Civ AI sucks

Ok, so you spend a few paragraphs talking about how the AI does such a bad job when it's crowded, and then sum up by claiming the AI isn't really that bad? If you need to give the AI plenty of space, and don't want it to deal with lots of diplomacy, that's not really a convincing argument for the robustness of the AI.

Bh

And instead of refuting my points, you are just begging the question. I didn't claim the AI was robust. All I said was people often underestimate the AIs because they are playing under conditions favourable to them. I really have no idea what you can possibly be objecting to here.
 
I played against 17 civs on a huge map on Immortal difficulty... I knew it should get easier over time, and it did AFTER I survived the BC era inital warmongering phase.

But the start was very difficult, definitely much more difficult than less number of opponents. Fighting vs 4AIs in the 2000-1000BC aint so easy, 2 of them being Genghis and Monty... you can check the writeup in my signature..

Today, I tried several games at deity difficulty... I again tried to make the map most populated possible by the HOF rules... I thought since it worked at Immortal, it should work at Deity as well... result?

well, the AI expanded everywhere so quickly that I could not build enough cities to produce any significant force.

Then, I reduced the number of opponents(lowest allowed by HOF rules) and was able to expand much more successfully.

so, obviously there is an optimum number of civs to play against at certain difficulties, and it varies from difficulty to another. I don't agree with cliche comments that overcrowding is always easier for the human player.

I agree. But this is due to the AIs' initial bonuses. The lower their initial bonuses, the more handicapped they are right from the start. So, unless you are talking about very high difficulty levels, it isn't a big problem, especially not on Monarch and not even on Emperor (Immortal might be different, but from my experience, adding a few more AIs makes the game easier). I still believe that the human player is capable of focusing production in such a way that as long as luck isn't too bad, he can carve out a competitive existence, whether by chopping settlers to REX asap and/or rushing his neighbours, and then surpass the weakened AIs from then on.

And, ironically, the cliche is more AIs = more difficult, not the other way round.
 
And instead of refuting my points, you are just begging the question. I didn't claim the AI was robust. All I said was people often underestimate the AIs because they are playing under conditions favourable to them. I really have no idea what you can possibly be objecting to here.

I'm not "begging the question". I'm just saying if you're going to put forth an argument that the AI is being underestimated, it's not going to be very convincing if, as part of that argument, you spend a great deal of time explaining how the AI does poorly under various conditions.

And it's not a matter of "favourable conditions". Favourable conditions would be restarting the map until you get a great starting location. Or lowering the difficultly level. Adding more AI players to the game shouldn't give the player an advantage.

Bh
 
I'm not "begging the question". I'm just saying if you're going to put forth an argument that the AI is being underestimated, it's not going to be very convincing if, as part of that argument, you spend a great deal of time explaining how the AI does poorly under various conditions.

And it's not a matter of "favourable conditions". Favourable conditions would be restarting the map until you get a great starting location. Or lowering the difficultly level. Adding more AI players to the game shouldn't give the player an advantage.

Bh

But you are. Again you assume or make it look like I'm saying that the AI is underestimated. What I'm saying is some people underestimate the AI because they are playing under certain conditions.

We all know the AI has limitations. It overcomes these limitations with bonuses, and one of the more crucial ones is its ability to expand with relatively little penalty. With so much land, the AI has more economic room to manuever, making it more competent. If the AI is unable to compete because it doesn't get this advantage, it is because of the limitations of its programming. Should or should not is not the question. The reality is the developers of Civ do not have the budget nor the time to achieve the ideal standards people like you have in mind. The reality is overcrowding the map is crippling the AI. I'm not interested in debating normative terms and statements with you.
 
The reality is overcrowding the map is crippling the AI.

Which does nothing but support the OP's position. I don't care about your interpretation of the AI, or overcrowding. What I do care about is the fact that you put forth the above opinion as if it somehow invalidates the OP's position, when instead, it does the opposite.

If the AI is unable to compete because it doesn't get this advantage, it is because of the limitations of its programming. Should or should not is not the question. The reality is the developers of Civ do not have the budget nor the time to achieve the ideal standards people like you have in mind.

