CIV 4 - AI Requests

[yoshi]

Since Civ4 will use a whole new (and cleaner) code, you can probably expect XPs for it till the cows come home; They've even stopped patching Civ3 so I wouldn't expect any more XPs...but to stick with the thread topic:

There was some talk about there being no good AIs on the market right now. If you read up on the new AI for Counterstrike:Condition Zero (its primary selling point since otherwise it's basically just CS with SP), you'll find there has been a marked improvement--in fact, the AI functions much like a human player does. Granted this is an FPS, but the AI's decision-making process is probably the eqivilant to Civ3 in complexity (since the AI is having to deal with a 3D environement and effectively react to unpredictable human moves on a second-to-second basis). The main difference where AI orientation is concerned is that in CS:CZ the bots are modded to exploit the maps that ship with the game whereas Civ's AI has to be able to react to random maps.

CS is 6 years old and CS:CZ had one of the worst development records yet it still pulled off execellent AI. Thing is, most of the CS market is MP-based, so the question is not whether good AI is possible but rather if it's worth spending resources on good AI given the negligable pay-off.

As I said before, I think a reasonable chunk of Civ players' systems don't have MP capability, thus it's likely that AI (SP) will be a factor but not a priority for the reason stated above.

One thing I want to see in Civ4 is a leap in diplomatic options; i.e. I want to know that I can at least mod the AI to wheel-and-deal as well as a human player at level 3 in C3C.

I'll also emphasize that giving more AI-rules-modding options will allow the AI to be more controllable without having to rely too heavily on AI characteristics (e.g. AI-AI tech-trading rates).

 

[Jon Shafer]

I personally think the investment on Civ 3 expansions are wasted for the returns. Speaking from where I work anyway, it reached a point where it cost so much to push the code forward since it was so intricate and complicated (without necessarily being well laid out). So we retired one code base, and announced production of the next full version on a completely new code base.

25 programmers for an expansion sounds poor marketing, if you ask me. But I don't have the numbers in front of me, so I'm really speaking randomly. Not like I'm changing anything, nor do I want to. Just hoping they can get to Civ 4 and hit the ground running.

He didn't mean 25 coders, he meant 25 people total (which is still quite a few). Programmers, artists, musicians, testers, producers, designers, marketing people. All told it's a group of 25. That's not even including the number of people working on it OUTSIDE of the development team (like Jeff from Atari, he's part of rather large crew). All said it's quite a commitment of resources.

All interesting. I guess maybe it's a natural cycle - code gets stretched to the breaking point, then you have to start over.

Yes, I wouldn't want to distract Firaxis from working on CIV 4!

Correct, and as has been pointed out, the condition of Civ 3 wasn't even all that great when it was first released.

 

[Jon Shafer]

Since Civ4 will use a whole new (and cleaner) code, you can probably expect XPs for it till the cows come home; They've even stopped patching Civ3 so I wouldn't expect any more XPs...but to stick with the thread topic:

There was some talk about there being no good AIs on the market right now. If you read up on the new AI for Counterstrike:Condition Zero (its primary selling point since otherwise it's basically just CS with SP), you'll find there has been a marked improvement--in fact, the AI functions much like a human player does. Granted this is an FPS, but the AI's decision-making process is probably the eqivilant to Civ3 in complexity (since the AI is having to deal with a 3D environement and effectively react to unpredictable human moves on a second-to-second basis). The main difference where AI orientation is concerned is that in CS:CZ the bots are modded to exploit the maps that ship with the game whereas Civ's AI has to be able to react to random maps.

I hate to break it to you, but this assumption is incorrect.

While the basic functions of a strategy game AI (optimizing food, city placement, etc.) are simpler than a game like CS, the difference comes in the broad goal of the AI - for CS, the AI must simply react to the environment and respond in the best way for a certain set of circumstances. A strategy game's AI cannot function like that. If it wants to be effective, a strategy AI has to be able to "PLAN" things ahead of time - it cannot simply react to the environment. If it did then you would never see an AI civ launching an invasion, colonizing a certain area, building the spaceship in hopes of winning. Making an AI that does these sorts of things is a much more daunting feat than a FPS bot.

CS is 6 years old and CS:CZ had one of the worst development records yet it still pulled off execellent AI. Thing is, most of the CS market is MP-based, so the question is not whether good AI is possible but rather if it's worth spending resources on good AI given the negligable pay-off.

