Editable AI - Firaxis / Soren

SithUK · July 2, 2002, 04:45

Hello,

It seems to me that a lot of people are pleading to have the AI fully open to editing?

I personally do not have the time to edit the AI. From reading the Civ forums both here and on CivFanatics I can see that there are many dedicated and very intillegent people who do. Doesn't it make sense to provide the tools to edit at least all the things that were editable in CTP?

I know the community did a lot of work whipping that AI into shape long after the development team had forgotten about the game.

Is it at all possible to give the AI set piece strategies? We all know that humans can do some nasty attack set pieces (combined invasions etc.). Is there no way to build a database of possible strategies the AI can call upon?

Lastly let me add that I think the Civ3 AI is quite playable. I feel I have more than got value for money! I have no idea if what I have ranted above is possible. Hopefully Soren could pop a reply stating a yes / no / maybe?

Thanks for reading,

SithUK

Deathwalker · July 2, 2002, 04:59

Though I think openening the AI would be a good thing, some how I doudt there'll do it

Nym · July 2, 2002, 07:56

Quote:

Originally posted by Deathwalker
Though I think openening the AI would be a good thing, some how I doudt there'll do it

It would probably be to complicated to modify this core part of the game now.

DeepO · July 2, 2002, 08:43

Indeed. that would require scripting, instead of using compiled AI. Too late for that now, plus the AI is quite decent, there is no real need to tinker with it. It's just for fun

If it would be open I'd surely have a go at certain routines, not the combat things (I'm sure other people would love improvements here), but I want more specialised worker AI (and the mood-governor could use some shaping too). As it is now, I set about 90% of my workers on automatic, using the other 10 to correct when needed, but certain things you just have to do on your own. It has improved considerably over the patches, but still... laying RR isn't their piece of cake

DeepO

dexters · July 2, 2002, 14:29

While a option to considerably alter AI may be out of our reach, I hope we can see a more modest AI editing, if not from the fans, at least from Firaxis.

For the record, I find the AI to be quite competent and satisfactory, although as mentioned above, AI's should be capable of multiple invasion strategies.

In most games, if attached on land, AI merely moves it forces over land, and occasionally drops the the surprise attack (usually a small force) behind your lines.
Of course, the AI is known to sign ROPS and attack you where you think you're safe. Good thing.

In Island wars, where enemy AI is on a different island/continent, AI appears to land in random points, usually 1 force landing in one location, while another force lands in another area in a surprise landing.

What could be improved is a set variation on this strategy.

For island wars, AI may or maynot land in two areas. They may land a combined army in one spot. Or they may land in two or three. We don't know, and we're sort of faced with what the germans faced trying to defend the beaches of France. While the number of alternatives are estimatable, defending against all probabilities could be costly, and thus adds challenge.

As far as I've observed, most wars usually have the odd unit destroying an improvement here or there, or maybe a road to a resource. They may send artillery and ships to passively bombard, but there is no active strategy on an invasion force whose sole purpose is to make the player commit forces to fight the invasion.

The AI should be randomized so that they may attack you with the intent of causing maximum damage to your infrastructure, which means, a force going on a rampage destroying improvements. Currently, the AI landings, often defeat themselves by attacking well fortified cities -- and their lust for workers, often lead then into deadly traps. This makes it rather easy for players to either bait the AI, or sit out the assault, and in both cases, the player is in control, they wear down the AI army then send their own "main force" to mop up. If the AI is programmed to sometimes attack with the intent of destroying imporvements, and they refuse to take cities and instead sit around cities destroying improvements and starving the population, the player have no choice but to attack, not only making the game more challenging, with the player attacking fortified and fresh units, it makes for a more intricate war.

Even more interesting is to have an invasion with both strategies going. One invasion attacking cities, another invasion force destroying improvements. And in both cases, AI should have some sort of a switch that allows them to adap and change their strategy on the fly. A rampaging army destroying improvements may try to take weakly defended cities once the player has send his reserve defensive units out of the cities to fight the invasion. Or the army originally set to attack cities may switch to destroying the improvements around the city first, in a siege, then switch back to attacking the city.

