|
Maybe someone could add that to the TL calendar?
|
On January 24 2019 02:36 Grumbels wrote:Show nested quote +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.
link to paper?
|
On January 24 2019 18:24 Hider wrote:Show nested quote +On January 24 2019 02:36 Grumbels wrote: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. link to paper?
https://deepmind.com/documents/110/sc2le.pdf
On January 24 2019 18:08 -Archangel- wrote:Show nested quote +On January 24 2019 02:34 Grumbels wrote: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? 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: Show nested quote +Tetsuo
Wow nice to finally see some results lets see if any A.I can beat Serral ^^
Show nested quote +Prrredictable
Negative. The Koreans deployed eight humanoid AIs to Blizzcon last year who were all met with a Serral victory. ggwp
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...
|
Anyone excited that with good AI, blizzard can finally test their balance changes properly before pushing them out?
|
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 !
|
On January 24 2019 21:56 LoneYoShi wrote:Show nested quote +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 !
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.
|
The name of the AI is AlphaStar, so either AS or A*. By the way, A* is a well-known AI algorithm
|
Seems like AlphaStar will face TLO?
No way in hell it can be on that level already though right?
|
So many people here who expect absurd things without (seemingly) having the slightest idea of the complexity of the task DeepMind is undertaking. Saying stuff like they "won't be impressed" if the AI can't perform to level x or y. Instead of being super excited about the work that is being done which is automatically PR for the game. It was news in state media in Germany when AlphaGo beat go masters. They had a road to get there. This is the road. You are part of it and are experiencing it. Dota 2 has seen strong evolution of AIs with OpenAI playing at TI last year. Why not just be excited for our game to be part of this scientific field?
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.
|
On January 25 2019 00:36 Grumbels wrote:Show nested quote +On January 24 2019 21:56 LoneYoShi wrote: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 ! 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.
|
On January 25 2019 02:47 MaryJoana wrote:Show nested quote +On January 25 2019 00:36 Grumbels wrote:On January 24 2019 21:56 LoneYoShi wrote: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 ! 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.
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.
|
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...
|
On January 25 2019 02:52 Grumbels wrote:Show nested quote +On January 25 2019 02:47 MaryJoana wrote:On January 25 2019 00:36 Grumbels wrote:On January 24 2019 21:56 LoneYoShi wrote: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 ! 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. 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.
|
On January 25 2019 02:54 ASoo wrote:Show nested quote +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...
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.
|
On January 24 2019 18:35 Grumbels wrote:Show nested quote +On January 24 2019 18:24 Hider wrote:On January 24 2019 02:36 Grumbels wrote: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. link to paper? https://deepmind.com/documents/110/sc2le.pdfShow nested quote +On January 24 2019 18:08 -Archangel- wrote:On January 24 2019 02:34 Grumbels wrote: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? 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 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?
|
On January 25 2019 02:55 MaryJoana wrote:Show nested quote +On January 25 2019 02:52 Grumbels wrote:On January 25 2019 02:47 MaryJoana wrote:On January 25 2019 00:36 Grumbels wrote:On January 24 2019 21:56 LoneYoShi wrote: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 ! 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. 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.
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.
|
On January 25 2019 03:00 GreasedUpDeafGuy wrote:Show nested quote +On January 24 2019 18:35 Grumbels wrote:On January 24 2019 18:24 Hider wrote:On January 24 2019 02:36 Grumbels wrote: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. link to paper? https://deepmind.com/documents/110/sc2le.pdfOn January 24 2019 18:08 -Archangel- wrote:On January 24 2019 02:34 Grumbels wrote: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? 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 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?
pls go live in twitch chat
|
PvP because AI has to start easy :D
|
|
|
TLO's aligulac elo is 2008, for comparison Serral and Maru are over 3000.
Also according to Aligulac, if Serral plays TLO in Bo7, Serral will win 99.41% of the time.
|
|
|
|
|
|
EPT
