CIV 4 - AI Requests

[dexters]

Top post updated.

Sirian, if you don't mind I've borrowed some of what you've wrote and included it in the preamble. I thought you described one aspect of the AI we want to see fixed perfectly. For most of us, its not about the challenge, but about facing a more interesting opponent.

P.S. if I run out of room on my first post, I'd like to add a second post immediately after it with a continuation. Can a mod possible fix it so that the two are in series even though post #2 will likely come many weeks after I posted the top post?

 

[Neomega]

An editable AI will allow a great deal fo fun in SP.

I played an RTS called Total Annihilation, that was a lot of fun, but even on hard, it was easy to beat up on after 3 month of play.

Then online a guy named Switek built an AI... he obviously had some experience, because it was definitley a lot tougher.

I think the Civ III AI does pretty good, considering the complexity of the game. It is, afterall a strategy game, and people are way better at stretegy than a computer could ever hope to be.

As the level of action relates inversely to the amount of time to make decisions, AI's will seem to get better. Because the only thing a computer can do good is be fast. So getting a good RTS AI is easier than getting a good TBS AI, and getting a good FPS AI is pretty simple... in fact, you could easily make an FPS AI that would be impossible to beat.. (a la Xaero for Quake III on "nightmare" level... never been beaten, to my knowledge).

 

[MMAfan]

I played an RTS called Total Annihilation, that was a lot of fun, but even on hard, it was easy to beat up on after 3 month of play.

Then online a guy named Switek built an AI... he obviously had some experience, because it was definitley a lot tougher.

Think how bad the AI was! It was only a rebalance of the weights of building units. It was a static weighting, limiting the control over it. Either forbid the AI building heavy stuff later or risk it going resource bankrubt by building the heavy stuff too early if you give it a (low) weight. The AI tweak valued the cheaper effective stuff more and had higher unit limits so the AI would usually build more units before sending them towards you.

What was really needed was being able to change the weights using a script like:

script:
resources = 100%
unit factories = 0%
if ( resource > x ) {resource = 50%, unitfactory = 50%, cheap units=100%}
if ( cheap units > 10) {cheap units = 50%, workers=50%}
scriptoff

you could follow up with interesting algoritmes to balance the fighting forces / resource income-output etc. etc.

Now this deals only with the horrible resource management of the TA-AI. Pages could be filled with complaints and solutions to all other problems. But this is a CIV forum not a TA forum .

CIV AI is actually more complex then TA AI. It functions the role of:
-opponent / companion
-simulation of behavior of civilizations
-tool (yes automating units is a sin but I would reaaaalllly like to have firaxis make worker smart enough to do stuff for you. Something as easy as being able to give a queue of orders to a worker help a lot. Not exactly AI but helpfull it is.(dont have to remember your plan you had when you gave the worker the order to mine a square))

Problem with AI (or everything) is you do 20% of work for the first 80% of effect then 80% of work for the last 20% of effect. When you have the first 80% of effect, that 80% of work looks like a biiigggg mountain to take... .

The only solution:
Devs define the scripting language, the mod communitie creates the AI.

 

[Sirian]

Think how bad the AI was! It was only a rebalance of the weights of building units.

Sometimes that is all it takes, rebalancing a few elements to squeeze out good performances. However, that is not always enough.

I worked with the MOO3 modders tweaking and playtesting the balances on that game, and we made big progress. However, there were some things in the AI that were broken, which were hardcoded. Tweaks can't fix everything, nor can mods. Firaxis needs to design the core AI effectively, or else they would have to expose pretty much the entire thing for modders to fix it.

A better AI is possible, but it may require new vision. Drawing on what has been done before in other games isn't going to carry Civ4 to the promised land. Civ3 was Soren Johnson's first AI. Despite its shortcomings, it is better than any other in the genre ever produced and WAY ahead of any previous Civ AI. I'm hopeful that Civ4 has a chance to break new ground. We'll just have to wait and see.


