to remove the damaged one. This AI just takes out the damage one immediately and isn't even that clumped usually in the first place. Spending money on vessels to kill one muta every 75 energy just wouldn't be worth it I don't think.
For the bajillionth time "What about its response to Valkonic"? Valkyries and MM and vessels. Stop trying to act like the killing components aren't there and its just irradiate vs mutas.
Exactly how do you want to counter mass mutalisks with irridate if the opponent never stacks them?
Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
Sure you can use irradiate to kill one muta at time. However I think the money going to vessels would be far better spent on more marines/medics/valks.
On January 20 2011 03:58 L_Master wrote: This isn't how you argue, no need to personally attack the guy. This has the potential to be a pretty interesting discussion but people almost invariably respond to flames like this de-railing the whole topic.
Also, what Chef said, I kinda want to see a full game, as well as the entire game vs Oriol.
This is how I argue on an online forum. I don't have to be nice because I have no tolerance for idiocy. If you can't understand basic concepts, you do not have a place to argue more in-depth or advanced ones. Simple as that. If I see something bad or wrong, I'm going to call it out for what it is. If being told the truth is "flaming", then the world doesn't flame enough. Also yeah wtf @ there being no video of the game... also have we figured out who this kid is yet?
Irradiate is pretty useless in this situation, valks would work as long as you kept them alive. MnM&valks should theoretically work as long as you target fire scourge. Of course if the AI is smart enough to tech switch then you're pretty much doomed
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
Goliath negate Muta micro with their range. Search for a video that showcased the power of AI muta micro.
It could micro 36 mutalisks indidually to kill off unbelievable amount of archon, turning aside at the correct moment. Basicly, AI mutas are invinsible if you cannot outrange them.
Also, Valks die to scourge pretty easily. And the AI can split the mutas anc control them seperately perfectly.
On January 20 2011 04:01 Diminotoor wrote: Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
The irradiated muta would be separated immediately and sent to his doom.
You act as if the initial splash damage and forcing a muta away is a bad thing? Valks die to scourge if you suck at positioning them and we'd have to see its response to that because its response to 3-gate zealot was make tons of zerglins and 4 sunkens except it put a sunken in its main....Don't get me wrong, its impressive, they essentially made a D-/D zerg out of a program. Thats cool, but to boast it beat a "progamer" when the kid was obviously not a progamer is flat out lying.
On January 20 2011 03:58 L_Master wrote: This isn't how you argue, no need to personally attack the guy. This has the potential to be a pretty interesting discussion but people almost invariably respond to flames like this de-railing the whole topic.
Also, what Chef said, I kinda want to see a full game, as well as the entire game vs Oriol.
This is how I argue on an online forum. I don't have to be nice because I have no tolerance for idiocy. If you can't understand basic concepts, you do not have a place to argue more in-depth or advanced ones. Simple as that. If I see something bad or wrong, I'm going to call it out for what it is. If being told the truth is "flaming", then the world doesn't flame enough. Also yeah wtf @ there being no video of the game... also have we figured out who this kid is yet?
You're just being overly aggressive to compensate for not reading the actual article, there's actually no reason to argue about it since your beef with the whole thing is that the thread title is misleading. The article wasn't bragging about how an AI can beat a person (it can't), but how complicated it is to teach a computer to problem-solve.
On January 20 2011 04:14 seRapH wrote: Irradiate is pretty useless in this situation, valks would work as long as you kept them alive. MnM&valks should theoretically work as long as you target fire scourge. Of course if the AI is smart enough to tech switch then you're pretty much doomed
I was under the impression that we were discussing a hypothetical countering AI for the muta AI.
You're just being overly aggressive to compensate for not reading the actual article
No, a lot of you people were apparently raised to be pussies who can't take feedback from somebody who sees a bigger picture than them. Its amazing how these discussions are almost always people trying to present a countering army composition and acting like "well this will never work because of this that and the other thing". What they completely forget about is the tactics and the utilization of the units that make the strategy effective.
Bottom line? There aren't any top players who would agree with your ways of thinking. Might want to change.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
Goliath negate Muta micro with their range. Search for a video that showcased the power of AI muta micro.
It could micro 36 mutalisks indidually to kill off unbelievable amount of archon, turning aside at the correct moment. Basicly, AI mutas are invinsible if you cannot outrange them.
Also, Valks die to scourge pretty easily. And the AI can split the mutas anc control them seperately perfectly.
About irradiate, look at above
On January 20 2011 04:01 Diminotoor wrote: Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
The irradiated muta would be separated immediately and sent to his doom.
You act as if the initial splash damage and forcing a muta away is a bad thing? Valks die to scourge if you suck at positioning them and we'd have to see its response to that because its response to 3-gate zealot was make tons of zerglins and 4 sunkens except it put a sunken in its main....Don't get me wrong, its impressive, they essentially made a D-/D zerg out of a program. Thats cool, but to boast it beat a "progamer" when the kid was obviously not a progamer is flat out lying.
Yeah, the title is sensationalist; though I'm guessing the guy is better than D-/D
On January 20 2011 03:58 L_Master wrote: This isn't how you argue, no need to personally attack the guy. This has the potential to be a pretty interesting discussion but people almost invariably respond to flames like this de-railing the whole topic.
Also, what Chef said, I kinda want to see a full game, as well as the entire game vs Oriol.
This is how I argue on an online forum. I don't have to be nice because I have no tolerance for idiocy. If you can't understand basic concepts, you do not have a place to argue more in-depth or advanced ones. Simple as that. If I see something bad or wrong, I'm going to call it out for what it is. If being told the truth is "flaming", then the world doesn't flame enough. Also yeah wtf @ there being no video of the game... also have we figured out who this kid is yet?
If that's your attitude about posting then you should seriously consider deleting your account and moving over to Facepunch. I hear they have fun debates over who is going to make the next LMAO pics thread.
OT: Ok, the AI is probably not able to defeat Jaedong or Flash. That's not such a big deal though, what's amazing is it's ability to adapt. I do understand that it counts the number of unit producing structures etc. and that part is not hard to make. However, making the AI play a few games against mass high templars and becoming better at it each time, that's good.
BTW, throwing around how good of a buddy you are with top programers and flaunting your degree will not make your arguments more solid, you just look like an asshole.
On January 20 2011 03:40 Trowabarton756 wrote: =_= did i just read that right? Dude tried to use mass goliaths to kill off mutas??? i mean wouldn't mm or valks or mm+valks or...irridate...or like a thousand other ideas be far more useful? Iunno I feel as if this "top 16 europe" guy isn't all he's cracked up to be.