As I said before, I think a reasonable chunk of Civ players' systems don't have MP capability, thus it's likely that AI (SP) will be a factor but not a priority for the reason stated above.

One thing I want to see in Civ4 is a leap in diplomatic options; i.e. I want to know that I can at least mod the AI to wheel-and-deal as well as a human player at level 3 in C3C.

I'll also emphasize that giving more AI-rules-modding options will allow the AI to be more controllable without having to rely too heavily on AI characteristics (e.g. AI-AI tech-trading rates).

Well, as Soren has said that just about everything will be moddable through XML and Python I don't think you have to worry about that... you have an actual programming language at your fingertips to make the AI function exactly as you wish.

 

[yoshi]

While the basic functions of a strategy game AI (optimizing food, city placement, etc.) are simpler than a game like CS, the difference comes in the broad goal of the AI - for CS, the AI must simply react to the environment and respond in the best way for a certain set of circumstances. A strategy game's AI cannot function like that. If it wants to be effective, a strategy AI has to be able to "PLAN" things ahead of time - it cannot simply react to the environment. If it did then you would never see an AI civ launching an invasion, colonizing a certain area, building the spaceship in hopes of winning. Making an AI that does these sorts of things is a much more daunting feat than a FPS bot.

CS:CZ's AI does have to plan ahead as it reacts to direct stimuli; e.g. sound (player's footsteps): AI must plan a route to follow that will cause the bot take the shortest possible route while determining the direction of the player (i.e. where the player will be) so that it will be an intercept course--rather than endlessly following the player).

Civ3's AI doesn't really plan that much: it sees an opportunity and decides whether or not to react.

For example:

AI wants to build a city.

1. Determine appropriate coordinates (food, shields, commerce, resources--although Civ3's AI isn't what you would call picky...and C3C didn't change that ).
- AI does not take defence into account (preset to always accompany Settler with 1 defensive unit);
- AI does not take enemy proximity into account (frequently builds right next to *enemy cities).
- AI does not take future population growth into account (city may even stay at 1);

2. Build units (x # of defenders).
- AI does not take upkeep into account;

3. Build improvements.
- AI does not take maintenance into account;
- AI apparently has to build something so it must decide based on its civ's characterisitcs and economic, military, cultural, etc. status (this is very likely predetermined: if x # of units per x # of cities / x number of SRs, and so on, then build x), as well as on direct requirements such as population growth requiring the building of an Aquaduct, for instance (pretty sure AI doesn't plan this: builds the Aquaduct before pop. reaches growth limit because it's next on the list--i.e. the illusion that it has planned this when it's just a fluke or at most, the game is balanced so that city growth requirements coincide with production and tech--or because the pop. reaches the limit thus triggers the switch to building Aquaduct;

[* 'enemy' refers to other civ, whether at peace or war]

(The not taking upkeep into account being one of the reasons why the AI is always broke--it just keep building regardless of how much cash it has...there may be a limit, I don't know.)

IMO there's really not that much planning going on the AI's part. If it seems like the AI has a 'plan' it's an illusion that results from preset actions that trigger under certain situations like the example I gave above (or "canned strategies" if you will).

Although, yes, Civ3's AI probably does still plan more than CS:CZ's AI; I would guess that most of the real planning, though, is used in unit strategy (similar to the CS example I gave in the previous post). The other thing is tech-trading, but there are only so many wildcards that the AI has to deal with there. In CS, every move the player makes is what you could almost call a wildcard (although CS's AI isn't taking ALL that into account as it would be too overwhelming--think of BB trying to play chess with a human on a board the size of a football field...Blue would fry pretty quick).

 

[dh_epic]

Programmers or not, that's still a lot of people to do an expansion pack where you don't really change the gameplay much. A few new units, a few new wonders, a few new civs, some new art.

Then again, building the editor must have been a HUGE chore. I guess therein lies the time, along with playtesting those scenarios. I think I just answered my own question.

 

[Sirian]

Adding MP on to Civ3 was a huge chore, and that ate up most of PTW. C3C added the conquest scenarios, which had to be designed, balanced, supported with art, tech trees, rules mods, etc.

- Sirian

 

[Neomega]

FPS AI vs TBS AI:

1. Evidence: A choice handful have ever beat Xaero (final boss) on Nightmare (hardest level), in Quake 3. (I looked it up, and there are a few. ) No cheating by the AI in this instant.

A good portion of Civ III players can beat the AI on Monarch (relatively little cheating by the AI).

