_H2O wrote:That AI is very impressive. If it can get to this level you can imagine it can go further no problem.
Some things are odd but still work. For example it seems to value presaturating before expanding. But to always be well marcoed and to make the combat decisions it is making is very impressive. Obviously it’s through testing 200 years of possible ways to interpret these scenarios. But things like the ramp block are so fascinating because there’s so many ways to build a base and it’s not entirely obvious to a brute force situation why you would ever reach outside of building in base.
It also showcases how you have more options if you have better mechanics. It can win situations a worse mechanical player could not. Mechanics and strategy in humans are linked and in the game they are also linked. It’s not possible to separate them entirely because the level of depth to your game expands as your mechanics open up new avenues.
Didnt the bot mana played train for even more than 200 ingame years? I wonder if its capped in its potential. I mean, 200 years of playing and the bots havent figured out that you can distract someone by dropping immortals with a warp prism. Talk about a slow learner haha.
I badly want to see them let this bot grind much more and then let it play with half the apm.
It trained vs other bots. Bots don't get distracted, so probably the decision of not dropping was because it is not beneficial for the offender. In that situation, at least. I think drops and anything that testes the opponent defenses is the way to go if you have unlimited APM. Then again the AI might have just come to the conclusion that vs perfect defense it's better to save the units for the main army.
_H2O wrote:That AI is very impressive. If it can get to this level you can imagine it can go further no problem.
Some things are odd but still work. For example it seems to value presaturating before expanding. But to always be well marcoed and to make the combat decisions it is making is very impressive. Obviously it’s through testing 200 years of possible ways to interpret these scenarios. But things like the ramp block are so fascinating because there’s so many ways to build a base and it’s not entirely obvious to a brute force situation why you would ever reach outside of building in base.
It also showcases how you have more options if you have better mechanics. It can win situations a worse mechanical player could not. Mechanics and strategy in humans are linked and in the game they are also linked. It’s not possible to separate them entirely because the level of depth to your game expands as your mechanics open up new avenues.
What's impressive? We all know that a bot with 1500 APM during fights can win vs pro gamer, that's not a secret lol. Strategically however, this AI was just bad, spamming one unit isn't what I'd call good.
You can indeed say that being able to take good fights is a form of strategy but to me, the AI wasn't even good at that. Like it almost throwed a game against TLO because it ignored the ramp and got trapped by force fiels, and it didn't know how to abuse sentries and warp prism+immortals.
_H2O wrote:That AI is very impressive. If it can get to this level you can imagine it can go further no problem.
Some things are odd but still work. For example it seems to value presaturating before expanding. But to always be well marcoed and to make the combat decisions it is making is very impressive. Obviously it’s through testing 200 years of possible ways to interpret these scenarios. But things like the ramp block are so fascinating because there’s so many ways to build a base and it’s not entirely obvious to a brute force situation why you would ever reach outside of building in base.
It also showcases how you have more options if you have better mechanics. It can win situations a worse mechanical player could not. Mechanics and strategy in humans are linked and in the game they are also linked. It’s not possible to separate them entirely because the level of depth to your game expands as your mechanics open up new avenues.
Didnt the bot mana played train for even more than 200 ingame years? I wonder if its capped in its potential. I mean, 200 years of playing and the bots havent figured out that you can distract someone by dropping immortals with a warp prism. Talk about a slow learner haha.
I badly want to see them let this bot grind much more and then let it play with half the apm.
It trained vs other bots. Bots don't get distracted, so probably the decision of not dropping was because it is not beneficial for the offender. In that situation, at least. I think drops and anything that testes the opponent defenses is the way to go if you have unlimited APM. Then again the AI might have just come to the conclusion that vs perfect defense it's better to save the units for the main army.
Yea, the AI only played vs itself, ie vs another AI with crazy APM and zoom hack, so multitasking is pointless.
https://deepmind.com/blog/alphastar-mas ... rcraft-ii/ An IA masters starcraft, so I guess it would have some similarity in aoe3. I don't know starcraft, but there seems to be some very interesting strategies done here by the IA.
bwinner1 wrote:https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/ An IA masters starcraft, so I guess it would have some similarity in aoe3. I don't know starcraft, but there seems to be some very interesting strategies done here by the IA.
We've been discussing that for 2 pages now lol. The strategies were bad, really bad, and the AI only won because of insane micro.
I don't know much of starcraft strategies, but how can you say strats were bad? What if an AOE3 AI comes and all it does is just sepoy rush wiht perfect micro? Is that a bad strategy if it works even vs its counter unit?
well what it did vs the drop raids wasn't great tactics and strategy. couldve kept some stalkers back and pushed the enemy base, or couldve made a phoenix and stop the raids. instead it kinda did nothing and went back and forth
the pro opponent said he even disccovered some new strategies in this games, so I don't think it was bad strategies overall even though there have been some bad decisions for sure.