And how does this counter the argument that the "AI sucks"? All you are doing is making excuses for why it "sucks" (assuming that the OP is correct, which obviously I don't).

Frankly, there are numerous arguments for why the AI doesn't suck that you (or anyone) could put forward. Suggesting it's because some people put too many AI in the game isn't one of them.

Bh
 
Which does nothing but support the OP's position. I don't care about your interpretation of the AI, or overcrowding. What I do care about is the fact that you put forth the above opinion as if it somehow invalidates the OP's position, when instead, it does the opposite.

Because the OP stated what settings he usually plays under and I'm just saying that I'm not surprised that he finds the AI underwhelming. It's not because he is such an expert at beating the AI, as he seems to imply.

Bhruic said:
And how does this counter the argument that the "AI sucks"? All you are doing is making excuses for why it "sucks" (assuming that the OP is correct, which obviously I don't).

Frankly, there are numerous arguments for why the AI doesn't suck that you (or anyone) could put forward. Suggesting it's because some people put too many AI in the game isn't one of them.

Bh

Well, I'm saying the AI doesn't exactly suck. It does have limitations, and if you grossly exploit these limitations, intentionally or not, you'll find that the AI will most definitely suck.

I don't really understand what you are arguing for. It seems more like you are trying to pick a fight here.
 
Well, I'm saying the AI doesn't exactly suck. It does have limitations, and if you grossly exploit these limitations, intentionally or not, you'll find that the AI will most definitely suck.

To which I'd agree.

I don't really understand what you are arguing for.

I'm arguing about this line: "Some people, and not just the OP, have the wrong conception of how stupid the AI is because they are overcrowding their maps."

It's not a wrong conception about the AI. It's an entirely right conception of the AI. If the AI is unable to handle common game situations, that's a flaw in the AI. And pointing that out is an entirely valid position. Dismissing their position simply because they happen to have selected some more AIs to play with isn't, imo, an acceptable argument.

Bh
 
still- the non-addressing of an improved AI without bonuses remains.

Strange. considering chess programs, which have been around since computers started, had levels of difficulty 20 years ago that did not involve giving the computer an extra Pawn, or even the first move.

I suspect that there may be an issue of not making the AI to good.
In such a case the AI advantages via bonus may be the only way they could make it stronger yet still beatable.

Troy,

You may not realize this, but chess AI was pretty terrible in the beginning also. Indeed, building chess computers became a major focus for computer programmers for years and years. I hope you enjoy this story.


The 'big era' of early chess AI started in 1968 (I think, it was along time ago) when Master David Levy made a bet with the programming world that no chess computer would beat him in 10 years. He was a 2300ish type player, not a great player by chess player standards (of course, he could clobber me easily) but FAR FAR from Petrosian, Spassky, or Fischer.

The computer programmers scoffed, saying that Levy was arrogant. Of course, they are the ones who proved to be arrogant. I was an 1800ish player, and I remember going to tournaments in the early days of computers, talking to programmers. I remember talking to some of the leads in 1975, and they admitted that 'oh, chess is a lot harder than we thought.'

To catch up, they started to 'cheat'. The programmers gave up on having their computers 'calculate' the opening, they started to type The Encyclopedia of Chess Openings (ECO) directly into the code. Originally, they claimed they didn't need to do this but learned better.

Well, 1978 came and went, and the best computers were about my strength, 1800ish or so. Remember, these were played on mainframes, we aren't even talking about PC's.

My chessplaying brethren and I had a good time on this -- yes, people do underestimate how good the REAL chess players are. They can't beat Levy, how could they even CONSIDER beating Karpov?

I remember going to an AI convention for work in 1986 or 1987 (by the way, this was REAL AI, not game AI. This was the era that AI applications were considered a promising technology, which for the most part hasn't proven out.) By now, the computers of course had well passed me, and even Levy, but again the programmers learned something -- going from 1800 - 2300 is a LOT easier than going from 2300 to the Karpov/Kasparov level.

I was shocked when the keynote speaker said, 'The most important development for the advancement of the application of AI programming in the general public would be to produce a computer that can beat Garry Kasparov at chess.' Talk about priorities!