Goliath negate Muta micro with their range. Search for a video that showcased the power of AI muta micro.
It could micro 36 mutalisks indidually to kill off unbelievable amount of archon, turning aside at the correct moment. Basicly, AI mutas are invinsible if you cannot outrange them.
Also, Valks die to scourge pretty easily. And the AI can split the mutas anc control them seperately perfectly.
About irradiate, look at above
On January 20 2011 04:01 Diminotoor wrote: Did you even watch the videos? Its clumping and spreading accordingly. Thusly, in a clump irradiate would be effective if only a contribution of 1 to the MM/Valkyries
The irradiated muta would be separated immediately and sent to his doom.
You act as if the initial splash damage and forcing a muta away is a bad thing? Valks die to scourge if you suck at positioning them and we'd have to see its response to that because its response to 3-gate zealot was make tons of zerglins and 4 sunkens except it put a sunken in its main....Don't get me wrong, its impressive, they essentially made a D-/D zerg out of a program. Thats cool, but to boast it beat a "progamer" when the kid was obviously not a progamer is flat out lying.
Well, the whole thing about the valks is arguable, because whether or not the valks can be snipes depends on how the game plays out.
And yes, I think it will die to most ground attacks, only only 3 gate speed zeal, but standart 4 gate 2 archon. In the game where they demonstrated the overlord scounting, I noticed there was no sim city, just 2 sunkens placed at seemingly random locations at the choke
Edit- Theres a rep of him playing against a terran AI, he was protoss. He goes 2-gate goon into 3-gate robo dt for drops, constantly has +1,000 minerals, constantly ques up 3+ units at his gateways(one time there was even 4 dragoons qued up in 1 and the other 2 not making anything. he didn't take his first expand till >10 minute mark. It honestly was like watching a game from 2001. His aka is =Dogo=
On January 20 2011 04:19 Stenstyren wrote:Ok, the AI is probably not able to defeat Jaedong or Flash. That's not such a big deal though, what's amazing is it's ability to adapt. I do understand that it counts the number of unit producing structures etc. and that part is not hard to make. However, making the AI play a few games against mass high templars and becoming better at it each time, that's good.
BTW, throwing around how good of a buddy you are with top programers and flaunting your degree will not make your arguments more solid, you just look like an asshole.
My point is the AI is not able to defeat ANYBODY OF ANY SIGNIFICANT VALUE. I bet any B/B+/A- on Brainclan would present more of a challenge than this guy did. Its an improvement on its ability to adapt but its simply a matter of storing responses. Now how you PROGRAM those responses is the complex part, but nonetheless, my original point still stands since literally nobody has offered any evidence to the contrary. Program enough responses into it, you have basically a human-emulation AI that simply has vastly superior multi-tasking/micro abilities. The thing that differentiates human players from AI is that our ability to "store responses" is infinitely more complex than computer code will allow currently.
Throwing around that I know many progamers, top progamers at that, and that I have studied the field for which this thread is about extensively does in fact show that I am well-aware of what I'm talking about. Sorry you don't have a tough coating to handle how I word things. However bringing your personal deficiencies to the table just makes you look more pathetic instead of the "heroic voice against the over-aggressive oppressor" thing you were trying to go for.
**Edit** Finished watching, the T went for a 2 fact mech build looks like. His ebay was late and his turrets were... well pretty much nonexistent. Any opinions on his play? I wanna see this AI vs Bio just once.
On January 20 2011 04:46 freelander wrote: I'm more and more sure that you won't last long here buddy. You are not even worth arguing, and I can tell you you are not alone with a CS degree here.
Kid, I've been dealing with dangers my entire life the likes of which you can't even fathom. You claim I'm "not even worth arguing" which doesn't make any sense, yet you can't stop responding to me. Neither you, nor anyone else here who lacks a spine or intelligence has a chance of even getting a real rise out of me much less "chasing me away". Get over yourself.
I'm more and more sure that you won't last long here buddy. You are not even worth arguing, and I can tell you you are not alone with a CS degree here.
Would be possible to get any of the matches commentated? Here are some of the Overmind replays from the competition: best_replays.zip
1. Chronos_v_Overmid This was the finals match in the bottom bracket. I was expecting Chronos to be able to stop the muta harass with wraiths, but it didn't plan out quite as expected.
2. Krasi0_v_Overmind This is the first game from the final match. It was a lot closer than it may have looked from the clips I have uploaded so far:
3. Oriol_v_Krasi0 This was one of the preliminary matches between DoGo and the Krasi0 bot, and was much more interesting than the match I tried to commentate on.
There may be several other great replays from the competition, but there's a lot to go through so I hope this helps streamline the process.
On January 20 2011 04:43 Diminotoor wrote: Kid, I've been dealing with dangers my entire life the likes of which you can't even fathom. You claim I'm "not even worth arguing" which doesn't make any sense, yet you can't stop responding to me. Neither you, nor anyone else here who lacks a spine or speck of intelligence has a chance of even getting a real rise out of me much less "chasing me away". Get over yourself.
Seriously? Now you've dealt with dangers we cannot fathom as well? Why is that even remotely connected to what we are discussing here.
I still claim that this is an achievement. It's not easy to design AI's that are half decent and non-abuseable but these guys have done a pretty good job. Coding the AI to react to stuff is not as easy as you make it sound, you do not just plug it into a database somewhere and that's it.
On January 20 2011 04:43 Diminotoor wrote: Kid, I've been dealing with dangers my entire life the likes of which you can't even fathom. You claim I'm "not even worth arguing" which doesn't make any sense, yet you can't stop responding to me. Neither you, nor anyone else here who lacks a spine or speck of intelligence has a chance of even getting a real rise out of me much less "chasing me away". Get over yourself.
Seriously? Now you've dealt with dangers we cannot fathom as well? Why is that even remotely connected to what we are discussing here.
It means you aren't gonna "scare me away". If you can't make that simple connection, you may want to seek help fast because you're going to have a lot of problems going through life.
On January 20 2011 05:02 Stenstyren wrote: I still claim that this is an achievement. It's not easy to design AI's that are half decent and non-abuseable but these guys have done a pretty good job. Coding the AI to react to stuff is not as easy as you make it sound, you do not just plug it into a database somewhere and that's it.