Garja wrote:I don't know much of starcraft strategies, but how can you say strats were bad? What if an AOE3 AI comes and all it does is just sepoy rush wiht perfect micro? Is that a bad strategy if it works even vs its counter unit?
Well strategically it was just worse, it would get big advantages out of inhuman mechanics and then lose a bunch of that with poor decisions (most notably ramp engagements but also the warp prism defense). It isn't especially interesting that blink stalkers become extremely good given inhumanly good control, everyone has already known this. It seemed rather worse strategically compared to its human opponents. That being said, just from an AI progress point of view, that it was able to play a vaguely competent recognisable game of starcraft by learning from replays and vs itself (as opposed to a human defined evaluation function) is impressive to me.
For an interesting competition I kinda wish they would really really nerf its mechanical capacity, give it pretty low hardcapped APM and some kind of ways to limit its multitasking (amongst other things) in somewhat arbitrary ways so that it can reasonably be described as weaker mechanically than a human it is playing against. That way it would actually be a battle of strategic "thinking".
Garja wrote:I don't know much of starcraft strategies, but how can you say strats were bad? What if an AOE3 AI comes and all it does is just sepoy rush wiht perfect micro? Is that a bad strategy if it works even vs its counter unit?
It didn't abuse all the sc2 mechanics. Like warp prim + immortal or archonte is fucking great with good micro, and that's actually how mana managed to win the live game. Likewise, the computer didn't make a single sentry, and you can also abuse force fields with good micro. Also it got trapped and wasted a lot of units to sentries because it just didn't know the unit lol.
bwinner1 wrote:the pro opponent said he even disccovered some new strategies in this games, so I don't think it was bad strategies overall even though there have been some bad decisions for sure.
They said that in a way to say that they were very surprised by the way the AI played, it doesn't mean that it was good. Like not walling, ignoring the ramp, and not making warp prism+immortal+archonte was just dumb. The AI won because of better macro/micro but even with perfect macro/micro there are much better ways to play.
Garja wrote:I don't know much of starcraft strategies, but how can you say strats were bad? What if an AOE3 AI comes and all it does is just sepoy rush wiht perfect micro? Is that a bad strategy if it works even vs its counter unit?
Well strategically it was just worse, it would get big advantages out of inhuman mechanics and then lose a bunch of that with poor decisions (most notably ramp engagements but also the warp prism defense). It isn't especially interesting that blink stalkers become extremely good given inhumanly good control, everyone has already known this. It seemed rather worse strategically compared to its human opponents. That being said, just from an AI progress point of view, that it was able to play a vaguely competent recognisable game of starcraft by learning from replays and vs itself (as opposed to a human defined evaluation function) is impressive to me.
For an interesting competition I kinda wish they would really really nerf its mechanical capacity, give it pretty low hardcapped APM and some kind of ways to limit its multitasking (amongst other things) in somewhat arbitrary ways so that it can reasonably be described as weaker mechanically than a human it is playing against. That way it would actually be a battle of strategic "thinking".
Yes. As I said, it had 270 APM but realistically it's 270 EPM, which probably becomes 350 EPM if you consider the zoom hack, that's too fast for a human. And it was just "average" APM, that is to say 50 APM in the first minute and 1000 APM (so 1500 EPM because of zoom hack) during fights, that's unrealistic. First they need to remove the zoom hack (they did so in the last game and the AI lost), to lower the average APM and to reduce the max APM to something like 400.
Challenger_Marco wrote:To know the ans bring ppl from NASA ,make them play vs h2o or high level player ,obviously high pr players r gonna win(my option).
I think the discussion is about whether being smart of having good mechanics is more important in reaching a high PR. Of course somebody who is better at the game is going to win against somebody who has never played it.
Challenger_Marco wrote:To know the ans bring ppl from NASA ,make them play vs h2o or high level player ,obviously high pr players r gonna win(my option).
I think the discussion is about whether being smart of having good mechanics is more important in reaching a high PR. Of course somebody who is better at the game is going to win against somebody who has never played it.
Conclusion: Mechanics>IQ but if player has both high IQ n good mechanics he is obviously better coz he will know to solve any type of situation easily.
Not sure how higher intelligence would be the deciding factor in a game such as aoe3. Kasparov had his IQ tested to be 123 and 135, which is great but still a lot lower than many other chess players he defeated. And chess games are a lot more strategy and tactic oriented than AoE3, so the arguement that IQ somehow correlates with game performance doesn't make much sense to me.
Mechanics on the other hand, are very important in rts games. The way you control your units and resources will decide the outcome of the game. You can have a great advantage through your strategy and tactics, but unless your mechanics are sufficient to make proper use of your units, you will never win. That is why a PR30 player will almost always lose against a PR40 player doing a bad strategy, the PR40 player will almost always take effective trades (better than he should have, or could have against a stronger player) and outperform the PR30 player mechanically.