Any thoughts on this? Soren, is this possible / reasonable modificaiton for a patch ? :P

W4r_Machine · July 2, 2002, 14:37

It would be hard for the "common man" to edit an AI as complex as this.

Improvement of the AI is all that is needed, at least that's what i think.

SABRA · July 2, 2002, 16:55

If we start to edit the AI , will this not lead to problems that could eventually result in lets say a chrash ?

Sabre2th · July 2, 2002, 17:00

no matter what, firaxis simply can't release the code for it....they wouldn't want to let their secrets out

Thrawn05 · July 2, 2002, 22:18

Quote:

Originally posted by Sabre2th
no matter what, firaxis simply can't release the code for it....they wouldn't want to let their secrets out

A year or two ago, Rebellion released the code for the game AvP:GE. I've played around with it, changing not only values, but trying to make the AI and in particularly, the AL|EN bots better (at least I think they are

).

For a few months, I've been open with the idea that eventualy, down the road, Firaxis should release the code. I would think when it's 2004, Firaxis sholud consider this.

DeepO · July 3, 2002, 05:13

Even if I would love it, releasing the source is not going to happen. That's why I said that you need scripting if you want to adapt the AI: this way you'd be able to change / add rules, while not being able to 'spy' on the inference engine design.

Scripting is not something you just include afterwards, either you design for it from the start, or you don't use it. And it will be a lot slower, the AI is reasonably fast now for a complex thing; you don't want to wait half an hour in between turns, do you?

SABRA, surely it is possible to make an unstable AI while playing with it, the biggest risk not being that the game would crash, but that your AI gets stuck into loops: for instance a unit between two cities that each turn will switch the intention to attack between the two cities. So it will go one step forward, one step back, etc. Very hard to debug...

W4r_Machine, it may be hard for 'the common man' to have a go at the AI, but OTOH it is extremely hard to make decent graphics, and this is open for modding. Very few people are good at it, but it brings something to the community if you can play with user-generated terrain. Conceptually I don't see much difference between graphics and AI, so allowing to change one should mean you allow to change the other. However, there are some practical considerations that make it much more easy to make graphics changeable.

DeepO

W4r_Machine · July 3, 2002, 14:30

Quote:

Originally posted by DeepO

W4r_Machine, it may be hard for 'the common man' to have a go at the AI, but OTOH it is extremely hard to make decent graphics, and this is open for modding. Very few people are good at it, but it brings something to the community if you can play with user-generated terrain. Conceptually I don't see much difference between graphics and AI, so allowing to change one should mean you allow to change the other. However, there are some practical considerations that make it much more easy to make graphics changeable.

DeepO

AI has very little to do with graphics. I don't even think there's ever a game producer that released its AI coding, game engine(i.e. quake 3) yes, but not AI.

No dought it will add something to the community and add a level of play, but releasing the code is not worth it from the developers side.

AlrightyThen · July 3, 2002, 15:42

The type of AI you seek does not exist. Before you see it in games, look for it in 'real-life' military hardware, then consumer products, and even then we're talking autos, PC's, etc.

Games will be low on the list, so right now we are stuck with scripting languages that necessitate computing power. The solution for our immediate gaming future is probably mulit-gigahertz computing.

Hang in there!

CapTVK · July 3, 2002, 16:15

Why bother fretting about Civ3 AI code when you have complete access to Freeciv AI code now?

Why not help out with that?

http://arch.freeciv.org/freeciv-ai-200207/msg00000.html

Immortal Wombat · July 3, 2002, 17:36

Quote:

Originally posted by W4r_Machine
AI has very little to do with graphics. I don't even think there's ever a game producer that released its AI coding, game engine(i.e. quake 3) yes, but not AI.

A full 'release' isn't necessary. The algorithms of the CtP1 and 2 AIs (both different) are open and known, and have enabled the modding community to substantially improve the AI performance.