Of course this is an achievement. Literally NOWHERE have I said it wasn't. Also if you had read ANY of what I've posted, I already said long ago that the actual coding of responses is incredibly complex and requires a lot of judging abilities on the AI's part. That being said, the end result is STILL "we programmed in responses to stimuli". Expand the AI's ability to process those possible responses to the stimuli, and you have yourself a better AI. I seriously don't understand how any of this is difficult to understand or see. I'd draw a chart but I'm horribad at MSPaint...
WCG has no record of a Oriol Vinyals competing. The only thing close is Oriol Prats Navarro a UT99 player from Spain. The website doesn't have players from the WCGC (2000) but also does not list Spain as one of the countries that competed that year.
Did he mean that he'd played in a preliminary of the Spain qualifiers for Spain? Because that's something significantly different.
On January 20 2011 04:19 Stenstyren wrote:Ok, the AI is probably not able to defeat Jaedong or Flash. That's not such a big deal though, what's amazing is it's ability to adapt. I do understand that it counts the number of unit producing structures etc. and that part is not hard to make. However, making the AI play a few games against mass high templars and becoming better at it each time, that's good.
BTW, throwing around how good of a buddy you are with top programers and flaunting your degree will not make your arguments more solid, you just look like an asshole.
My point is the AI is not able to defeat ANYBODY OF ANY SIGNIFICANT VALUE.
Yeah no crap. The article says as much. It beat someone using Goliaths. People are getting all keyed up about that, like why did he go mass goliath etc. He probably did so as part of testing the AI. Like the people in the project probably asked him to "go goliath this time, to see how the AI responds". It responded well enough to beat him, in that situation (while losing every other game to him).
Saying that it couldn't beat a progamer, or a ICCUP d+ or whatever is setting up, then demolishing a strawman. The article never said it could. The AI progressed to the point where it could take out other AIs, and beat a pretty good human in a specific situation. But if you don't see how that is interesting I'm not sure what to say. Before an AI can beat a human consistently in any/most situations, it has to be able to do it in one. You start somewhere.
This is interesting because it is the best AI (results show that), and it is getting closer to being good. It's interesting that an AI this good (for an AI) still has massive difficulties in beating a good human, but also interesting that it is getting there. It's not some slam at progamers. It won't be a slam at them if/when the AI can beat them. It is an AI, it is not a level playing field. If/when it can ever beat progamers, it won't make what they do less impressive just like no one thought Kasparov was now terrible and worthless after a computer beat him. You reall ydon't need to white knight progamers so hard.
On January 20 2011 05:11 Beside_kr wrote: WCG has no record of a Oriol Vinyals competing. The only thing close is Oriol Prats Navarro a UT99 player from Spain. The website doesn't have players from the WCGC (2000) but also does not list Spain as one of the countries that competed that year.
Did he mean that he'd played in a preliminary of the Spain qualifiers for Spain? Because that's something significantly different.
=DoGo= participated in the WCG 2001 finals for Spain. Given the amount of time since his peak training, the thread title is definitely an overstatement.
the golaiths were propbably the best choice in this scenario. based on the attack style, it should be able to beat mnm really easily, hence the player chose to golaiths with its range
It's still *incredibly* hard to strategise past build orders as an AI, and micro is easy to program with say, 1 muta vs 1 archon, but when you start to put in different mixes of units and different unit formations which humans will make, it gets exponentially harder. While it might be able to micro mutas perfectly, I'm betting it still can't strategise as well as Stork.
On January 20 2011 04:19 Stenstyren wrote:Ok, the AI is probably not able to defeat Jaedong or Flash. That's not such a big deal though, what's amazing is it's ability to adapt. I do understand that it counts the number of unit producing structures etc. and that part is not hard to make. However, making the AI play a few games against mass high templars and becoming better at it each time, that's good.
BTW, throwing around how good of a buddy you are with top programers and flaunting your degree will not make your arguments more solid, you just look like an asshole.
My point is the AI is not able to defeat ANYBODY OF ANY SIGNIFICANT VALUE.
Yeah no crap. The article says as much. It beat someone using Goliaths. People are getting all keyed up about that, like why did he go mass goliath etc. He probably did so as part of testing the AI. Like the people in the project probably asked him to "go goliath this time, to see how the AI responds". It responded well enough to beat him, in that situation (while losing every other game to him).
Saying that it couldn't beat a progamer, or a ICCUP d+ or whatever is setting up, then demolishing a strawman. The article never said it could. The AI progressed to the point where it could take out other AIs, and beat a pretty good human in a specific situation. But if you don't see how that is interesting I'm not sure what to say. Before an AI can beat a human consistently in any/most situations, it has to be able to do it in one. You start somewhere.
This is interesting because it is the best AI (results show that), and it is getting closer to being good. It's interesting that an AI this good (for an AI) still has massive difficulties in beating a good human, but also interesting that it is getting there. It's not some slam at progamers. It won't be a slam at them if/when the AI can beat them. It is an AI, it is not a level playing field. If/when it can ever beat progamers, it won't make what they do less impressive just like no one thought Kasparov was now terrible and worthless after a computer beat him. You reall ydon't need to white knight progamers so hard.
Of course it is interesting but the article writer was doing a dirty job of giving the impression that the AI can beat everyone who play SC except for a few. Although in fact it won't have a chance against a macro D+ on iCCup (I don't think I can beat a high D either). Making the world think that the average skill level of BW gamer is as low as this AI is unacceptable.
Not all too impressed yet. Did you know that mutas can actually fire backwards if you give the commands at accactly the right frames ? There was briefly a TL BWAPI team I was in and we had a private video of mutas owning infinite scourges just running away and firing backwards while moving. It might still be on youtube but it's most likely still private. All this AI do is mimic smartness. It's still just a simple algorithm changing certain key timings and targeting. I do have to say that it does so very well.
I added the links to some of the videos I made with dropship / tank or goli micro since this is a topic about AI
On January 20 2011 05:30 kamikami wrote: Of course it is interesting but the article writer was doing a dirty job of giving the impression that the AI can beat everyone who play SC except for a few. Although in fact it won't have a chance against a macro D+ on iCCup (I don't think I can beat a high D either). Making the world think that the average skill level of BW gamer is as low as this AI is unacceptable.
Again with the strawman. from the article: "after dozens of test matches, it has finally defeated our human StarCraft expert for the first time ... “Okay,” he says. “We can beat goliaths. What’s next?”"
