ListlessSalmon wrote:Given the previous discussion in here its perhaps worth noting there's some kind of Deepmind sc2 livestream tomorrow hosted by Artosis and Rotterdam: https://www.teamliquid.net/forum/starcr ... ion-jan-24
No idea precisely what it will be, I would have assumed not something substantial (like a full 1v1 program capable of beating reasonable players) but who knows.
I just realized I really dont want them to have a bot that solved sc2. It takes away some of the magic.
Same. I really want to see what strats a bot that solves sc2 could came up with, but at the same time would be pretty boring to reach the next wcs and have the "one true build to use and how to adapt" for every mu, and the game becoming just a mechanical execution contest. Not sure how it will play out in the long run.
Fwiw just based on all the information they have given previously about their progress I really doubt they are close to having something that can beat a top human, at blizzcon last year they were talking about progress on sc2 minigames they made to train it and it all seemed really really far off from a competent full game program. Given that they have Artosis I doubt that it will be nothing though. Probably they have something that can play a recognisable game of sc2 and it will be some game/match against another bot or perhaps like a master league (or lower) human player. I could obviously be wrong though.
Regarding lemmings concern about the possibility of players just learning the "one true build" from an sc2 AI I think chess is probably instructive here. Humans have lost to the best machines for 2 decades at that game (and human vs AI has been utterly hopeless for at least a decade) and yet top GMs still play a variety of openings (and styles).
Are we assuming that an aoe3 AI would have unlimited APM? Or like, exceedingly high APM, much higher than any player could physically accomplish? Because if so, I don't think we need to speculate much on the AI coming up with novel strategies that players haven't thought of. It would probably play in a very straightforward way, and just absolutely crush any player because it would macro and micro perfectly and calculate every single fight perfectly. It would take lots of fights that normal players would avoid because it would calculate that it could win or trade effectively with zero overkill, perfect kiting, and perfect counters. All while at the same time perfectly responding to raids, perfectly macroing vills, and running 10 different hussars independently around your base raiding at every angle. With this level of APM build orders and long-term strategies would become much less relevant. All the AI would have to be programmed to do is trade vill seconds slightly more efficiently than you and it will snowball into a win.
Squamiger wrote:Are we assuming that an aoe3 AI would have unlimited APM? Or like, exceedingly high APM, much higher than any player could physically accomplish? Because if so, I don't think we need to speculate much on the AI coming up with novel strategies that players haven't thought of. It would probably play in a very straightforward way, and just absolutely crush any player because it would macro and micro perfectly and calculate every single fight perfectly. It would take lots of fights that normal players would avoid because it would calculate that it could win or trade effectively with zero overkill, perfect kiting, and perfect counters. All while at the same time perfectly responding to raids, perfectly macroing vills, and running 10 different hussars independently around your base raiding at every angle. With this level of APM build orders and long-term strategies would become much less relevant. All the AI would have to be programmed to do is trade vill seconds slightly more efficiently than you and it will snowball into a win.
But that's where computers fail is they can't strategize on the fly against unexpected opponents. You can only plan for what you can see and unlike chess nearly the entire map is unknown for a large portion of the game. Yeah 1 input per frame would allow 10 huss raids at every angle and 0 overkill skirmishes but the ai can't ever know for sure what units they will come up against unless it is constantly scouting (i would actually expect perfect ai to use stealth units constantly). Its an interesting proposal to think of the possibilities but i still think the ai would get stuck on sieging walls and stuff (technically sieging wall segments is always villie second efficient) unable to grasp big picture strategy unless it was programed with human builds
Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
LoOk_tOm wrote:I have something in particular against Kaisar (GERMANY NOOB mercenary LAMME FOREVER) And the other people (noobs) like suck kaiser ... just this ..
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
There is a card called advanced balloon. Iamturk can teach you
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
There is a card called advanced balloon. Iamturk can teach you
He also has the best mechanics : 95w+150w treasures as India
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
All this "it is impossible to" is just because of human limitations. In the end, even vs another AI, a strong machine would end up playing in a way that wins most of time. If anythong the no draw possibility means some ways can be too risky to follow.
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
There is a card called advanced balloon. Iamturk can teach you
Damn funnily enough he didn't send it in our french mirrors, dunno what happened
Garja wrote:
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
All this "it is impossible to" is just because of human limitations. In the end, even vs another AI, a strong machine would end up playing in a way that wins most of time. If anythong the no draw possibility means some ways can be too risky to follow.
Well yeah like I said you can go for the safest strat, but it won't be perfect since you don't have perfect intel. While mechanics can be perfect.
