You could but that makes for a bad strategy game.

The whole point of having the AI play by the same rules is to allow you come up with ideas how to counter your opponent.

For example, if you blow up economic starbases and shipyards you are currently actually hurting the AI and that's exactly how it should be. Your game moves and actions within should have meaning.

If you just generate the stuff for the AI regardless of what happens then that kills your counter moves. Or, perhaps we should reward the player for these things and generate a penalty to the AI "for the estimated damage inflicted"? Again, we could but then we just have the same problem the opposite way: if the developer hadn't thought of your move in advance you can't be rewarded for it. That makes a bad game. Which doesn't mean there aren't such games. There are plenty. And generally they aren't very good.

If you can make the AI play by the same rules and with some rationality, however it accomplishes it, that creates an open-ended game where the only limiting factor are the rules themselves. You are free to make up completely new and never-heard-of strategies and they will work in the game. This is one of the best signs of a good game.

Besides, as we have established, the AI-AI-part isn't that bad. It's actually quite good at what it does. The problem lies in its poor economical planning and in the inability to react to what happens in the game. Mostly this means scripts and missing support for a host of necessary triggers and actions, and the missing computer equivalent of the micromanagement that the human player routinely applies.

Add those, plus some better ship blueprints, and you get a pretty decent AI opponent.

After you have that you can start adding AI bonuses to make the higher difficulty levels.

NO.

People shouldn't have to pay for better AI.  That is part of the purchase.

This Fall if the GalCiv III AI hasn't reached a point where it's satisfying our hard core fans, I'll get in and start working on it myself.

Brad has been involved with the AI. He's not been much involved with the scripting. Likewise, I've been vocal in outlining issues with the scripting. I've no complaints about the AI. Brad has also struggled with the scripted bits (such as when he couldn't get the computer players to build the ships he wanted - this is presently heavily script dependent, and the general commentary from inside stardock seems to imply that they'd like that to change).

Remember, when players speak of 'AI', they're talking about a combination of wildly different modules that are created by multiple people with very different specialities. It is unlikely that, for instance, a C# AI coder would be involved in xml scripting to any extent, since coding AI in C# is difficult and requires years of practice, while xml scripting can be handled by a hastily-shaven chimpanzee with twenty minutes of on-the-job training.

 This image may help:

Every red oval is a separate, largely independent module, which may have been implemented months before or after any other by programmers who may never have even met.

Now, I don't work for Stardock, and I'm sure this isn't 100% accurate (there's probably a few dozen examples on each side), but what you have is a large collection of semi-autonomous subroutines in the actual AI, and an equally large number of semi-autonomous scripts in the Scripts. So when players say 'the AI is bad', Stardock need to try and figure out from that which AI subroutine we're referring to, or which script, or even if we're talking about scripted behaviour or actual AI behaviour.

They're all basically independent; there's no over-arching 'AI' entity co-ordinating them. They take the resources they find at the beginning of the subroutine, and attempt to use them - so the diplo AI checks the scoreboard and diplo scripts to determine how it should act with another player; the tech picking AI refers to the technology scripts to decide which tech to go for, the ship designer AI looks at the blueprints and populates them with the best weapons of the types listed etc. Most of the AI's economic management is entirely handled by scripts.

The bit in the light blue on the right hand side is very good. The bit in yellow on the left hand side, where it is picking those initial conditions from, is where the problems largely lie. Brad, as a talented C coder with 20+ years of experience, works primarily on the blue stuff. Modders like myself, who have no access to the C, work on the yellow stuff.

If anyone from SD wants to chip in and contradict or correct any of this, do feel free, but this is largely my understanding of it.

Added the diagram to help illustrate what I mean. Drawings always help.

I'm not sure about others posters, but more scripts and more triggers is what I have in mind when I write about an improved AI DLC. Obviously wording could be better and more specific, but the general idea is to have a DLC that is actually an expansion. Could be bundled with a scenario editor, improved map editor, and maybe a couple more races or new race specific tech trees that the new bundle of scripts and triggers could take advantage of. 

Maybethere isn’t enough demand for this or its just a non starter for SD, but there is at least some demand.