- Sirian

 

[Jon Shafer]

Improving the AI isn't just about making it more competitive. There is also making it behave more in line with the game's context.

I agree - but what I'm mainly talking about is improving the AI in the sense of making it smarter overall and more capable at the game as a whole, as opposed to fixing broken parts of it. An AI should be able to handle any part of a game, making certain areas "human only" means the AI is broken - and it needs fixing, not "improving" in the sense of actually making it "smarter."

Besides, your line about most players not being able to beat Regent is a red herring. A typical game of Civ sees one human player and seven AIs on a normal map size. That's 7 to 1 odds, all other things being equal. To measure the player's performance against seven AIs and claim it as a meaningful measurement of how inherently strong or weak the AI is doesn't add up for me.

Well, in Civ 3 all AI civs are against each other, so that's not entirely true. It's not until you get to higher difficulties that you see the AI getting boni for working together (e.g. cheaper tech trading). Up until that point all the AI are basically against each other as well as you. In Civ 2 it was a different story.

Players could lose scads of games and still notice the AI making "dumb" moves left or right, or lacking personality at the diplo screen, displaying a major gap in its capabilities to handle game elements, becoming predictable, or any of the other areas where it underperforms.

See what I said in response to your first comment.

Fact is, -I- did not buy a new Civ franchise installment after Civ1, until Civ3 came along with a dramatically upgraded AI. The AI was -THE- selling point for me. I would simply not have bought the game without being told that the AIs would not "play as a team vs the player" -- something I had grown to despise from the earlier Civ games with a negative intensity that could frighten dogs and small children if honestly expressed.

Limitations on the AI are why I gave up on Civ1. I outgrew it and left it behind. It wasn't its inability to beat me, but rather its inability to play fair and its inability to throw me any new surprises or keep me guessing.

I did not buy Civ2. Someone gave it to me, years after it had been out. I have never even seen SMAC in action. Sid made a great game in Civ1, but I'd been waiting a decade for the AI to get some major renovation before I bought back into the concept of Civ. Perhaps I'm alone on this score, but somehow I doubt it.

Well, uh, we both know that you're not the usual Civ player.

I repeat, the AI is the player's opponent and companion, in most cases. Your definition of "improve the AI" seems somewhat one dimensional, Trip. I'm not sure you and I are thinking of the same concepts with those three words.

Again, I think you simply misunderstood what sort of "improvement" I was referring to.

 

[yoshi]

Thing is, the Civ3 franchise has made crap-loads of cash. Why? No real competition IMO. So, it's reasonable to expect improvements. What isn't reasonable is to expect significant improvements in the AI based purely on the fact that it's supposed to be even better than Civ3. (There are elements of C3C's AI that could certainly use some patching up but never have been because there is probably no incentive to do so.)

Civ4 will be made from scratch AFAIK. This could mean that the increased flexibility of a totally new engine will allow for a 'better' AI but, as I understand it the AI could also be worse because this is a whole new animal and the results are not as predictable as they were with Civ3--which was basically an upgraded Civ2.

The rules may be different but it seems unlikely that the general format will change much (i.e. why mess with what works).

Thing with 'canned strategies' or preset AI behaviour under x circumstances is that a human player can discover what these strategies are and exploit them to his own advantage, thus defeating the purpose of single player gaming in a way.
I would personally intermix presets with random so that the AI will occasionally deviate from what is expected. Of course, this has nothing to do with AI; it's just a switch being set for the AI at random (e.g. AI preset to settle at x but instead settles at y). The AI has not made the decision based on logical factors but the player won't know the difference--he'll just scrath his head and say, "well I wonder why the Romans settled there instead of here where I thought they would?" Of course, this simplifies things significantly for the programmer but it also risks having the AI ***** itself (i.e. act contrary to reason). So, you'd have to limit the random AI effect (e.g. Caesar offers gold for no partilcular reason ONLY when he relations are very good; this random event never triggers if relations are bad). AFAIK Civ2 made heavy use of "canned strategies."