Well, then chess AI got a boost that no game ever had or probably ever will get. $$! IBM started to work big time on building efficient chess computers. I suspect that this was largely done at IBM to try to advance the art of programming and applications of computers, but the resources spent on building chess computers now entered a frenzied stage. And was some of it for the pride of the computer?

Chess players like me always knew that computers would pass my chess heroes, but they put up a really good fight. By the mid-90's, the computers were doing hundreds of thousands of calculations/second. By this time, the programmers got smart, and hired real chess GM's to review their games and see if they could improve the AI (note that this isn't cheap either. )

By the mid 90's, the computers were real good. Again, I talked to the programmers who now had GM's at their side. They knew they could beat any player in the world -- at that point, except Kasparov! Now, the programmers had learned how kick ass the best chess players were.

And of course, now programs like Hydra can beat most GM's spotting them a pawn.


My point here is that it is unfair to compare the AI of chess to a commercial game. The resources spent on developing chess AI are far greater than any PC game can ever spend. As others have pointed out, chess is a game well made for a computer's calculations. Your version of Fritz or the Chessmaster also have the advantage of programming going back in a large way to 1968, hundreds if not thousands of people at some point working on development, the typing in of ECO and similar texts, and millions upon millions of dollars. I suspect if Firaxis had those resources, it could build some REALLY good AI.

So don't expect that kind of skill from any game you buy!

Best wishes,

Breunor

PS I see others have talked about Deep Blue's budget. Sorry for repeating this! But the money spent on chess AI is even far greater than even Deep Blue's budget since this was built on years of other programming!
 
I'm arguing about this line: "Some people, and not just the OP, have the wrong conception of how stupid the AI is because they are overcrowding their maps."

It's not a wrong conception about the AI. It's an entirely right conception of the AI. If the AI is unable to handle common game situations, that's a flaw in the AI. And pointing that out is an entirely valid position. Dismissing their position simply because they happen to have selected some more AIs to play with isn't, imo, an acceptable argument.

There's a difference between pointing out the flaws in the AI and labelling it as completely unsatisfactory, especially when the latter is coupled with a superior undertone, as often accompanies such claims. If such vehement labelling proves to be unjustified, as in this case, since the OP is virtually abusing the AI's widely known limitations, then you can expect a fitting rebuttal.
 
yes, an interesting tale, brings back memories. I happened to read Levy's book years back and remember too his prediction.

(can't believe anyone else read that book....some of those chess guys are brilliant..think the state champ here is a doctor- one of my greatest moments was beating a lawyer after three hours when he had a pawn on my 7 rank-)

Without underestimating will, money and brains (a title to refute guns, germs, and steel, i suggest that the exponential increase of computers capabilities will negate some of that criterea as a necessity (sp) or as a prerequisite in order to produce a highly capable AI.

on a side note i think the AI is good, and the attempt to make it play "more like a human" is an inspired move in so far as making a game fun- but may contradict making it stronger. Maybe they ought to make it play with the intellect of a man- the cunning of a fox, and the work ethic of a bee.
 
yes, an interesting tale, brings back memories. I happened to read Levy's book years back and remember too his prediction.

(can't believe anyone else read that book....some of those chess guys are brilliant..think the state champ here is a doctor- one of my greatest moments was beating a lawyer after three hours when he had a pawn on my 7 rank-)

Without underestimating will, money and brains (a title to refute guns, germs, and steel, i suggest that the exponential increase of computers capabilities will negate some of that criterea as a necessity (sp) or as a prerequisite in order to produce a highly capable AI.

on a side note i think the AI is good, and the attempt to make it play "more like a human" is an inspired move in so far as making a game fun- but may contradict making it stronger. Maybe they ought to make it play with the intellect of a man- the cunning of a fox, and the work ethic of a bee.

I think this is a VERY good point. Thanks Troy!!

I think that in the intermediate stage of computer development for chess, there was a big debate in the programming world. Indeed, in the 1950's, Claude Shannon predicted two types of machines for playing chess (which then, as of now, was a benchmark for the progress of computers). One was brute force (which Shannon called type A), and the other was to search carefully. Shannon predicted that type A would fail.