So in summary we need more talented XML script monkeys.  

Is there an emoticon here with pictures of a monkey? The best I can find is this: 

Well, I have already received a cornucopia of information.

I understand modular programming and the overall messiness and elegance involved.  I have done a few clever things myself, but mostly I have been one of the ones who makes things work for mathematicians and programmers and such when no one else can, and that mostly involves only being able to talk to programmers about their interesting subjects instead of pursuing all of them.

I like the blue and yellow diagram.  I understand the difference between the C++ and script stuff.  So, when we send in savegames to Brad, is that helping with the blue stuff or the yellow stuff?  Or is there more of an overlap as it comes to what a user perceives as behaviors?

The notation gives that XML code "knows nothing".  It was my understanding that code at that level should have hooks into information presented from lower levels.  Things like economy numbers from the lower levels should somehow be available to upper levels for decision purposes.  I would understand it as the C++ code supplies basic behaviors in things like managing an economy and tells the XML how that economy is going with various bits of processed data.  The scripts end up encoding certain behaviors with certain decision points.  That is what I hear the forum work heading towards.  I am still not sure I understand what parts of the behaviors Stardock is affecting.  I do notice it is overall working.  Version 1.2 seems significantly better to me on some subjective level.  Since I hear little about big changes in what the XML is showing yet, I have to assume I am seeing some the C++ work.  Maybe.  I am just guessing and trying to appreciate.

One of the things I remember hearing in earliest days was something about a lot of data mining from the Steam integration.   I know a lot of that is being used for gameplay stats by Brad and already coming up with interesting and relevant data.  I was under the impression that some of it would be harnessed somewhere for some layer or another of the decision making.  Is that really a thing?  Did I hear that right?  If so, is there any clues on what kind of use is planned or possible?  It seems a potentially powerful tool for all sorts of interesting techniques.

I want to thank everyone for your great replies.  This is the most fun I have had with GC3 outside of actually playing the game.

Naselus has a pretty accurate description above.

There is improvement.  In the diplomacy screen at least the AI is a lot more demanding.  

Well frog, it seems to me that kind of DLC surely would result in alot of negative publicity but i think the idea behind that DLC is that fans that wanted it could invest much more heavily and in return, the staff of galciv3 would be able to invest much more heavily themselves with more personnel, budget, etc. For me the AI hasnt been a big problem at 1.0 on onwards, but neither have i played much after dropping it due to other perceived issues to see these improvements.

I was scratching my head awhile over this one, and ive seen you say similar things before. About how you had joined the team at so-n-so time before launch, etc, things should get better when or if you get involved more. Since you seem like ''The Man'' at stardock, and all i know of you is being a software engineer, i myself dont quite understand why the teams best player is seemingly more invested into company management. If you get a star quarterback in the NFL, you wouldnt normally pull that guy out and put him in the front office, you would want him down there doing what made him great...throwing passes, or maybe coding stuff. Im not trying to tell you what to do, i just dont understand the situation is all.

How's that for an AI coding incentive? Just be careful you don't get caught up in the positive feedback loop 

I'll also second the call for more script hooks. Modders can do some crazy stuff if you expose enough of the game to scripting...

If I had to guess, though, it's probably not that easy because GC3 should support running 100 AIs on huge maps, so perhaps every little script call could put that at risk if it's not properly engineered.

real intelligent AI (not just components that don't coordinate in any optimal way) is a very hard problem

simple modding can influence the scripting of individual components, but to actually write something that assesses the entire gamestate and decides what to do (like a "player"), you kind of need an API (eg. BWAPI) or SDK (eg. civ4)

even with those tools, it's hard to imagine how there would be a good AI that gets written for free. only starcraft 1 got stuff done 10+ years after release because of a few organized efforts (mainly for universities), but even that stuff is only strong because the AI has infinite APM, not because it plays intelligently.

even turn-based stuff as popular as civ can't attract AI modders to really make it play competently. but chess got a lot of attention because "man vs machine" and beating the world's best player was a milestone in computer science, just like sending a human to the moon and back home was a milestone in human achievement