Of course, that won't be needed in Civ3, but scenario AI are notoriously poor, because the default strategies won't work.

Thrawn05 · July 3, 2002, 21:25

Quote:

Originally posted by W4r_Machine

AI has very little to do with graphics. I don't even think there's ever a game producer that released its AI coding, game engine(i.e. quake 3) yes, but not AI.

No dought it will add something to the community and add a level of play, but releasing the code is not worth it from the developers side.

Did you read my post?

DeepO · July 4, 2002, 09:57

Quote:

Originally posted by W4r_Machine
AI has very little to do with graphics. I don't even think there's ever a game producer that released its AI coding, game engine(i.e. quake 3) yes, but not AI.

I know that, of course. I was pointing out that on a conceptual level, releasing graphics is the same as releasing AI rules. You don't need to know how the graphics engine works to make graphics, as you don't need to know the inference engine for making rules.

However, in Civ III graphics are editable, and rules are not, as they are compiled into the .exe, while graphics are built at initiation of the program. I don't dispute that. I was just pointing out that for some it is easy to make decent graphics, while others (maybe a smaller group) are more able to tinker with the AI. I know I suck at graphics, while I work in AI (so I should be able to understand that)

Quote:

Originally posted by AlrightyThen
The type of AI you seek does not exist. Before you see it in games, look for it in 'real-life' military hardware, then consumer products, and even then we're talking autos, PC's, etc.

I don't know who you were replying to, but I don't agree with that statement. Military AI is no different than commercial AI, and game AI is about the most modern you can have. The only difference is in how much time goes into development of the 'content' of the AI, military systems are typically far larger than game AI, which also means they are far more complex. In my experience, commercial AI (and with that I don't mean the basic car-AI but for instance plant-control systems to control chemical installations) is the most complex, and up to date of all.

People think that the field of AI is very new, but the type of AI used in games is basically the same as when Marvin Minsky was researching himself, instead of only describing other's research. It hasn't changed, apart from a new methodology and new accents, for the better part of 30 years. The reason why game AIs are progressing is that computing power has augmented drastically, and certain libraries of known algorithms are kept: you don't have to invent the weel over and over again when you want to build an AI.

There are some new techniques, but they are under-used in real systems, and remain largely a research topic. With that I'm referring to agents, certain types of neural nets (SVM, SOM), and the shift to understanding human knowledge true knowledge management instead of trying to mimick that with 'dumb' rules. These, however, have no application in games.

DeepO

W4r_Machine · July 4, 2002, 15:29

Quote:

Originally posted by Thrawn05

Did you read my post?

I missed that one. Was it the raw coding of the AI? just curious...

Anywho, If you know C++, Java or whatever programing language then go for it. Or, someone can make GUI so some of us who don't know jack squat can edit the AI.

Partially realesed codes for previous games wouldn't help the current one. It's a different code all together.(maybe not)

Thrawn05 · July 4, 2002, 17:30

Quote:

Originally posted by W4r_Machine

I missed that one. Was it the raw coding of the AI? just curious...

Anywho, If you know C++, Java or whatever programing language then go for it. Or, someone can make GUI so some of us who don't know jack squat can edit the AI.

Partially realesed codes for previous games wouldn't help the current one. It's a different code all together.(maybe not)

It was the full code for the avp.exe file.

And I think a GUI would be unresonable. The AI is currently hard coded, and I dought Firaxis planned to allow the AI to be edited that way. The source code would be easier to release then a GUI editing program that constantly needs to compile the new EXE so your changes can be in effect.

Solver · July 5, 2002, 12:52

I wonder why don't AIs get released. I don't care too much for the Civ 3 source code, except for the AI code. Apolyton has a very strong community, including many able programmers. All due respect to Soren, he did great job, but a dozen or two of programmers will soon make the AI do even better things, and will certainly correct the main weaknesses of Soren's AI.