FPS AI doesn't have to "hear" anything, it already knows the location. FPS AI's are nerfed, so the game is enjoyable. CPU's are faster than people, especially in their own realm of tangents and vectors.

Really, this is a no-brainer. FPS AI is not incredibly complicated. Things like hiding behind a box etc is almost more for the enjoyment of players, than any realistic need for AI improvement. So it "seems" like players are playing people. But it is not necessary.... they can fire faster, can "estimate locations" way better, etc.

A FPS AI can win a "showdown at noon" draw any day. It takes a millisecond to decide you are in line with the target vector, and fire. A good person takes at least a hundredth of a second.

 

[Jon Shafer]

CS:CZ's AI does have to plan ahead as it reacts to direct stimuli; e.g. sound (player's footsteps): AI must plan a route to follow that will cause the bot take the shortest possible route while determining the direction of the player (i.e. where the player will be) so that it will be an intercept course--rather than endlessly following the player).

Yes, but there's only so much to consider in a FPS game - your location, your enemy's location, the possible routes and what might happen should you take those routes. In a strategy game there are 100 times as many things for the AI to consider. Multiple cities/provinces, what to build, why to build it, what to do with it once it's built, what to research, how much to expand, who to attack, why to attack them... the list goes on and on and on and on.

Civ3's AI doesn't really plan that much: it sees an opportunity and decides whether or not to react.

For example:

AI wants to build a city.

1. Determine appropriate coordinates (food, shields, commerce, resources--although Civ3's AI isn't what you would call picky...and C3C didn't change that ).
- AI does not take defence into account (preset to always accompany Settler with 1 defensive unit);
- AI does not take enemy proximity into account (frequently builds right next to *enemy cities).
- AI does not take future population growth into account (city may even stay at 1);

2. Build units (x # of defenders).
- AI does not take upkeep into account;

3. Build improvements.
- AI does not take maintenance into account;
- AI apparently has to build something so it must decide based on its civ's characterisitcs and economic, military, cultural, etc. status (this is very likely predetermined: if x # of units per x # of cities / x number of SRs, and so on, then build x), as well as on direct requirements such as population growth requiring the building of an Aquaduct, for instance (pretty sure AI doesn't plan this: builds the Aquaduct before pop. reaches growth limit because it's next on the list--i.e. the illusion that it has planned this when it's just a fluke or at most, the game is balanced so that city growth requirements coincide with production and tech--or because the pop. reaches the limit thus triggers the switch to building Aquaduct;

[* 'enemy' refers to other civ, whether at peace or war]

(The not taking upkeep into account being one of the reasons why the AI is always broke--it just keep building regardless of how much cash it has...there may be a limit, I don't know.)

IMO there's really not that much planning going on the AI's part. If it seems like the AI has a 'plan' it's an illusion that results from preset actions that trigger under certain situations like the example I gave above (or "canned strategies" if you will).

Although, yes, Civ3's AI probably does still plan more than CS:CZ's AI; I would guess that most of the real planning, though, is used in unit strategy (similar to the CS example I gave in the previous post). The other thing is tech-trading, but there are only so many wildcards that the AI has to deal with there. In CS, every move the player makes is what you could almost call a wildcard (although CS's AI isn't taking ALL that into account as it would be too overwhelming--think of BB trying to play chess with a human on a board the size of a football field...Blue would fry pretty quick).

To be brutally honest, the Civ 3 AI does a lot more "reacting" than it does "planning," which is why it's not really that difficult to beat it on even footing. A true strategy AI would have to do much better. Many players can defeat the AI quite easily on higher levels because of this.

With most of my comments, I'm referring to the "ideal" strategy AI, as opposed to the ones we've already run across. In order for a strategy AI to be able to compete with a human as well as a FPS bot AI can, it has many more things to consider.

As Neomega mentioned, a "perfect" FPS bot is achievable - those that aren't are simply dumbed down so that the human has a chance to compete with them. The AI can know exactly where to shoot to get a perfect headshot and doesn't have to guess like a player. While the player's actions aren't pre-determined, there are only so many moves a player can make in the split second when an AI player has to decide what to do. A player's head being a millimeter further away in a "random" direction doesn't make much difference to a bullet.

 

[yoshi]

FPS AI doesn't have to "hear" anything, it already knows the location. FPS AI's are nerfed, so the game is enjoyable. CPU's are faster than people, especially in their own realm of tangents and vectors.