The article is simply not making these claims you think you are seeing. It is an article about how the AI beat other AIs. And once, *once* it beat a "good human" (good human being someone who the article points out is retired), in one specific situation. That might change someday, but that's the situation now, and the article lays it out very clearly. You are getting worked up about something the article does not say. You will have to provide quotes if you want to change my mind.
More from the article: "By the time the submission deadline came around, the Overmind was good enough that he (Oriol, retired) had to play seriously" - It says he had to play seriously, not that he lost. He had to play seriously to beat it, which he, A retired guy no one has heard of, did (except for one time when they wanted to test it against goliaths)
“Dan believes that in a few years there will be agents that can consistently compete against the highest levels of human competition." - this means there isn't such a thing now. He might be wrong about whether such a thing will ever exist, but it is quite clear that he doesn't think it does now, and quite clear for anyone who actually reads and comprehends the article.
What the AI is, is not as dumb as the built in AI. You can't kite its whole worker force around with a single drone. Its build order doesn't break down if a building is missing or taken out or gas is not available. It is a massive leap over the built in AI, and a good first step towards maybe someday being able to compete with good humans. It is (or could be) the groundwork. It is not there now and no one says it is.
Bottom line: it is an article about "how the Berkeley Overmind won the 2010 StarCraft AI competition"
On January 20 2011 05:36 Marradron wrote: Not all too impressed yet. Did you know that mutas can actually fire backwards if you give the commands at accactly the right frames ?
Perfect micro is far less impressive from an AI standpoint than figuring out macro and when to build things. It is a much easier problem to solve, and there's no reason the Overmind couldn't incorporate it once it has figured out when to build what. What the Overmind is doing is far harder (for a machine) than microing some units around.
Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
ai can abuse bw imbalances perfectly and probably has a maphack *g* so if they work on it more they are able to beat any human being that is restricted to the old bw input system for humans hehe.
Anyway ... deep blue had help by a person while playing so it doesn't really count as mashine beat a grandmaster. But probably computers are now advanced enough to do so.
The bot still has a lot of issues, but it's definitely pushing on research in this domain. If you have an idea of what the bot is going for, then it's usually not to challenging to overrun the bot with a timing attack. I played against the system, but had a pretty idea of what the bot would do beforehand: http://eis-blog.ucsc.edu/2010/11/man-vs-bot/
Unfortunately there is not yet an easy to set up matches for playing against BWAPI. This is something being explored for the next competition.
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
On January 20 2011 05:30 kamikami wrote: Of course it is interesting but the article writer was doing a dirty job of giving the impression that the AI can beat everyone who play SC except for a few. Although in fact it won't have a chance against a macro D+ on iCCup (I don't think I can beat a high D either). Making the world think that the average skill level of BW gamer is as low as this AI is unacceptable.
Again with the strawman. from the article: "after dozens of test matches, it has finally defeated our human StarCraft expert for the first time ... “Okay,” he says. “We can beat goliaths. What’s next?”"
The article is simply not making these claims you think you are seeing. It is an article about how the AI beat other AIs. And once, *once* it beat a "good human" (good human being someone who the article points out is retired), in one specific situation. That might change someday, but that's the situation now, and the article lays it out very clearly. You are getting worked up about something the article does not say. You will have to provide quotes if you want to change my mind.
More from the article: "By the time the submission deadline came around, the Overmind was good enough that he (Oriol, retired) had to play seriously" - It says he had to play seriously, not that he lost. He had to play seriously to beat it, which he, A retired guy no one has heard of, did (except for one time when they wanted to test it against goliaths)
“Dan believes that in a few years there will be agents that can consistently compete against the highest levels of human competition." - this means there isn't such a thing now. He might be wrong about whether such a thing will ever exist, but it is quite clear that he doesn't think it does now, and quite clear for anyone who actually reads and comprehends the article.
What the AI is, is not as dumb as the built in AI. You can't kite its whole worker force around with a single drone. Its build order doesn't break down if a building is missing or taken out or gas is not available. It is a massive leap over the built in AI, and a good first step towards maybe someday being able to compete with good humans. It is (or could be) the groundwork. It is not there now and no one says it is.
Bottom line: it is an article about "how the Berkeley Overmind won the 2010 StarCraft AI competition"
On January 20 2011 05:36 Marradron wrote: Not all too impressed yet. Did you know that mutas can actually fire backwards if you give the commands at accactly the right frames ?
Perfect micro is far less impressive from an AI standpoint than figuring out macro and when to build things. It is a much easier problem to solve, and there's no reason the Overmind couldn't incorporate it once it has figured out when to build what. What the Overmind is doing is far harder (for a machine) than microing some units around.
Thank you for your informed post. Also, I assume the title has been changed from "progamer" or whatever, so I don't think there's much to argue about anymore.
(On a related note, what if the title had said "programmer" instead of "progamer"? I bet a lot of people would be reading it as "progamer" looool)
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
Clearly I can't argue with someone who forms an opinion on an article by reading a single paragraph. And clearly I don't need to.
On January 20 2011 04:43 Diminotoor wrote: Kid, I've been dealing with dangers my entire life the likes of which you can't even fathom. You claim I'm "not even worth arguing" which doesn't make any sense, yet you can't stop responding to me. Neither you, nor anyone else here who lacks a spine or intelligence has a chance of even getting a real rise out of me much less "chasing me away". Get over yourself.
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
No, nobody said the whole idea was invalidated. Try reading more thoroughly. First impressions mean everything. Getting things wrong and giving a false pretense are the worst things you can do to an audience. We're pointing out that the writer did just that.
Of course we can go through the article and quote a bunch of stuff too and claim that its the only way to interpret it, or you can accept the fact that the intros carry an air about them that isn't reflected in the rest of the article. Your choice.
Yes. Right from the very start you've been saying the intro is bad. Not that the article is crap.
Guess you're choosing the "recycle my bad ideas" route. Seriously if you're not comprehensive at English, don't start trying to debate in it.
So you want me to read more thoroughly into your idea that people don't need to read?
Making things up out of the blue and trying to say that I said them doesn't work so well on forums when you can see everything I typed and thusly can clearly see you're flat-out lying.
I don't see what the argument is for. The article is obviously written for an audience that does not know how to play high-level starcraft. I wouldn't be suprised in the least if they could do one going over the intricacies and problems of the AI for higher-level players, but right now it's best to keep in mind who the article is written for.
There is no issue with the arstechnica article. It was well-written, accurate, and enjoyable.