However, Firaxis is not doing this, I'm sure of it, they've even mentioned it.

Thrawn05 · July 5, 2002, 19:32

Quote:

Originally posted by Solver
I wonder why don't AIs get released. I don't care too much for the Civ 3 source code, except for the AI code. Apolyton has a very strong community, including many able programmers. All due respect to Soren, he did great job, but a dozen or two of programmers will soon make the AI do even better things, and will certainly correct the main weaknesses of Soren's AI.

However, Firaxis is not doing this, I'm sure of it, they've even mentioned it.

I agree, at least the AI. I'm no open code freak, in fact I defend Microsoft when it comes to the states trying to do just that, but I think that when Civ3 has come and gone, in a couple of years, Firaxis should consider it. I have lots of fun with the AvP and Quake I codes, but I want to take a look at a TBS/RTS AI.

I'm in no rush, I'm in the middle of getting Cavedog to do the same with their Total Annihilation AI.

XarXo · July 5, 2002, 20:01

I agree

Solver · July 6, 2002, 07:15

Thrawn, I have published the whole 210 page AOE 2 AI, BTW...

When will XarXo be restricted and given a 10% PCR? I feel soon...

zulu9812 · July 6, 2002, 07:37

Quote:

Originally posted by Solver
When will XarXo be restricted and given a 10% PCR? I feel soon...

Yes, it's begiining to annoy me too.

As far as AI scriptig goes: Immortal Womabr, you are correct. Strategy AI's are notoriously bad at handling scenarios, simply because the player (either human or computer) is often called to do things outside of the normal gaming conditions. In CIv3, with so much importance placed on the AI flags for units, e.g. np flags selected or all greyed out means the unit won't be used. a similar problem could arise with scenarios.

Solver · July 6, 2002, 07:45

Well, when programming an AI, it's about telling it what to do in any situation. Like how to start the game, what to do when you're behind in tech but with a superior army, etc.

Scenarios, however, are usually very much different from a normal game, causing the human player to play much different, but this harms the AI, which still tries to play in a standard way. This is why scenarios are generally for MP, the most, erm, non-standard ones.

Jon Shafer · July 6, 2002, 13:00

I completely agree with the things that dexters has brought up. That and I agree with the concerns and realities that DeepO has mentioned. Scripting/AI editing is out of the question for Civ III... I'm sure most of us know that. Firaxis... PLEASE make that available for IV though!

DeepO · July 7, 2002, 11:29

Quote:

Originally posted by Solver
I wonder why don't AIs get released. I don't care too much for the Civ 3 source code, except for the AI code.

Solver, I think that part of the reason why game developers are rarely releasing the AI is that it can be used again in other games. Graphic engines have to evolve very rapidly because of the evolution of graphic cards, standards, etc. You can savely release the source of a graphic engine after the game came out, because by the time a competitor can implement (parts of) it in his own game, it is outdated. But, as AI does not evolve that quickly, it would be a bigger risk to release it: part of the AI of Civ3 could be used for several years, maybe even in Civ4. Developers don't want to release code that can be turned into commercial, competitive products.

Another reason, I think, is that developers underestimate the number of fans that can program in AI. It is a smaller group than graphic-people, because you need quite a bit of experience to program it well. But, most AI-devs I know play games, and many play strategy games. I'm sure there would be quite a bit of AI-tinkering if Firaxis would release the AI code. Look at Quake I: I've seen quite a few scientific papers where that engine is being used to program better bots, it is a challenge for researchers in planning, while it remains fun to test it out. It is becoming a de facto test environment standard, which only helps to build a name
( for some of those articles, look at
Citeseer ) If only the Firaxis team would see the light

DeepO

Solver · July 7, 2002, 12:08

DeepO,

I do not think that such code, if released, could be used by other companies. Obviously, Civ 3 AI code is only useable in a civ-style game. So even if another company did something like this (though after Activision's fiasco I strongly doubt it), they got no right, since the code is copyright.

