Balance vs. AI

[Armuss]

Chances victory (Fall Weiss) by game vs. AI is:

SC2 (Allies/Axis) 3:1 - AI is too weak, problem is not to decide...

SC2:WaW (Allies/Axis) 1:1.25 - my opinion: USSR is too weak.

As there is human vs. human game?

Armuss

[ May 28, 2008, 02:50 AM: Message edited by: Armuss ]

[G B Scurlock]

??? I dont follow you here.

[arado234]

I dont think its possible to make any A.I.good enough to beat a good human player unless maybe its the first game,otherwise once you know how it will react to a certian move you can setup as many ambushes as you want.A.I cant reason.In most cases it just reacts to your moves which puts it at a big disatvantage.

As far as A.Is go this is by far the best ive ever played.

[pzgndr]

arado, I respectfully disagree.

In many cases the AI reacts to what you do and can choose a course of action and choose what forces get assigned to a plan, just like a human opponent on the other end of a TCP/IP connection. You have to react to that. And with FOW on you don't know what the AI is doing until you confirm or deny what it's doing. Just like playing a human opponent, yes?

Something I tried to emphasize in my AI articles was that SC2 really brings together a "perfect storm" of features for a challenging computer opponent: decent generic AI based on fuzzy logic, AI scripting including decision events and global variables, and event scripting for randomness and variants. You cannot possibly know for sure how the AI is going to react to your moves. And that is the beauty of the whole thing. :cool:

It's not perfect yet, for sure. But it's getting there. Hubert's got a great system developed which can only get better.

[arado234]

I agree with some of what you say pzgndr but from what ive found once you have played it a few times it will generally react the same to whatever you do,meaning if it beat you once at a strategy you tried by doing the same moves again and prettywell knowing what it will do you can usually set up a trap and wipe it out.A human player may react totally different.

Yes you are absolutley correct about it being very close to as perfect as it can get.Yes Hubert not only has a great system developed but the best I have ever played.

[Youri]

I recently bought an electronic chess game and was surprised to see it was impossible to beat the AI at entry level (countrary to the older version I bought 15 years ago...). Apparently AI improvement results mainly from storing more and more pre-programed game situations (especially begining/end of games). Following the SC2/PDE game I just finished agains AI, I believe some improvements are still possible !

- by storing more situations of global strategy (for instance Axis should leave north Africa once its position on the continent is too weak, or it should retreat from eastern europe once Germany is threatened by an invasion invaded

- by improving logic at tactical level (for instance, in north Africa, I managed to stop Axis which used during more than a year an anti-tank unit to attack my infantry (had it used a tank unit instead, Axis would have won north africa)

Another remark regarding the speed of the game which I find too slow (perhaps due to the number of the scripts, or the programing language if interpreted & not compiled)

[Hubert Cater]

Hi Youri,

The speed of the turns is likely due to all the scripts as well as the map size and number of units. Unfortunately even for 2D games things can slow down quite a bit as the path finding algorithms can be quite expensive. It is a balance between accuracy and speed but something I am constantly working on to improve.

Hubert

P.S. For reference the game EXE is compiled and comparable in speed to any C programmed executable as it actually compiles to C byte code.