I used the example of CS:CZ because they chose not to nerf the AI so it doesn't 'know' where you are. It bases its decisions on sounds that it hears. Translation: action (sound) at x level from x direction at x distance triggers AI to take your coordinates into account in its decision where to move. It's pathfinding takes your direction into account so that it will intercept rather than...look, it was an example. My point was that there are a lot of constantly changing variables the AI has to deal with every second and that Civ3's AI isn't really reacting and plannnig around certain variables but rather is just following preset patterns (i.e. an illusion).

BTW, your example of the "showdown" isn't saying anything because 'good' AI isn't supposed have perfect accuracy. Instead there should usually be a timed delay and then some randomization where accuracy is concerned (i.e. shoot like the player does). Bots just shooting is only a matter of x # of milliseconds to reach x % of accuracy (CS AI obviously waits until 90-100%) to hit target at bearing x (you).

CS bots frequently get headshots the moment you turn a corner because, as you said, they 'know' your location--and which direction you're facing/moving--thus are already 'aiming' at your head well before you turn the corner, and since there is no 'bullet' delay taken into effect, the result is that the bot fires the moment your head is past the wall. Likewise, CS Bots sometimes miss intentionally but overdo it (bot shoots in every direction but where you are).
CS:CZ is different because the bots are programmed to react only to stimuli that is in visual range or that is audible (e.g. footsteps). Only then do they 'know' where you are. They have preset search patterns (e.g. checking doorways as they pas them) to make up for the lack of omniscience. The intentional missing is fixed so that they shoot more like humans.

The strategy-game equivilant (in terms of startegic planning for instance) of that kind of AI improvement in Civ4 would make my f-ing day.

 

[Jon Shafer]

But the underlying point is that even with all of that to be taken into account, a strategy game AI still has much more to do. Soren didn't program the Civ 3 AI without such features because he isn't capable or because he didn't want to - he programmed it as a series of environmental triggers linked together because there's so much to do that's the only feasible expectation. It would take far longer and much more work to do something otherwise. Will that be in CIV? Is it possible? We're revisiting a debate already brought up in this thread. Maybe, maybe not. It depends on what Soren wants to do and what he thinks he's capable of doing in the time allowed. All we can do is hope that he believes it's possible and does it.

 

[dh_epic]

I still think there's are some actual philosophical questions of what constitutes a good Civ AI -- instead of just saying "make it play more like a player".

1: Assume it makes the most sense for the AI to play at the level of the player.

To play at the level of a player, it is insufficient for the AI to just work at a steady pace and hope its hard work pays off. The most intelligent players will throw away their alliances, they'll deprive opponents of resources technologies and money even when they are at peace, and will TRUST NO ONE. Thus, for the AI to play with the quality of a player, it must emulate them.

Imagine a world where nobody trusts one another. With a deadline at 2050 and the scores close, imagine Britain turned around RIGHT NOW and Tony Blair said "suck it, Bush!" To make matters more interesting, the rest of the players figure "not my damn problem" and idly watch as Britain and the US go at it. Russia, not fond of the US, decides war is a good idea -- except they'd need more land to pull it off well, so they invade Germany and France for the hell of it. The whole world breaks down, but not in an interesting WW2 kind of way, but more in a dark ages "who gives a crap who's on who's side" kind of way.

From a gameplay standpoint, this is quite boring to me, and I'm curious what kind of person would enjoy it.

2: Therefore, the AI should not play at the level of the player.

But then if you do this, you end up with what you had in Civ 3. The AI joyfully watches as you run away with the game and says to themselves "yay, I'm Greece! Duh duh duhhhhhh, duh duh duhhhhhh... oooh a new technology, duh duh duhhhhhhhh." By the end of the game, the only way you're not ahead by a margin of at least 10% is if the AI is scaled up to Deity, and the "duh duh duhhhhhh"s are abbreviated to a short "d!"

The game is either non competitive, or you're playing against a one dimensional boring player. Or you end up with the Civ 2 "everybody attack the player!!" solution -- you still can't lose, but at least they can stop you from winning. Boring.

3: Therefore the AI should play at the level of the player.
4: But 2 and 3 contradict!
5: Therefore, peanuts are tasty.

I'd like to think I have a solution to this contradiction, in my mind, but I'll leave it there for now. I'm curious how many people agree with me up to this point.

 

[Jon Shafer]

You're confusing the two elements of any player (human or AI) - goal and skill.

