AI Strategy

[nemukud]

I assume it's obvious for everyone there's no way we'll get our hands on a more intelligent AI from Blizzard.
It won't be in the beta files, Blizzard is not that stupid. Especially now, when people got their hands on a working crack.


Well... we'll have to make our own. It won't be easy, but will keep us busy until the game is released.
We have a baby AI from Blizz, we only need (1 - ) people who know how to teach this baby and (2 - ) people who know what to teach this baby in order to turn him into a pro player.


So the point of this thread is to call for everyone to give their full support to those who are already working so hard to teach this baby some lessons.
As most of us have no (or limited) skills in programming, the only thing we can do is to help those working on the AIs find good strategies to make it look more realistic and harder to beat.


I have some suggestions and I think a lot of you guys will have better ones.
Here they are:
1. The basic micro management of the AI should be better worked as this is not race dependent an therefore easier to implement (at least I think so).
This will make AI lose less units and deal more damage. Atm with a little micro you can easily destroy AI's much more bigger army making it easy to beat even if it has a good build path and a strong army.
(e: attack that SCV repairing the bunker / attack with all units one enemy unit / retreat the unit being attacked while the other deal the damage etc. - more experienced players could add more to the list )
2. Protect from harassment. The current AI is very weak against resource line drops and harassment. This is also not so race dependent.
(e: some turrets/canons around the base and near the resource line)
3. Protect from cloaked units. Get detectors as soon as possible to be able to counter cloaked units.
4. Scout for expansions and attack the least defended. Maybe more attacks at the same time (hard to counter for real players).
(This 4 points are the most important imo. Even if it cheats or have 1 mil apms it will be easier to beat by exploiting this weaknesses)


5. Use a basic/popular build path. Based on the race, use popular builds used by real players (maybe someone can post some nice builds, I haven't played so much yet).
Even if it will be the same build every game. Once there is a good build you can work on new ones and make it choose the build based on the information gathered by scouting/scanning in order to counter opponent's build.
Probably this is the hardest thing to achieve and for sure it requires a lot of work and time.
This could also be divided in builds for early,middle and late game.
Probably the best way is to get a good early build for the beginning and while the other builds are being worked out we (those who are good only at playing  ) could test it and give feedback about its weaknesses.


I will test some early game P build and make another post about it, more in detail. I hope maybe someone can post some good builds meanwhile.
I'm sure those more experienced know some really good ones.
I think it will be really helpful for devs if we can gather some builds and present them like here:
http://wiki.teamliquid.net/starcraft/2_Gateway_(vs._Zerg)
(can someone working at AI scripting confirm this?)


Sorry for my English, not a native English speaker. And also for the long post (I was getting bored at work  ).


PS: Probably this is not some breaking news but I thought it would be nice to gather all this kind of info somewhere so devs will need less time testing builds and strategies and have more time for scripting the AI.

[Kernel64]

I assume it's obvious for everyone there's no way we'll get our hands on a more intelligent AI from Blizzard.
It won't be in the beta files, Blizzard is not that stupid. Especially now, when people got their hands on a working crack.


Well... we'll have to make our own. It won't be easy, but will keep us busy until the game is released.
We have a baby AI from Blizz, we only need (1 - ) people who know how to teach this baby and (2 - ) people who know what to teach this baby in order to turn him into a pro player.


So the point of this thread is to call for everyone to give their full support to those who are already working so hard to teach this baby some lessons.
As most of us have no (or limited) skills in programming, the only thing we can do is to help those working on the AIs find good strategies to make it look more realistic and harder to beat.


I have some suggestions and I think a lot of you guys will have better ones.
Here they are:
1. The basic micro management of the AI should be better worked as this is not race dependent an therefore easier to implement (at least I think so).
This will make AI lose less units and deal more damage. Atm with a little micro you can easily destroy AI's much more bigger army making it easy to beat even if it has a good build path and a strong army.
(e: attack that SCV repairing the bunker / attack with all units one enemy unit / retreat the unit being attacked while the other deal the damage etc. - more experienced players could add more to the list )
2. Protect from harassment. The current AI is very weak against resource line drops and harassment. This is also not so race dependent.
(e: some turrets/canons around the base and near the resource line)
3. Protect from cloaked units. Get detectors as soon as possible to be able to counter cloaked units.
4. Scout for expansions and attack the least defended. Maybe more attacks at the same time (hard to counter for real players).
(This 4 points are the most important imo. Even if it cheats or have 1 mil apms it will be easier to beat by exploiting this weaknesses)


5. Use a basic/popular build path. Based on the race, use popular builds used by real players (maybe someone can post some nice builds, I haven't played so much yet).
Even if it will be the same build every game. Once there is a good build you can work on new ones and make it choose the build based on the information gathered by scouting/scanning in order to counter opponent's build.
Probably this is the hardest thing to achieve and for sure it requires a lot of work and time.
This could also be divided in builds for early,middle and late game.
Probably the best way is to get a good early build for the beginning and while the other builds are being worked out we (those who are good only at playing  ) could test it and give feedback about its weaknesses.


I will test some early game P build and make another post about it, more in detail. I hope maybe someone can post some good builds meanwhile.
I'm sure those more experienced know some really good ones.
I think it will be really helpful for devs if we can gather some builds and present them like here:
http://wiki.teamliquid.net/starcraft/2_Gateway_(vs._Zerg)
(can someone working at AI scripting confirm this?)


Sorry for my English, not a native English speaker. And also for the long post (I was getting bored at work  ).


PS: Probably this is not some breaking news but I thought it would be nice to gather all this kind of info somewhere so devs will need less time testing builds and strategies and have more time for scripting the AI.

There's still some limits the AI cannot do atm. They can respond to base raids when units are there. This means if we can force the AI to leave some units while it is doing an attackwave, there should be some defense there.

I suspect the AI we have at the moment in the engine is not complete with functions that takes every condition, and thus appropriate response, into consideration.

But I'm shooting in the dark here as well.

[bustya]

No one is going to barely play vs AI when the multiplayer function comes out.

[Gamewiz]

No one is going to barely play vs AI when the multiplayer function comes out.

That's not true at all. I know MANY people who play RTS games and prefer a computer opponent as opposed to a human one.

[kassen]

true there are many that prefer a computer opponent.

[RaTcHeT302]

No one is going to barely play vs AI when the multiplayer function comes out.

Well, sometimes playing with a CPU is better and funnier than playing with a human player.

[itsarabbit]

OP is talking like there is no AI at the moment.

[MiCrOMaN1]

No one is going to barely play vs AI when the multiplayer function comes out.

No joke.

[Kernel64]

Well, I edit the AI to at least do things that  I would. This allows me to check things like BOs and econ stats timings, etc.