I however have a problem with how the pcgamer article basically repackaged it with a bunch of lines saying READ MEEEE LOLOLOL (like of course, the title saying "progamer") and didn't actually add any new content
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
No, nobody said the whole idea was invalidated. Try reading more thoroughly. First impressions mean everything. Getting things wrong and giving a false pretense are the worst things you can do to an audience. We're pointing out that the writer did just that.
Yes. Right from the very start you've been saying the intro is bad. Not that the article is crap. hahahaha
On January 20 2011 05:36 Marradron wrote: Not all too impressed yet. Did you know that mutas can actually fire backwards if you give the commands at accactly the right frames ? There was briefly a TL BWAPI team I was in and we had a private video of mutas owning infinite scourges just running away and firing backwards while moving. It might still be on youtube but it's most likely still private. All this AI do is mimic smartness. It's still just a simple algorithm changing certain key timings and targeting. I do have to say that it does so very well.
I added the links to some of the videos I made with dropship / tank or goli micro since this is a topic about AI
That muta micro vs the huge pile of templar was at once both terrifying and somewhat expected if the programmers had accomplished their goal of making the AI understand threats properly. Really interesting to read about their development process - and about how the AI embraces the concepts of strong play.
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
No, nobody said the whole idea was invalidated. Try reading more thoroughly. First impressions mean everything. Getting things wrong and giving a false pretense are the worst things you can do to an audience. We're pointing out that the writer did just that.
Of course we can go through the article and quote a bunch of stuff too and claim that its the only way to interpret it, or you can accept the fact that the intros carry an air about them that isn't reflected in the rest of the article. Your choice.
So you want me to read more thoroughly into your idea that people don't need to read?
On January 20 2011 02:23 Torte de Lini wrote: No surprise here. A computer does not suffer from strains of the fingers and perfect microing is no problem for it either.
Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
SCBW is totally different from Chess, cuz it's in real-time and you have to anticipate MUCH MUCH more possible actions of your opponent.
I guess it's possible for a AI to beat a good player in one out of many matches, but for a PC to constantly beat a top-player just by perfect macro/multitasking is IMO not possible in the near future, just because most top-players already have incredibly strong macro, but they can anticipate, act and not only react etc.
My question is: Did the AI also have to scout through the fog of war, or was it able to "see" stuff without scouting?
Honestly, I don't think it will be very long before an AI is able to beat top BW pros. If the research continues at the rate its going it won't be long. In fact, I'd argue that if they can figure out a way for the AI to survive until it has a reasonable amount of Mutas then it will win every time.
Mass mutas with that AI controlling is basically unstoppable. There's literally no way to outplay micro like that.
On January 20 2011 02:23 Torte de Lini wrote: No surprise here. A computer does not suffer from strains of the fingers and perfect microing is no problem for it either.
Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
SCBW is totally different from Chess, cuz it's in real-time and you have to anticipate MUCH MUCH more possible actions of your opponent.
I guess it's possible for a AI to beat a good player in one out of many matches, but for a PC to constantly beat a top-player just by perfect macro/multitasking is IMO not possible in the near future, just because most top-players already have incredibly strong macro, but they can anticipate, act and not only react etc.
My question is: Did the AI also have to scout through the fog of war, or was it able to "see" stuff without scouting?
Read the article. It talks about the development of scouting and how it uses that information.
On January 20 2011 02:23 Torte de Lini wrote: No surprise here. A computer does not suffer from strains of the fingers and perfect microing is no problem for it either.
Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
SCBW is totally different from Chess, cuz it's in real-time and you have to anticipate MUCH MUCH more possible actions of your opponent.
I guess it's possible for a AI to beat a good player in one out of many matches, but for a PC to constantly beat a top-player just by perfect macro/multitasking is IMO not possible in the near future, just because most top-players already have incredibly strong macro, but they can anticipate, act and not only react etc.
My question is: Did the AI also have to scout through the fog of war, or was it able to "see" stuff without scouting?
I believe from their discussion near the bottom of page 3 through page 4 of the article that the AI had to scout through the fog of war, along with their entry into the 4th tournament (the full game tournament) implying that, like a regular game, fog of war would be turned on.
On January 20 2011 02:23 Torte de Lini wrote: No surprise here. A computer does not suffer from strains of the fingers and perfect microing is no problem for it either.
Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
SCBW is totally different from Chess, cuz it's in real-time and you have to anticipate MUCH MUCH more possible actions of your opponent.
I guess it's possible for a AI to beat a good player in one out of many matches, but for a PC to constantly beat a top-player just by perfect macro/multitasking is IMO not possible in the near future, just because most top-players already have incredibly strong macro, but they can anticipate, act and not only react etc.
My question is: Did the AI also have to scout through the fog of war, or was it able to "see" stuff without scouting?
It could scout and could only react to information that it has scouted. There were categories in the competition for AIs that could only do micro and AIs that didn't scout, but all the super-advanced ones could scout.
The computer that beat Kasparov supposedly cheated, which was covered in another documentary (I forgot the title). The human won easily the first day by exploiting ambiguous positions that the AI couldn't navigate, but the second day the AI changed dramatically (apparently even making mistakes) and won. After that IBM immediately disassembled the computer so nobody could confirm that it played the game un-assisted.
Who cares if one article was stupid and had an awful headline/intro? If you're discouraged or sent away by that, you're not really interested in the AI or the work which is vastly more interesting than simply beating a progamer with ai.
The content is incredible and a little scary, not because of current code/ability, but the potential. Like the arstechnica article headlines - Skynet.
Quite a change from the perfectly micro'ed dragoons and mutas, this overmind bot, I'm impressed! They should provide more videos, with explanations of what is going on, what the code is doing and not doing, it's very interesting what's up so far, but it's not enough.
Hopefully these guys get hired by Blizzard or some other rts studio, this is pure gold compared to BW AI.
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
No, nobody said the whole idea was invalidated. Try reading more thoroughly. First impressions mean everything. Getting things wrong and giving a false pretense are the worst things you can do to an audience. We're pointing out that the writer did just that.
Yes. Right from the very start you've been saying the intro is bad. Not that the article is crap. hahahaha
Once again, I would differentiate between the pcgamer article (crap) and the arstechnica (mindblowing, well done)
On January 20 2011 05:11 Beside_kr wrote: WCG has no record of a Oriol Vinyals competing. The only thing close is Oriol Prats Navarro a UT99 player from Spain. The website doesn't have players from the WCGC (2000) but also does not list Spain as one of the countries that competed that year.