Oh, and also keep in mind that the more the Ai has to 'think' the more CPU time it takes up, hence a slower game. That said, how effeciently the program carries out all this scripting may determine the nature of the AI's 'intelligence.'

Modding the AI via Python is fine. It's just that with all the MP gaming going on now--one assumes that this mode will be vastly improved in Civ4 if only because the market craves it--AI will become less of a priority to publishers (which is a mistake IMO) thus you will get the 'it's good enough as is' response to player requests. Granted, the modding community will no doubt come out with all sorts of variations to suit different tastes. I just like the idea of having a really solid product without having to DL anything (heh, that sounds kinda dirty) but that just isn't realistic given the fast-paced environemnt these monsters are created in. Nevertheless, the core AI should at least be up to par--modders will only do so much before they start feeling like they're doing someone else's job.

As for AI features I would LIKE to see in Civ4...I'd settle for, USE ALL THE FEATURES IN THE *UCKING GAME!

...But here's something a little more specific:

- Effective unit-grouping
- Effective use of various unit roles
- Path-finding that takes only 'visible' obstacles into account
- More personality variation between different leaders (assuming they go with fixed leaders again) than has been seen up to now
- Fiscal limitations taken into account in realtion to production (i.e. generally won't build what it can't afford)

I can think of others but the best way to keep the AI functional is to just add features that limit it and the player alike.

For instance, if you don't want the AI settling all over the place, limit the range of Settlers. You could do this by just preventing the unit from moving beyond a certain point thus creating a 'physical' barrier to further settlement--the AI just has to know to not keep trying to overcome this barrier (as for instance the AI in Civ2 when sending a sea/air unit to an empty enemy city, which the unit would repeatedly try to enter but could not as neither type could occupy cities). This is a simpler alternative to limiting the AI by scripting strategies for it to follow when settling the map; the AI will just settle where it can within the parameters set by its range (from it's point of origin).

The AI and player alike are 'forced' to act within certain parameters as a result of game mechanics rather than getting the AI to function according to certain settlement patterns. Basically create shortcuts for programmers so that there's no need to write out unnecessary scripts (i.e. nothing for the AI to think about).

 

[Sirian]

Well, in Civ 3 all AI civs are against each other

That's what makes the odds N to 1, with N being the number of AI civs in the game.

If the AI were of comparable skill to the player and working as a team of seven united on one team vs lonely player by himself, the odds of victory would be approaching zero.

Thus my 7 to 1 comment is entirely accurate. If each player is playing a true "every civ for itself" approach, then each has a fair shot to win, if they are all of comparable performance. Players who can perform about as well as a single AI could be expected to win even up odds, which would be N to 1, or still mostly losses in your typical civ game.

My point is that using "doesn't beat Regent" as a benchmark is not everything you cracked it up to be.

An AI should be able to handle any part of a game, making certain areas "human only" means the AI is broken

I've been telling that to Brad Wardell and crew since January. The chief lack in Galactic Civilizations is that the AI can only pursue one goal: military. It doesn't even pretend to try to go for anything else. At least Civ3 is competent at both military and space ship victories.

I really hope to see improved AI in GalCiv 2, too, but as that game is slated to go beta in January, I'm kind of doubting that Mr. Wardell can write a revolutionary new AI by that time. And for me at least, I've reached the same point with GalCiv1 that I did with Civ1 back in 1992: "Call me when you've improved the AI significantly."

At least Mr. Johnson's AI defends its production centers. Civ3 leans more defensive, while GalCiv is all offense, both the game and the AI.

Yoshi says there's no competition for Civ, but that's not true. Both MOO3 and GalCiv came out last year. One stank, the other made a decent mark as a good game. Firaxis has more resources than Stardock does, though. Thus I think Firaxis is the better bet for coming up with a stronger AI.