LoOk_tOm wrote:I have something in particular against Kaisar (GERMANY NOOB mercenary LAMME FOREVER) And the other people (noobs) like suck kaiser ... just this ..
Kaiserklein wrote:Well yeah comps would be mechanically perfect, but strategically there is no such thing as perfection in aoe3. It's often rock-paper-scissor, at best you can find the safest strat, but not a perfect one. Which is why perfect mechanics (perfect positioning, perfect macro, zero overkill, and so on) would be much more relevant than a perfectly clever comp, since it would still have to guess stuff. And there's no way to scout perfectly when your opponent has units and will kill your scout.
There is a card called advanced balloon. Iamturk can teach you
He also has the best mechanics : 95w+150w treasures as India
Well honestly the first serie vs tlo isn't impressive. It's true that the AI doesn't play like we would expect. Super cheesy as you predicted but it doesn't play perfectly at all. Like it would have lost to a top korean protoss.
Strategically this AI is just bad, I'd even claim that I'm better strategically at sc2 than this AI, but AI macro and AI micro helps a lot.
The same AI in aoe3 would get bashed by any pr35+.
Garja wrote:Now it's gonna play a series with a protoss player, after it got trained for another week.
Yea true, the version which played against tlo only practiced for 10 days (which is 200 years in real time), 17 days can definitely make a difference. Still, tlo is 6k mmr, mana 6k3, which is significantly better but far from Serral's 7k3
Honestly I'd only be impressed if it can beat neeb or someone of his level. Still it can only play on one map, one patch and in one MU, so it would take the AI years to become decent at aoe 3.
Lol, honestly the AI just won out of micro, its units get like a 150% value, it's irrelevant, any human would lose doing these strats. Also better macro, it's strategically irrelevant.
But micro is connected with strategic decisions. Of course it's inhuman to have that stalker micro but what if it is was AI vs AI. I think that if the other AI where to pick Mana's strategy he would lose anyway to stalker flanks. So in the end it shows what's best strategically if you can use a certain unit to its max potential.
[Armag] diarouga wrote:This is just lame. The AI just makes full stalkers and wins because perfect macro in sc2 means that you don't lose a single unit.
Yes. This just seemed like the shooter variant of 'just click heads' and win. Strategically it seemed okay, I think it basically understands the game at a ~masters level. That's still impressive, because it basically discovered all of this on its own, but it doesn't seem like this AI managed to break strategic barriers here. They need to tone down the apm, maybe even half it, and try to restrict its awareness/multitasking.
Garja wrote:But micro is connected with strategic decisions. Of course it's inhuman to have that stalker micro but what if it is was AI vs AI. I think that if the other AI where to pick Mana's strategy he would lose anyway to stalker flanks. So in the end it shows what's best strategically if you can use a certain unit to its max potential.
I highly doubt it. Even with that perfect micro, the AI didn't win easily, and mana's micro wasn't perfect at all (although he played great). Anyway, the AI is just going to crush everybody now but I expected the AI to play with a human micro/macro and win on decision making.
[Armag] diarouga wrote:This is just lame. The AI just makes full stalkers and wins because perfect macro in sc2 means that you don't lose a single unit.
Yes. This just seemed like the shooter variant of 'just click heads' and win. Strategically it seemed okay, I think it basically understands the game at a ~masters level. That's still impressive, because it basically discovered all of this on its own, but it doesn't seem like this AI managed to break strategic barriers here. They need to tone down the apm, maybe even half it, and try to restrict its awareness/multitasking.
Yea definitely, the AI has 270 APM, but it's actually 270 EPM, which is more than anyone but 2-3 players in the world, and with no missclic, it can see the whole map, and perfect APM spenging. It should be limited to like 150-200 EPM lol.
It can rape any progamer now, but it's just not fair.
[Armag] diarouga wrote:This is just lame. The AI just makes full stalkers and wins because perfect macro in sc2 means that you don't lose a single unit.
Yes. This just seemed like the shooter variant of 'just click heads' and win. Strategically it seemed okay, I think it basically understands the game at a ~masters level. That's still impressive, because it basically discovered all of this on its own, but it doesn't seem like this AI managed to break strategic barriers here. They need to tone down the apm, maybe even half it, and try to restrict its awareness/multitasking.
Yea definitely, the AI has 270 APM, but it's actually 270 EPM, which is more than anyone but 2-3 players in the world, and with no missclic, it can see the whole map, and perfect APM spenging. It should be limited to like 150-200 EPM lol.
MAybe even 100 epm or less. If literally every click is accurate, that's unlike any human. Humans will often click a few times for actions and spend a ton of apm actually navigating the map. This AI is basically godlike mechanically with it's top down view and something thats probably equivalent to say a 700 apm human player.