Did he mean that he'd played in a preliminary of the Spain qualifiers for Spain? Because that's something significantly different.
=DoGo= participated in the WCG 2001 finals for Spain. Given the amount of time since his peak training, the thread title is definitely an overstatement.
I apologize, WCG and TLPD have DoGo under the name Antonio Crespo Gomez
One problem they will have to look out for is overflowing commands if the game gets to a certain supply amount. As evidenced by map techniques recently, if you flood the game with 10k APM functions of the game stop working correctly or at all. So they will have to streamline it more as well.
Does anyone know if any academic papers in regards to the algorithms used in this have been released yet? I took a scan through the papers released in 2010 by Klein, but couldn't find any. The closest one I could find would be the Hierarchical Bayesian Approach, but that one involved only the PI and only as a 3rd author, and didn't make any mention of possible applications to what was mentioned in the arstechnica article.
and how the developers of the AI could possibly harvest code from some of these hacks to buff up certain aspects of its early game. I'm not quite sure how much this would benefit the AI, since theoretically wouldn't it already have enough APM to do perfect splits / mass unit selects / mass building selects? And also wouldn't this detract from the learning algorithms that the programmers are trying to implement and thus from the advances in AI from a broader perspective?
Also, from reading the article, the team has so far limited itself to a primarily mutalisk based army composition with slight deviations to respond to early aggression (am I reading this right?). As others have mentioned, videos of more games in their entirety would be nice to see how the AI reacts to various things (including cheese early all-ins). I would be interested in reading how the 2nd place AI worked and approached the learning / adaptation problem, since their army composition consisted of more than 1 type of unit. Also it would be fun to see how (and if) the AI could learn to use different builds and unit compositions in response to scouting, since so far it reads like the scouting is mainly used in a defensive manner until a critical mass of mutas is reached. Looking forward to the advances in SC AI in the future!
EDIT: Sorry, stupid question about the academic papers. I forgot about the review period they have to go through with academic journals before they can get published. Hope that finishes soon >_<.
On January 20 2011 05:53 kamikami wrote: Well MamiyaOtaru you can defend the article as you like but I (and the guys arguing with you before) will just hate it because it gives false impression that SC skill level is low and that's it, you cannot change what people feel while reading it.
Why ? Because the title and the first paragraph give the impression that the human opponent is very high level (WCG competitor, 1st in Spain...), it just doesn't state specifically that it was 10 years ago and nowadays he is no one.
So the whole idea is invalidated because someone might not read the whole article?
The entire "argument" in the thread should've gone like this:
"I don't think the thread title is accurate." "yep it isn't." "okay."
And yes they did say that he's no longer competitive at the game.
No, nobody said the whole idea was invalidated. Try reading more thoroughly. First impressions mean everything. Getting things wrong and giving a false pretense are the worst things you can do to an audience. We're pointing out that the writer did just that.
Yes. Right from the very start you've been saying the intro is bad. Not that the article is crap. hahahaha
Once again, I would differentiate between the pcgamer article (crap) and the arstechnica (mindblowing, well done)
I THINK that's what he's trying to say too.
Thank you for this post. That would explain a lot. I am reacting solely to the Ars article, I didn't even notice the underlined "here" in the first line mea culpa. Still , the worst I can say about it is that it has a terrible title, and cherry picks a bit from the Ars article it then links to, where one can get the full story.
The article is definitely flawed. For people reading it who have little or no prior knowledge of starcraft it would most probably lead them to misunderstand the game. The article does mention a level of strategic depth that the AI isn't programmed to address, and it does imply at one point that it cannot currently consistently compete with humans at the highest level of play, however the great significance of that strategic depth as an obstacle to progamer-level competition is never fully aknoweledged, and is badly overshadowed by a bias of pride in the AI's impressive, yet humble accomplishments.
The article does make it clear however just what is involved in those accomplishments, that these methods of AI programming are not what most people would expect. Rather than simply identifying working strategies in every possible situation, essentially telling the computer what to do without any idea of why it's doing it being necessary, the programmers identified many of the underlying parameters behind the types of decisions that need to be made in starcraft, essentially telling the computer the why without the what. This kind of comprehensive understanding of the fundamental elements of gameplay, with practical demonstration of the effectiveness of such understanding, is very interesting from a game theory point of view.
The article's dishonesty or ignorance is pretty reprehensible and definitely deserves pointing out, but it still has merit that makes it very worth reading.
Can someone please watch this video? If you think you've seen it before, then you probably haven't. This is much better and harder to do than what you see in pro games. JD has done this before but nobody bothered to watch the video I posted because they all think they've seen it before.
It would have been much simpler for it to use hydralisk than mutalisk. Most humans can't beat a hydra break, why would an AI be able to? Besides I think the AI would be totally screwed by an AI or player told only to build photon cannons and dragoons...
On January 20 2011 09:36 Chimpalimp wrote: It would have been much simpler for it to use hydralisk than mutalisk. Most humans can't beat a hydra break, why would an AI be able to? Besides I think the AI would be totally screwed by an AI or player told only to build photon cannons and dragoons...
The AI executes the same general gameplan vs all races
At some point (sooner as opposed to later) the AI will be able to consistently beat humans. Then we will have to worry if we are playing a real human in a tourney.
Really fun to watch! Very impressive. Gjob to the organizer Ben Weber and all the contestants/universities that participated in the AI competition. I see that krasi0 took runner up!
Amazing how much progress has took place since kovarex and lowerlogic's humble beginnings.
Look forward to seeing how this continues to evolve!
Until it beats Flash not really impressed. Is the AI restricted to actions only on the current view? If not than it should have no problem out multi-tasking anyone.
On January 20 2011 09:09 zobz wrote: The article is definitely flawed. For people reading it who have little or no prior knowledge of starcraft it would most probably lead them to misunderstand the game. ... The article does make it clear however just what is involved in those accomplishments, that these methods of AI programming are not what most people would expect
It's not an article about Starcraft, it's about an AI. It's written by people who are making an AI, not Starcraft gurus. It is about an AI that beat a bunch of other AIs, and hey, one time it beat some human dude. Beating other AIs is, at this point, a lot easier than beating good human players, which is quite clear from the article. They have to start somewhere. They may be a little optimistic in extrapolating future performance, but calling it dishonest or reprehensible is.. hyperbolic, inaccurate and mean.
Big whoop, the article uses somewhat ambiguous words to capture your attention. Read the newspaper sometimes.