- Sirian

 

[Offa]

I did not buy Civ2.

Why not. It's pretty good, and in some ways has the edge on civ3.

 

[MMAfan]

Sometimes that is all it takes, rebalancing a few elements to squeeze out good performances. However, that is not always enough.

It sure wasn't enough. Anyone that ever watched a recorded game (with the game recorder from those swedish? guys, that thing rocked) of a good multiplayer player could beat 3 AI's with a boatload of extra resources and a safe build up time without problems. It still sucked

To ever have a good AI will need a scriptable interface that has the interaction possibilities to:
-control all movement of every unit.
-control all production / working of every city
This will allow scripts to be written that make the AI MM its units well. Implement sensible impprovement building to cities. Assuming that YOUR governators use the same AI one could create custom governators for players to use...

-an elementairy understanding of strategic layout of the map.
-an elementairy understanding of the strategic organisation of an empire.
So the AI could play a bit more like a board game.

You can't let the AI play a game of chess in CIV, to many individual moves to make. The problem of the game GO but then even worse!!
If it were somehow possible to reduce all the little things in CIV to a lesser amount of data / a more high level view. Then the AI could make more intellegent human like decisions based on the info.

Of course I doubt if it all can be done...
(the part of creating the tools for the script, if the tools are there i have faith that the script will be there too soon)

Oh, and also keep in mind that the more the Ai has to 'think' the more CPU time it takes up, hence a slower game. That said, how effeciently the program carries out all this scripting may determine the nature of the AI's 'intelligence.'

One nice thing about CIV is that while (allmost) all other games intensify in the graphics department asking for more and more performance, CIV performance is basically fixed. All the extra clockcycles that keep coming every year can be used to futher the AI. I personally dont need wonder movies, the 3d iso terrain of SMAC looked realistic (the fact that higher terain gave more energy wasn't!) but was also very very confusing at times.

As long as the interface for the player remains responsive; I dont mind a few extra seconds wait at the end of turn.

 

[Neomega]

Sometimes that is all it takes, rebalancing a few elements to squeeze out good performances. However, that is not always enough.

It sure wasn't enough. Anyone that ever watched a recorded game (with the game recorder from those swedish? guys, that thing rocked) of a good multiplayer player could beat 3 AI's with a boatload of extra resources and a safe build up time without problems. It still sucked

I played enough TA online to know, that an AI could never compare to the wits and determination of an advanced player... in any game. Perhaps the time is coming, but AI programmers do not have the luxury of playing for years, and playing against people, who inevitably will find a way to massively exploit the tiniest of chinks in any AI's armor.

 

[warpstorm]

To add on to Neomega's comment. Not only do AI programmers not have the luxury of playing for years, etc.,they have to make the AI while the rules of the game are still in flux.

 

[Sirian]

Warpstorm: true! Thus it might be wise to allocate some resources to upgrading the AI after release, to tweak small details that aren't just right yet and fix the fixable exploits and gaffes. Don't spend the whole budget up front! Make sure the game gets ongoing support. (Civ3 took a year of patches to hit its sweet spot, performance-wise. That may be a good model to follow!)

Also would help to have rigorous beta testing involving game balance issues.

I don't suppose Civ4 is ready to deal with either of these points, if they aren't releasing it for another year, though.


- Sirian

 

[troytheface]

someone mentioned the popularity of mp these days, and i suspect that the game creators will cater to this market..i thought mp on civ made single player obsolete...while there are some that do not have the time or inclination for this, youth
(new gamers) will probably dictate the direction of what will or will not be implemented
..ol civ 1 and civ 2 players will be dinosaurs , lets face it, we grow older, game less,
and are replaced..i for one hope i am not one of those kinda guys that bemoan changes in the world and pine for the old days- they all sound the same. In so far as making the AI smarter, give me a break, Deep Blue beat the World chess champion- they could probably make an AI that would be unstoppable without any "help" in so far as extra units ect.

