We know the AI is stupid, but why?

[Terrapin]

I think the list of AI stupidities is the longest thread on this board. I cannot help but wonder if it really needs to be that way. For instance, the AI famously dribbles units into your territory to be slaughtered. That is bad enough, but why does it insist on moving low-defense units to within range of an attacker? Why launch attacks the attacker has virtually no chance of winning? Can't the civ2 AI check the potential outcomes of a move the way a chess program does?
I can understand why the AI is weak on strategy in a game where the victory conditions are diffuse and the methods for achieving them indirect. I cannot understand why the AI has to be tactically incompetent.

Also, it seems like the AI should be able to make up for some of its deficiencies in research and economics by being the micro-manager to end all micro managers. For instance, is it so much to ask the AI to shuffle around the citizens of a city to produce extra trade when the current layout is going to cause shields to go to waste when production is within one turn of veing complete? Would it be so hard for the AI to fenagle its tax rate/squares worked to boost shield or food production when a new tech is about to be discovered by limiting science to the exact # of beakers needed?

 

[Mercator]

My guess is the designers were no experts in AI, nor very good strategists either. The only reason why they might be able to beat the game at Deity is because they designed the damned thing.

I don't think the AI really thinks ahead more than the current turn, which explains why units simply rush towards the enemy.

And I don't think the designers really made any effort in optimizing such micro-managing. After all, even most official strategy guides complete miss the point on those things.

 

[Old n Slow]

Actually, I think the AI is pretty good. Consider the state of the art, eight years ago -- consider the multiple things to pay attention to -- not only the military, but population growth and satisfaction, the economy, diplomacy, tech races and wonder building.

Consider also the fact that many players are still challenged by King level, and it took a community or two to really master the game at Deity. In fact, new discoveries about the game are still (eight years+ after release) being posted and incorporated into best play.

If the military moves, tactics and strategies of the game were pushed up to the modern military doctrines, would that mirror the ancient or even medieval history? Read some of the battles conducted and the analysis about them -- in light of the losers, the ai ranks OK (not great, but it could be worse.)

Final thought -- if the ai were too good, we would have too few players to enjoy the game, and too few to justify a website.

 

[la fayette]

Originally posted by Mercator
My guess is the designers were no experts in AI, nor very good strategists either.

Think they were that bad and yet managed to build the best computer strategy game ever built up to now: astonishing

 

[Archer 007]

I think that other traits in the game really help to balance out the weak AI.

 

[Terrapin]

consider the multiple things to pay attention to -- not only the military, but population growth and satisfaction, the economy, diplomacy, tech races and wonder building.


As far as overall strategy is concerned, I agree. However, this does not explain why a catapult walks up by itself and stands right next to a horseman, for instance. Not committing obvious blunders like these would make the game more challenging, but not unplayable.

 

[ElephantU]

There are a few open-source civ clone projects out there where you can contribute to writing a smarter AI. Try FreeCiv (www.freeciv.org) or C-Evo (c-evo.org)...

Considering the "limitations" on program size and complexity back when Civ started, the AI is probably pretty typical of its time. Chess is a much easier to quantify and predict kind of game, so "look-ahead/plan-ahead" algorithms work better.

 

[Terrapin]

Before contributing to a more intelligent open source ai, I am first going to have to invent a smarter Terrapin!

"Chess is a much easier to quantify and predict kind of game, so "look-ahead/plan-ahead" algorithms work better."

True, but the more open ended nature of civ2 does not, it seems, mean that the AI could not look ahead at the fate of one particular unit before moving it.

 

[Mercator]

Originally posted by la fayette
Think they were that bad and yet managed to build the best computer strategy game ever built up to now: astonishing

AI, creating a solid game, and being able to play that game are entirely different things.

 

[yaroslav]

I've been told that Brian Reynolds some years ago declared that no body wants a strong AI

Anyway, creating a difficult-to-win AI is very, very time-consuming

 

[Ace]

Originally posted by Mercator


AI, creating a solid game, and being able to play that game are entirely different things.

Aye!! There is the rub. It appears that Sid and company grossly underestimated the ability of the human player to learn! IMHO, he had no idea of the talent and determination of the people who became addicted to civ. A rereading of the manual and official guides leaves one with the distinct impression that Sid and company truly believed that diety level was almost unbeatable and did not have a clue that people would figure out how to play at Deity +2 and Barbarian Wrath and win!!

That may also be why the AI in MPG is more hostile the in 2.42.