And for those of you who claim you are not impressed with this... REALLY? The idea that a bunch of metal and electricity (ok... nonmetals too) can carry out THIS level of play in our most beloved game is not interesting to you at all? Wow.. you're harder to impress than GSP
The hard part is usually not "How to do task A" but "When to do Task A" and "What to do in a situation".
Take this example: He has 5 goliath and turret guarding some scvs, is it worth killing the goliath, shoot some scv then run? Or should we kill turret first? Or just kill scv and run? These are the questions the bot needs to address, and the article covers it a little.
So again, coming up with adaptive strategy is much harder than coming up with some micro-trick that anyone could implement. I get the feeling we're focusing too much on "Wow yeah def computer OP it has 100000 clicks!!" But think about it, it is still difficult to decide "Oh yeah so u have 10000 clicks, how are u going to use them?"
On January 20 2011 02:23 Torte de Lini wrote: Wasn't there a documentary about this supercomputer and chess that played brilliantly against grandmasters?
'Game Over: Kasparov and the Machine' was one, though probably not the only one:
Can someone please watch this video? If you think you've seen it before, then you probably haven't. This is much better and harder to do than what you see in pro games. JD has done this before but nobody bothered to watch the video I posted because they all think they've seen it before.
*sees jaw fall off*
*blink blink*
WAT??? 30k apm??? seriously, what are people making these days... i feel like this will be the end of starcraft if we have that sort of computer out there...
On January 20 2011 06:11 djsherman wrote: The bot still has a lot of issues, but it's definitely pushing on research in this domain. If you have an idea of what the bot is going for, then it's usually not to challenging to overrun the bot with a timing attack. I played against the system, but had a pretty idea of what the bot would do beforehand: http://eis-blog.ucsc.edu/2010/11/man-vs-bot/
Unfortunately there is not yet an easy to set up matches for playing against BWAPI. This is something being explored for the next competition.
these matches looked like fun, but the bots still have severe weaknesses a human can recognize after a few games
Man i can't see enough of this. Its really fun to watch. I think watching two computers duke it out is better then watching it play a human.
That muta micro. I hope these ai's get even better. Image having a practice partner that could play every race, and got better with time, and you could play him 24/7.
Can someone please watch this video? If you think you've seen it before, then you probably haven't. This is much better and harder to do than what you see in pro games. JD has done this before but nobody bothered to watch the video I posted because they all think they've seen it before.
This is badass. When did JD do it? I want to see a cross section of an observer computer, a FP VOD, and a video of his hands during the incident, cause damn.
damn thanks for posting that chess trailer! im going to get back into chess. scary how the comps > grandmasters. will NEVER happen in starcraft though.
I just want to briefly mention a couple of things. The title is accurate, but for those seeking further clarification:
-I competed in the WCG 2001, and stopped playing a long time ago. Obviously, in terms of skill, I'm fairly average right now. However, my knowledge of the game is still rock solid, and given that I'm also a PhD student in AI, the two factors helped the Overmind tremendously.
-The AI system is far from beating anyone decent. I lost a single game where we were trying several conditions, how the mutalisks were acting, etc. In no way I was playing to win, but to "teach" the Overmind, so to speak. So the title may be a bit misleading in this regard, but it was meant to attract attention which it certainly did.
For those wondering what I'm doing now, I'm playing WoW (yes, I know!) on my (scarce) free time, and doing decently so (which helped me not loosing too much APM's or gaming "skills").
Oriol
P.S. I'm not the author of neither the article nor this post
On January 20 2011 09:39 Sanguinarius wrote: At some point (sooner as opposed to later) the AI will be able to consistently beat humans. Then we will have to worry if we are playing a real human in a tourney.
uhhh I think 40k apm gives away the fact it isn't human.
I just want to briefly mention a couple of things. The title is accurate, but for those seeking further clarification:
-I competed in the WCG 2001, and stopped playing a long time ago. Obviously, in terms of skill, I'm fairly average right now. However, my knowledge of the game is still rock solid, and given that I'm also a PhD student in AI, the two factors helped the Overmind tremendously.
-The AI system is far from beating anyone decent. I lost a single game where we were trying several conditions, how the mutalisks were acting, etc. In no way I was playing to win, but to "teach" the Overmind, so to speak. So the title may be a bit misleading in this regard, but it was meant to attract attention which it certainly did.
For those wondering what I'm doing now, I'm playing WoW (yes, I know!) on my (scarce) free time, and doing decently so (which helped me not loosing too much APM's or gaming "skills").
Oriol
P.S. I'm not the author of neither the article nor this post
Cool that you posted to give us some clarity It would be nice if you did tell the lead director that it kinda misrepresents what BW is like, but obviously we still think the accomplishment is a terrific feat! Also, if you're in Berkeley I would love to meet you ^^ We have Starcraft LANs here a lot!
On January 20 2011 05:36 Marradron wrote: Not all too impressed yet. Did you know that mutas can actually fire backwards if you give the commands at accactly the right frames ? There was briefly a TL BWAPI team I was in and we had a private video of mutas owning infinite scourges just running away and firing backwards while moving. It might still be on youtube but it's most likely still private. All this AI do is mimic smartness. It's still just a simple algorithm changing certain key timings and targeting. I do have to say that it does so very well.
I added the links to some of the videos I made with dropship / tank or goli micro since this is a topic about AI
Those were the exact videos I was talking about. They're made like one year and a bit ago. Guesse the owner changed them from private to not private a while ago.
On January 20 2011 15:31 Dox wrote: Lots of disappointing posts in this thread. Try to focus on the accomplishments of the AI & it's programmers, not the player who was beaten. Really...
Ironically, the player it has beaten is displayed as the AI's biggest accomplishment in that article. The level of the player surely is important. If a computer beats me in chess, nobody cares. If I lose to the BW AI, nobody cares.
I think it is inevitable that AIs will be able to beat a majority of players in the future, though not necessarily progamers. However, I do expect to see a future AI be able to competently compete against most progamers at some point, especially considering the sheer amount of perfect micro and multitasking that an AI is capable of doing.
Once someone starts tacking on brilliant game sense on top of beastly muta micro, AIs will become REALLY scary. The ability to perform perfect storm dodges and harass is not to be underestimated, and it's really exciting to see what kind of improvements can be made in the future. Heck, most Koreans already play like computers .
