Stardock, are you paying attention? There's a market for this out there.  

Last I heard, the devs, and Brad Wardell specifically, are going to spend the next six some-odd years making the AI better and better.  They don't need the carrot of the DLC dangling in front of them.  What they DO need is time and (useful) feedback from the community so they can refine the AI.  

So, hopefully, you'll get that AI you are craving for.  You'll just have to be patient. 

Actually, that is a quite workable suggestion to improve the AI, provided you supply "galciv community" for "marigoldran" and "pitch in" for "changing professions".

The major issue with the current AI strategy scripts is that they lack the tools to make them effective. The current scripts work blind as there are no hooks or checks available to base those strategies. Mostly they just guess by the turn count what the AI ought to be doing...

If SD started by adding lots of checks, hooks, function calls, triggers, data queries, etc. that can be used in the XML scripts as well as providing lots of responses, economy settings and overall configs that can be triggered in those XML scripts, there would be plenty of people in the galciv community to start experimenting and seeing what they could come up with simply because it would be interesting.

I can understand that SD doesn't want to give their source code away or even open up everything in the AI for business-reasons, and that's as it should be, but they could still add and open quite many things. Starting by just adding parameters and call-backs and exposing them in the XML files would also cost lot less in work hours and could be done reasonably fast compared to a full-out AI design project. And once there in the open the modders would be loose. SD could then use that as the input for what works and what would be good. The best AI strategy solution mods could even be incorporated into the base game with full honors. Or not, but at least you could build on top of the ideas that seem to work best and save time and money.

Doesn't mean SD couldn't have their AI improvement project running at the same time but they could crowd-source part of the brainstorming, experimenting and raw scripting. It's basically a win-win for everyone.

An advanced 7-year degree in computer science is the usual route.

Quoting Turkwise,

reply 9

One reply = "there's a market for this!"  You're always entertaining, marigoldran.

 

However, I admit there is a lot about this game I don't like, and it's painful because there's so much about the game I do like.  I haven't played very much since 1.0 released.  I could deal with silliness and blatant imbalances in beta, but I feel like "release" should have settled the most pressing issues, instead we are still dealing with many big problems we had in beta.

 

I disagree with the idea of AI DLC, though.  AI improvement should be part of the game for everybody, not an optional extra.

I think the point is to show, there is as much demand for improved AI and balance as there is for a Mega Event DLC. The many egregious things that even you allude to need to be fixed asap. They have not been addressed to date, yet the powers that be at SD decided to dedicate precious resources to creating the ME DLC. I believe what most individuals in this thread are saying, is they would happily pay for and improved AI DLC if that is what it takes for SD to dedicate the necessary resources.

Even the most ardent SD defenders would have to admit, the bug removal and game balance effort to this point appears to be poor at best. I'd bet they are running the risk of many players losing interest if they don't start o make progress soon.

OH, if you're talking state of the art, then it's literally years and years behind. I'm fairly sure Brad would agree with that too; he's been pretty clear that even an AI as good as GC2 DA is a best-case scenario, since it not only requires a near-unaffordable amount of effort, it also requires a lot of luck. If we're talking state-of-the-commercially-viable-art, then it depends.

The actual AI bit of the AI, the bit that makes decisions, is as good as almost any other game on the market; GC2 with expansions may be better, but not much else. The scripted part of the AI is primitive, not very detailed and not impressive at all; I can think of a dozen games off the top of my head that do it better, and most mods for most 4X games that are better, too. 

Partially, this is just a matter of more scripting being included full stop. The lack of triggers means that the strategic AI will only run one of 5 scripts at any time; the actual 5 vary from a small pool, but not very much. The strategic AI in my mod, on the other hand, runs through about 25 scripts in any given game, from a much, much larger pool. More scripts = more different things the AI can attempt to do over a given number of turns.

The other part of the problem, however is the very small number of triggers. There's a lot of stuff that the scripted AI simply can't detect, and so the strategic AI cannot react to things. It can't gear-up it's war production in preparation for an attack, since it has no 'prepare for attack' trigger in the scripts - even though I'm fairly sure that such a trigger DOES exist for the AI part of the AI, as it moves ships and suchlike to the front (now).

Now, there is a performance cost to adding more scripts, but it's not very high. There is also a performance cost to exposing more stuff to the xml, as keeping it in C is much more efficient. This means that there are limits to what we can have in the way of triggers. However, there are presently only 5 for the strategies; this number can probably be comfortable doubled without any discernible performance impact. And having twice as many triggers means twice as many things that the AI can react to - it could be told to notice that it shares a border, for example, or that it is in war-preparation mode, or that it needs to react to someone else's military build-up and adjust it's production settings, or be told to concentrate more on research if it notices it's falling behind the tech leader, or if it's researching a particularly important tech. We might even be able to tell it how to build planets more sensibly, since it's present method of picking is terribly restricted and is leading to other problems.