 

[ElephantU]

The AI in MPG is so hostile there really is no such thing as multi-turn diplomacy. If you want to build a friendly relationship with an AI civ use Classic/2.4.2. Respectful AI "attitudes" have been shown to vanish in one turn playing MPG. Might as well play "Bloodlust"...

 

[cdscivnut]

Originally posted by ElephantU
The AI in MPG is so hostile there really is no such thing as multi-turn diplomacy. If you want to build a friendly relationship with an AI civ use Classic/2.4.2. Respectful AI "attitudes" have been shown to vanish in one turn playing MPG.

I can't stand this. It is almost impossible to trade maps with the AI other than the first time you meet them. The bright side is the AI is so dumb you can just tear them apart it they are unfriendly.

 

[Old n Slow]

Originally posted by Terrapin

However, this does not explain why a catapult walks up by itself and stands right next to a horseman, for instance.

Some of those"blunders" may be due to the sight -- how many times have you moved the second move of a two move unit only to 'discover' an enemy? For me it happens too often, so much so that I tend not to fight so often at the early years, but usually wait unto Tactics or later.

 

[Ace]

It happens much too often. I cannot count the number of times a none two mover from an early hut has ended its turn right next to a barb archer!!

Re: trading maps, the AI in MPG will trade if its attitude is worshipful. I suspect that sometimes the AI does not want to trade maps because you do not have enough explored area to meets its criteria. Has anyone done any research on that possiblity??

 

[Duke of Marlbrough]

My usual problem is just that I forget to make sure they have Map Making first.

 

[la fayette]

Originally posted by Ace
Re: trading maps, the AI in MPG will trade if its attitude is worshipful. I suspect that sometimes the AI does not want to trade maps because you do not have enough explored area to meets its criteria. Has anyone done any research on that possiblity??

Map trading requires that both partners have acquired Map Making previously D DoM). Then the AI will always say OK if it is worshipful. It will also say OK quite often if it is enthusiastic, and even accept trading now and again if it is 'cordial' (sorry, I suddenly miss the English word).
The rule is the same in MGE and 2.42, but there is a big difference: in 2.42 you are sure that the AI's attitude will improve regurlarly step by step when you gift techs, while this is not true at all in MGE. You must therefore be prepared to massive tech gifting before Map trading in MGE (it has happened to me to give 7 or 8 techs in a row and still get a 'No' reply ).
I have never heard of anyone having researched whether there is any rational reason why the AI says Yes or No when it is less than worshipful (I personally took it for granted that it was random, but I may be wrong ).

 

[raiderbob]

Originally posted by Terrapin
I think the list of AI stupidities is the longest thread on this board. I cannot help but wonder if it really needs to be that way.

Agreed. Does anyone know if this has improved in civ III?

1Mercator = I concur
2Old n Slow= the AI isn’t very good, even considering age.* There are tactical and strategic commanders from the past that, given comparable technical assets, would expose the incompetence of today’s “modern military doctrines”
3la Fayette = astonishing post! ;-)*
4archer_007= I concur, but then so does Final Fantasy.
5Terrapin= Agreed, I might get around to asking those in civ III.
6ElephantU= my basic programming is a weakness, but then so is that damn “median line”! As far as chess… well it took a “super” computer to beat a grandmaster, above average chess players would tear this game up in less time. And ”Considering the "limitations" on program size and complexity”*
7Terrapin = yes, no and yes.*
8Mercator = concur.
9yaroslav = I guess. And probably
10Ace = Yes, but in 2.42, does more hostile mean better tactics?
11ElephantU = Yes, but in 2.42, does more hostile mean better tactics?
12Cdscivnut = apparently not (10&11)
13Old n Slow = ok, some maybe. But when the AI city can see and still sends out cannon, catapults etc?
14Ace = sadly yes. And see #16
15Duke of Marlbrough = I use it as one of my first “strikes” upon contact.
16la fayette = w/ my limited experience, concur. None in MGE/2.42 (I think)
17raiderbob = >*< Empire Deluxe which predates any civ game by at least a decade (two?) had expert computer opponents (AI) that inflicted losses upon humans. (wasn’t even in the top 100 of some bogus site I linked to from here somewhere) it seemed like a waste of time to make 17 different posts, but if anyone cares to debate the original question, or any point i've made... i'll reply

 

[The Person]

I'm not sure if you all agree to this, but sometimes the AI blunders are hilarious enough to bring up a good laugh, though mostly they're annoying.