What I find really great about this, is that in the future, they may be able to combine all of the bots to form a something incredible. This competition also shows us that humans aren't able to play to the game's full potential, and that there is always something to work on (for example, Flash's dominance just signals that there are certain aspects of the game he does better that other players need to work on).
Can someone please watch this video? If you think you've seen it before, then you probably haven't. This is much better and harder to do than what you see in pro games. JD has done this before but nobody bothered to watch the video I posted because they all think they've seen it before.
This is badass. When did JD do it? I want to see a cross section of an observer computer, a FP VOD, and a video of his hands during the incident, cause damn.
On January 07 2011 10:47 Wonders wrote: The only hard thing I've seen is a mutalisk actually shooting backwards like a vulture does, with almost no discernible turning. This can't be done with hold position or patrol. One has to move command back, attack command on the thing you want to hit (NOT just attack move near it), then move away, all in a fraction of a second. This is needed to kill scourge with 1 or 2 mutas, since the patrol trick requires at least 3 mutas. Here's what it looks like, starting at 8:00:
He's actually clicking move towards the scourge, then on the scourge, then away from it.
On January 20 2011 15:31 Dox wrote: Lots of disappointing posts in this thread. Try to focus on the accomplishments of the AI & it's programmers, not the player who was beaten. Really...
Ironically, the player it has beaten is displayed as the AI's biggest accomplishment in that article. The level of the player surely is important. If a computer beats me in chess, nobody cares. If I lose to the BW AI, nobody cares.
The fact the AI was able to beat a competent SC player - even in a single game with the human player trying a non-optimal strategy - is already quite an accomplishment.
On January 20 2011 19:55 nalgene wrote: The facebook snake video is the best 1450 points...
I don't get the AI on the Facebook snake.
Why doesn't it always leave a space open at the edges so it can escape when it wants? It should be relatively easy to fill the screen completely if you stick to a given trajectory. It takes a bit longer, but still, from there you can vary from that scenario to make it faster every time.
Completely disregarding all the credibility talk here. I find it very interesting to follow the development of more and more sophisticated AI. Speaking in a game sense, it'll make it infinitely more interesting to play games, and we might actually get to have proper challenges in games that aren't made up around the fact that the opposition gets 100% + to its stats, and instead it'll be a headlong battle of everything from micro to strategic gamesense.
well, AI will beat a human EASILY if done right, it will NEVER forget drones, have a PERFECT split, see even the tinniest detail, like a ling in the corner of your view for a qaurter of a second, and have perfect macro while having perfect micro in battle at the same time.
I saw a video of i dont know what vs i dont know what, and it micro'd it perfectly and took no dmg cos the units out ranged them, was air i think so it stacked. something like 9000 apm.
also if pro grammed right, the AI wont be able to think, but it will know what to do in enough situations, that it ownt even need to think.
This reminds me of a supercomputer IBM is working on at the moment called "Watson".
It's an AI designed to process millions of data within seconds, and aim to be able to compete with world class Jeopardy! players. It is simply incredible! 3000-core 2.7Ghz super computer. It's so fast, if our average high-end computer takes 2hours to process A, it would take Watson to process the same A in 2 seconds
On January 22 2011 06:37 threepool wrote: Hey all, I just did a series of casts on the bot's gameplay, if you're interested. (might stick that in its own thread in a minute)
Nice to get to see the actual gameplay; thanks for casting those.
I've played against Overmind and krasi0 with all races, honestly they are very bad.
In TvZ, Overmind does not seem to transition to lurkers or even muta/ling, and if you know heavy muta harass is on the way, it is very easy to beat with spamming turrets and m&m. In PvZ you just need to double stargate and cannon up. In ZvZ you can beat it with straight up mutalisks, can't remember any specifics.
krasi0 tries to macro up a mech force from 2 bases then exp like an idiot but does not seem to try any timing attack or other proper push. It does not defend its expansion attempts either so it is very easy to just snipe its SCVs to delay its already late expansions. Once in a while he attempts to do drops, but they are always way too late and are very bad drops. In TvT it is very suspectible to siege contains (to be fair it won a game doing the same goddamn tactic, but my terran sucks) and does not seem to transition to dropship play at all. In ZvT it doesn't seem to care if I have a spire up and 9 mutas morphing, he only starts reacting once he sees a mutalisk, or a mutalisk is already attacking him. Expansion sniping, macroing and straight up muta/hydra kills it easily. PvT is its best matchup, but it is easy to outmacro him in this one as well. Straight up zeal/goon/arb kills it.
In my opinion they are dumb as fuck and does not seem to adapt or use any meaningful strategy at all.
On January 21 2011 23:16 therockmanxx wrote: Isn´t a program that can beat Humans in starcraft was already built and installed in Flash Hard Drive xD
Haha, I was just about to say that. There is already an advanced Starcraft AI. His name is Flash.
Just watched one of the Berkeley Overmind games. IMO, the AI has quite a bit of room to improve. For some reason, he built a single sunken in his main, and it did nothing. The Terran AI suicided 3 marines. They also don't have pretty bases (i.e. they place their supply depots and stuff like noobs). Z went 9 pool against the T, and instead of pulling 3 scvs to his ramp with rines behind, Terran let the lings into his base, and then ended up pulling like 8 scvs. Z placed three sunkens in his natural that didn't cover his extractor, so T could have stayed out of the range of sunkens and picked off a few drones, lings, and the gas
Overall, I think the Overmind's mechanics are great. Uses larvae as soon as they appear, and the muta micro is pretty impressive (irradiate becomes noticeably less effective). The decision making is lacking, though. If I wall-in ling-tight at my natural, go 14 cc, and transition into 6 fact pure goliath, I'd be interested in what it would do, especially if I defend for a bit, and then move out. I doubt it could consistently make the proper decision between counter-attacking and pulling back the mutas to defend.
On January 27 2011 12:47 Azzur wrote: There was a time when people thought that computers could never beat humans in chess...
Chess is distinctly different. Many chess computers can get away with tactics and calculation alone. They often do not play positionally. If you play Starcraft with perfect micro and macro, but don't think strategically, you will run into problems. The improvements to these AIs seem pretty simple for a human. Any C- player knows how to place buildings better than those AIs, and how to scout and respond properly to certain strategies. It's noticeably more difficult for computers, though.
And even if you have a computer with great macro, micro, and situational awareness, what happens when it 12 hatches against an unscouted BBS? This type of situation never comes up in chess, so the computer just has to do what it's good at, and you get AIs that have virtual FIDE ratings over 3000.