They probably don't underestimate AI programmers, either, they see how big some communities, like Apolyton are.

Firaxis, come on

.

DeepO · July 7, 2002, 13:37

Solver,

I'm not talking about specific rules, but about the rule engine. That's one thing that is similar for all TBS games. And while specific rules can't be used as is, I'm quite sure many of them are more general, and cover many games. For instance, many of the rules could be used in MoO. Of course, you don't have specific land-units combat planning in MoO, but at least you have (in general) also combat planning spanning multiple turns.

Another common thing to all TBS is the pathfinding algorithm, but these are so well known that I find it very hard to believe that any game developer can build a crappy pathfinding into it's game. Still, it happens, and this only signifies the lack of interest from that developer in AI. I'm very happy that Firaxis is not like that.

I'm sorry to have confused you by posting about two things at once: the quake papers mostly dealt with specific rules, and only in certain cases with changes to the engine. But still, they're nice examples about what a game can be used for after it served it's commercial role.

I heard that Apolyton had an active community of modders for CtP, but as I was (and still am) opposed to that game out of principle I never looked at what has been done, let alone participated in modding. I can't judge how many people were active in the AI field, but reading some of the comments on this board made me feel there are just a handfull of people here capable of AI programming. It is a completely other way of thinking than traditional programming...

and about copyright: of course no game developer would be allowed to use it, but after the 'stolen' code is compiled, who is going to find out? From the source it may be obvious, but from an .exe?

DeepO

Peter Triggs · July 7, 2002, 13:38

Quote:

Civ 3 AI code is only useable in a civ-style game.

Solver, maybe not. Apparently, the AI algorithm used in CTP2 is essentially the same one that was initially used in Dark Reign (an RTS game, I believe), CTP1, and parts of Star Trek Armada. It sounds like an AI algorithm can be abstract enough to be used in all sorts of games.

DeepO, this algorithm uses a technique that was described as "priority based list processing". Do you know anything about this technique? I haven't been able to get any general information on it.

DeepO · July 7, 2002, 14:00

Peter, I can't say for sure as it is a company specific term, but I think that the term would be known to others as a blackboard system.

Basically, you have a number of possible tasks, which all get put on the blackboard. Then, you have a second algorithm which will give scores to each task, and rank the tasks according to the score. So, the first (most important) task is performed, and wiped from the board. Then the next, and so on.

This is like what happens with the workers in CivIII: when you put workers to auto, each will see what can be done (I'm not sure whether this is decided on a game level, or on a worker level). For instance, a worker sees that there is some pollution, and there are 2 tiles that still need to be improved, one of which is in use, the other is not. It will put these 3 tasks on the blackboard.

After this, a scoring algorithm gives for instance a score of 1 to improving the non-used tile, a score of 5 to improving the worked tile, and a score of 100 to cleaning the pollution. So, the worker will start on the pollution, as it has the highest score.
This scoring algorithm can be quite complex, certainly with multiple tasks. For instance, the third free worker could also be put on the pollution, but as there are 2 workers working on it already, the score of the pollution would be only 3. So, it will clean that tile of pollution if there is not a tile that needs improvment, which is in use by a city. In this case however, there is such a tile, so only 2 workers will go to the pollution, and the third will road the unimproved tile.

This happens throughout the game, each unit can do multiple tasks, and selects one based on the priority it receives. Additional tasks and the priority it should receive is where modders can help, while the tinkering with specific priority parameters (should pollution be 100, or only 10?) is what makes playtesting needed.

Also note that in many cases, such systems are given time-bonusses. Even if you have so much pollution that your workers are only cleaning pollution, the task of improving a used tile gets a priority of +1 each turn it stays on the board. So, after an amount of time it gets a higher priority that cleaning pollution, and will be done first.

DeepO

PS: If CtP2 used the same algorithm as RTS games, no wonder it performed poorly. Those two genres are so far from each other, real time AI is very different from traditional AI. Of course, it could both use blackboard systems, but the implementation and tweaking would be very different...