Of course, in the real world a combination of 'brute force' and 'looking smart' are always going to be employed. So, typing in ECO was the ultimate in intelligent search, completely ignoring the internal evaluation algorithm. Now, there is huge controversy over the typing in endgame solutions.

People who are more 'up' with the modern state of the best chess computers can probably help here, but I do remember reading that the big computer contenders do use different technologies and approaches toward play.

I don't think that there is any way to avoid the conclusion that computers are much better at chess both because they have learned how to evaluate positions (remember those hired GM's) and of course the work of Botvinnik in the 60's, and also from tremendous computer power. Clearly, those computers that held me even at 1800 ish in 1975 with the exact same code would be much tougher on a modern computer, it would lick me off the board.

And so we have it with computer games too. Should AI developers try to get the computer to 'think' like a human? Or should it recognize that the computer is more of an idiot-savant, and take advantage of its ridiculously powerful calculation as its base? I don't have an answer to this and I don't follow these arguments.

I would be interested to see what AI programmers like Blake think about this!

Best wishes,

Breunor

PS Never wanted to play lawyers, I always was afraid I would get sued if I won, especially if he had a pawn on the 7th ..... :)
 
LOL I must stink pretty bad. I have a terrible time winning Noble and get my butt kicked every time I've tried higher. I don't use any mods, just plain old Civ 4...got the expansions on order since I hadn't gotten around to getting them due to being addicted to WoW for the past 2 years, but got the Civ bug again. I love Civ and I've always picked it up for a whirl every now and then since original Civilization (I own all of them, including SMAC). I just never have been able to get good at them =(

Haha, don't feel bad, I know I am not very good. But its okay, I still have fun playing Civ, and that's all that matters to me!
 
When I said you could string a bunch of "if-thens" to improve the AI, I wasn't talking about strategic decision making that could be exploited, I'm talking about the obvious and flagrant mistakes the computer makes.

Examples of these would be things like building national and even world wonders in stupid cities (or starting them without a realistic chance of winning), questionable improvements, and DOW vs opponents who FAR outstrip them and are ready. Having the computer NOT do these things wouldn't allow for any particular exploits...indeed it should make exploits of other situations harder.

Applying high end game-theory to the creation of a computer AI for civ4 might be a bit much, so for situations where if-thens would actually be exploitable, you could just program the computer to weight about 2-5 options that are tactically viable and pick one at random (maybe weight them based on leader personality). Of course, if there is only one optimal choice, it's easy. If it's a wash or they're close, just randomize it so players do not know what to expect. Will the computer use the stack to pillage, defend, attack directly, or flank around via a naval assault? I almost never see same-continent AI's backdoor me via the oceans, but just having there be a chance of that would make warring much harder!

I'm sure the game already does this (I don't have BTS, but there must be an algorithm like this for when you mass troops on an enemy border and such), but it could certainly be streamlined further...or they can just lump them bonuses, which is MUCH easier to do once you have decent ai :).
 
like aelf said, AI has limitations that human can learn and exploit to make AI seem dumb. One reason for the limitations is that altho there are different leaders, the AI code is the same. Different leaders just make different decisions at branching points. that means they are different, right? well, yes, but not as different as they could be, and therefore human players don't have to defend against crazy strategies like "aim for all wonders," "first to lib and rifle," or even just "axe rush."

So I think AI is not truly stupid, be definitely predictable. human players kinda know what to do after playing against AI for a while. for example, we know what tech AI are least likely to research. we know that we don't need to put 4 defenders in every city, and that if we just keep our power somewhat respectable, AI is not likely to attack, etc.

I think having different sets of AI codes (maybe one geared toward rushing, one trying to get to a specific tech first, one aimed at building pyramid/GL, etc) would help make AI harder, or at least less predictable. but that's hard to do because it takes more effort to code and to maintain. and even then human will still be able to identify what kind of AI they are getting after playing for a while. So ideally, AI needs to behave more like human in order to pose a challenge, and that's near impossible, becasue doing that means learning from it's mistakes, predicting opponent moves, identifying and exploiting the synergies of leader traits, starting tech, and the starting location, and checking civfanatics forums everyday to learn the new tricks.

all that, and having the ability to reload, reroll, reset :P
 
And so we have it with computer games too. Should AI developers try to get the computer to 'think' like a human? Or should it recognize that the computer is more of an idiot-savant, and take advantage of its ridiculously powerful calculation as its base? I don't have an answer to this and I don't follow these arguments.