The part of the AI which is top-notch is the bit which SD are quite proud of,  and so they should be. Unfortunately, the scripted bit is the part that we are complaining about, because this is the bit which is making the AI fall apart in the mid-to-late game. A lot of work appears to be going into the 'fine' bit of the AI, which doesn't particularly need it; meanwhile, the amount of work going into the scripts appears pretty marginal (some re-balancing of obviously broken stuff but not any real development).

Actual strategy-level AI gets quickly hard to do the better you want to make it. The question of "where should I send these ships" isn't a simple one for a computer to solve. You are right that game AI isn't a simple thing, at least if you want it to actually offer an interesting challenge.

Fortunately, that part is actually reasonably good in GC3 and way better than in most games. If you hand the AI fleets with good ships it'll use them.

The problem is that it doesn't know how to setup an economy to get and support those fleets. This is were the "basic stuff" comes in.

Building a competitive production economy is the Big Issue. Solve it and you solve lots of other problems. The good news is that maximizing output by mindless repetitive boring micromanagement is what computers do best. They are terrible at figuring out what they should be doing but extremely good in doing it if you can tell them exactly what to do, no matter how boring it would be for a human.

If we start from a situation where the AI already has a planet with a reasonable infrastructure, that the strategic AI has already decided its specialization (manufacturing, research, etc.), and that the planet either isn't going to build anything or that the build queue is already given, then it turns out that optimizing output for the chosen specialization or to build fast is easy. All you need to do is max 2 gradient searches with the economy wheel assuming you also want to cover the "built in 1 turn" overflow and reroute it to secondary output target to squeeze out every drop from it. Basically, it's the same that a micromanaging human does except the computer calculates the optimum wheel direction instead of adjusting by eye and it won't forget it or get bored even if it did it every turn. More to the point: this isn't AI programming at all. It's just your basic optimization algorithm.

Doing that is fast but you, of course, need to do it for each planet. This means it should run in O(n) with the number of planets assuming internal access has no "quirks" but hard to see why it should. It simply needs the planetary data and the ability to calculate the production outcome from a given wheel setting.

This leaves us the initial part where the AI should decide what to build and what the planetary specialization should be. The latter is already there. The AI does specialize it's planet (believe it or not...) but the problem is that it currently takes advantage of that specialization very poorly. However, that we have now covered by our micromanagement optimization above.

The last item in the economy infrastructure planning is thus "what to build given a specialization". This could be handled by scripting that constructs the build queues. Except here we are massively lacking input data and the relevant parts aren't exposed for XML modding so we can't. SD could do it inside the binaries but I think a better approach is to expose it and open up possibilities. Consider the human player. He tries to plan ahead and build hubs as much as the planetary hex map allows and to take advantage of any special resources or bonuses. He also takes into account planetary data and builds improvements depending on how that data develops e.g. population, approval etc. The current AI doesn't. Well, it's a bit better at trying to build hubs and using adjacency bonuses now but nothing to write home about. If it actually did react to planetary data and how it develops it could do better. OTOH even leaving it as it is the simple use of optimizing specialization output would give a massive boost to the AI economy.

Add to this better shipblueprints and wartime strategies (See "Vid's Better Wartime AI" mod which has already solved many issues related to these) and a more aggressive expansion for the early game (See "Marigoldran's AI Strategy Mod" by naselus, although more of an experiment at this stage but that can be built on) and I think the AI would get pretty good...

You are aware I hope that the teams that create new content are not the same individuals,or in this case individual as I believe Brad has taken on all AI coding upon himself, as the rest of the development team.

This thread is screaming for a lock. 

Seriously Ronnie if you think you can do better, please do so. I am seriously enjoying the game but seems some folks are always unhappy. You all need to stop implying that the folks at stardock made a concsious choice to delivery what you feel is a poor ai. 

They did not.

and further more it is not (a poor ai). 

Can it be improved, of course but to come here and postulate that Stardock would rather crank out DLC maps rather than 'fix' the ai is complete manure. Ai is a work in progress and always will be. The hooks the ai needs can be added and likely will be (later this year). 

The issue is more of a poster than the general thread.  I'm mostly ok with what Stardock is doing but I'm going to be critical of the parts I don't like.  

Could be that you are just kinda coming off as a dick in the end, even if your critiques came from a sincere place.

Probably not a popular idea, but we might be better off having the AI play an entirely different game.

By which I mean assign the AI production, research, etc... per world...without tracking what is going on with buildings, adjacency,etc... AT ALL. Just use an algorithm to produce statistical output that keeps the AI competitive. When the player conquers an AI colony just populate the tiles with visually appealing ruins, make the player rebuild from scratch.

As long as the AI has the proper amount of tech and ships, who cares about planet micro-management we can't see or interact with anyway?