 

[punkbass2000]

I'm sorry, but are you implying that the AI required for chess is even remotely similar to that for Civ? I think you are grossly mistaken.

 

[troytheface]

no, i am implying that human ingenuity when it come to computer programing/design has progressed immensly and suspect that this will continue. In so far a "remotely" similiar...yes, i would say they are ...a computer's operation is still based in 1 and 2 a computer is programmed by people, Deep blue was programmed to beat a human player, ect. "Remote" implies that they are distantly familiar...i would argue that they are very similiar in the large scheme. The main difference i can see is that Deep blue was (in part) programmed with the help of 5 american chess masters, and was geared to beat one person in particular-seems to me that if they wanted to they could get 5 of the best civ players and attempt a similiar stunt geared towards the average civ player (after discerning what "the average civ player is apt to do-)

 

[punkbass2000]

You're missing the point altogether. A game of chess is very simple, rule-wise, compared to civ. Imagine playing a units-only game of civ on an 8X8 board. For a computer, the first 20 moves are easily 'book' moves in chess. There aren't relatively all that many options for the first few moves in chess that are 'sensible' (which is why Kasparov ended up playing slightly disadvantageous but obscure openings in order to draw the computer into making crucial decisions in the opening phase of the game, which is the most abstract.) Also, the 'board' changes in every game of civ.

 

[troytheface]

i hear what ur saying but i am thinking in a broad sense, as oppossed to addressing specifics. The limitations in so far as programming are probably based more in manpower and budget restrictions ...(and current computer processing speed) rather than doability.

 

[yoshi]

The main problem seems to be programming the AI and how to pay for that; is it even worth paying for since a) Python scripting will allow the AI to be modded (supposedly), and b) MP is becoming increasingl popular?

In the end, I think I over-exaggerated the importance of MP on the market. I would say it's just as likely that people will refrain from buying a TBS game with ******** AI as they would from buying one that lacks MP.

It's likely that a tenth--a shot in the dark--of the people who bought Civ3 don't have internet or have a connecion that is simply too slow (or expensive, if paying by hour) for MP gaming, especially considering the long waiting periods associated with this particular game. To these people, disfunctional AI would break the game, thus no sales. (This of course, is not mentioning all the initial problems with Civ3's MP mode prior to patching.)

That said, it's likely that the core AI, at least, will be sufficient for a newbie (i.e. not as picky as experienced players). Newbies will likely constitute a big chunk of Civ4 sales.

 

[dh_epic]

Nah, there's a world of difference between chess and civ. Both have turns, both involve moving units around, and some "where to move these pieces this turn", but that's about where it stops.

Because Civ is always introducing new concepts, there is always a chance that concept can be exploited. There are no exploits of the system in chess, because the rules exist and have no value. A poorly designed Civ could mean that a 0% science rate is actually the smartest thing you could do, because it's up to the designers to place different values and rewards on different things, and they can make mistakes.

In other words, the problem with designing a great AI for a civ game is that the rules are not finalized and may be imbalanced. They're probably working on the AI as they develop and modify the rules and equations that determine whether you prosper or not.

That's why a Sid level AI can exist: because the AI isn't actually conscious of many of these exploits that humans discover. So as the humans exploit more and more flaws in the gameplay, the easiest solution is to give the AI a handicap instead of intelligence.

I also think this is why a lot of people lose interest in the game (nobody on this board). Either you get bored because the game becomes an cakewalk to domination, or you up the difficulty and get pissed off that the AI is "cheating" instead of playing more intelligently.

 

[Matternich]

I'd like the A.I in cvi 4 to be able to discuss recipes for choclate chip cookies, work out my end of year tax returns and sing a sweet lullaby to soothe me to sleep after that 'just one more turn.'
Except possible Bismark as his lullaby would be too Wagnerian and probably give me nightmares.