I would be interested to see what AI programmers like Blake think about this!

Best wishes,

Breunor

Well I wouldn't consider myself an expert AI programmer, but I have written half decent Reversi and Checkers programs, and what I know about search algorithms has been acquired from people with far more expertise than myself. Anyway, although opening books and tablebases have no doubt increased the strength of a computer's play against strong opposition, most of their strength is still derived from the Shannon A approach. Although programs do "think" and play differently, much of the thought revolves around differences in eliminating moves that are so bad a human wouldn't even consider them, or trying to make sense of what is happening at the leaf nodes (quiescent search). Strong programs still use Alpha-Beta pruning with various enhancements to achieve this. Without brute force, computers would still be no match for a strong human player. Most of the increase in AI chess strength can be attributed to superior processing power for speed, memory for hashtables, and storage for databases and tablebases etc.

CIV is a different proposition, the game tree is so computationally complex that brute force isn't an option, furthermore it won't be an option in our lifetimes. Then there is the problem that cIV isn't a game of perfect information (unless we allow the AI to cheat ;)). Considering that most players would become frustrated if each AI spent copious amounts of time thinking about its moves (before doing something dumb), it seems that the most viable option available to the programmer is to simply script the AI's behaviour, which is what has been done. Perhaps the programmers can prevent the AI from doing some of the obviously dumb things that it does at the moment by adding to the script. However, a script is never going to match an intelligent, adaptive, human thought process, so as long as games like cIV are limited to such methods, I don't foresee any dramatic improvement in the AI. Incidentally, there is an example of a Checkers program called Blondie that uses neural networks to teach itself to play, but it is very weak compared to programs that apply conventional methods. If a computer can't play a computationally simple game like Checkers smart, what chance would it have with a game like cIV?
 
My point here is that it is unfair to compare the AI of chess to a commercial game. The resources spent on developing chess AI are far greater than any PC game can ever spend. As others have pointed out, chess is a game well made for a computer's calculations. Your version of Fritz or the Chessmaster also have the advantage of programming going back in a large way to 1968, hundreds if not thousands of people at some point working on development, the typing in of ECO and similar texts, and millions upon millions of dollars. I suspect if Firaxis had those resources, it could build some REALLY good AI.

So don't expect that kind of skill from any game you buy!
The point, and it has been made before, is that chess is an easy game to program. I have no doubt whatever that I could put together a program within a couple of months which could whup you easily and I know almost nothing about pruning decision trees. If 'AI' folks from the seventies thought it was difficult, that was because the field was in its infancy and computers were orders of magnitude less powerful.

Edit: crosspost with The Rook. I'm pretty sure that you don't need to program opening books in order to make a chess program which plays 1800.

BTW. Checkers has been solved. The game is a draw.
 
I think having different sets of AI codes (maybe one geared toward rushing, one trying to get to a specific tech first, one aimed at building pyramid/GL, etc) would help make AI harder, or at least less predictable. but that's hard to do because it takes more effort to code and to maintain.

the code is also "dumbified" abit in some aspects, because otherwise you wouldn't be able to play on high lvls.

I'm quite sure they can code an axe rush, the warmongers even do it on times, however, if it'd be a truely optimized axe rush(or not even that, plain "build only axes" code), given the bonuses they get on immo/deity, I'm 100% you'd stand no chance no matter what you do. If the ai starts with 2 cities, a good production bonus, worker and builds only axes to rush me, he'll succed in 200% of the cases. Same if the ai, starting with industrious and with 2 cities, directly aims for wonders, 99% you'd never get mids or oracle.
 
If a computer can't play a computationally simple game like Checkers smart, what chance would it have with a game like cIV?

None. But the scripts are poor. As in intentionally nerfed by Firaxis. That's what happened to Blake's initial ideas as far as I know.
 
Back
Top Bottom