Goal is when the human player declares war on everyone a few turns before 2050 before winning. Skill is when the AI sends a flood of Settlers through your territory every turn to build a city on the 2 tile tundra peninsula.

To make a good AI you have to give it different GOALS from the player, but the same SKILL level. So you have an AI that tries to maintain stability in the world and replicate the history of civilization, instead of degenerating into a morbid domination free-for-all. But in doing so, it performs at the LEVEL of the player, conducting wars, diplomacy, micromanagement, etc. in a superb way, giving it the capability of COMPETING with humans without replicating their goals in the game.

 

[MMAfan]

A FPS AI can win a "showdown at noon" draw any day. It takes a millisecond to decide you are in line with the target vector, and fire. A good person takes at least a hundredth of a second.

Hey, they use a computer guidance system in practically everything in the military. Calculate parabolic trajectory for artillery, guide the cruise missile to the target and even shoot down a incoming rocket with bullets with a gun turret on a ship, damm speak about fast targetting. In a FPS with one shot one kill an AI would only need 1 bullet to win with a 99.9999% certainty.

 

[dh_epic]

To make a good AI you have to give it different GOALS from the player, but the same SKILL level. So you have an AI that tries to maintain stability in the world and replicate the history of civilization, instead of degenerating into a morbid domination free-for-all. But in doing so, it performs at the LEVEL of the player, conducting wars, diplomacy, micromanagement, etc. in a superb way, giving it the capability of COMPETING with humans without replicating their goals in the game.

I don't think that's possible. You see, setting the goals IS the skill, watching the score, declaring war, attacking the weak, all quite skilled. The player has an advantage in this respect -- they don't pay attention to reputation or honor. The human player simply WILL engage in more conquest than the AI, and in serial to boot. The AI will give up a conquest for a peace treaty, whereas a player will play ruthlessly.

You're right about one thing, though, the AI makes up for it in skill. But since there are very few alternate strategies or paths to victory for the AI, it boosts its skill artificially, with accelerated production and research.

Show me a super-skilled player who has even one shred of mercy, and I'll show you a loser.

 

[dexters]

DH_Epic made some good points 2 on the posts above.

I want to mention that we are getting awfully close to the 'killer' big blue AI concept again. I'm not interested in that.

I think Sirian said it best several pages ago when he said he wanted a more human AI (not neccessarily a more difficult one, although we presume being more human could make it more difficult). This implies some sort of randomness to decision making, and to create moment where people go '

"wow, the diplomacy session felt really human. The AI pulled was ready for war, she called by bluff"

The Civ3 AI tries to emulate the human element by using RNG variable on many of its decision criteria. That's how you get the sneak attack without warning scenario and that's how asking an AI to move may or may not result in war, ceteris paribus. But the RNG is only one variable that is truly random and unpredictable and it becomes by its nature predictable and simple a question of gambling and a player's risk aversion. The humanity is lost.

What I would like to see, ideally, is a more competent AI on all levels, but really an improved leader AI. The Civ3 AI, as Soren described it in an interview back in 2001 is multi-layered. There is a Leader AI > City AI > Unit AI and each element communicate with each other. By improving individual segments of the AI, we will probably get a better overall performance if Civ4 goes in the same patch as Civ3.

On the macro leader level, the point has already been made that the AI needs to be able to play more strategically. I'm not sure how feasible this is, but the current way is plays is basically on a turn to turn basis. It is reactive rather than proactive, although proactive elements have been programmed into it (like an AI frantically building airpowe or seapower for no apparent reason than because the programmer made it want to do so -- with the implied expectation of war later on) This however is hardcoded human intervention and not truly an AI behaving on its own. What would be interesting is for the leader AI to governed by more variables, say an Expectation set of variables and a goal set variable.

The expectation set is simply that. Using technical (historical game data and perhaps other data and calculations) the AI projects ahead several turns and see what its expectations of future variables will be.

Civ A, B , C is currently at X, Y, Z points. what is the AI expectations in 20 turns? What will the expected relative ranking be in 20 turns? Our economy is currently generating X gold per turn. What is our expectation in 20 turns (given current growth, adjusted for a number of variables). With these expectations, the AI can become more human in its decision making. Rather than reacting randomly, and sometimes in contradictory behavior (as is sometimes the case with Civ3 AI) we will see a more focused AI.

If AI for Civ D predicts Civ A (can either by human or AI player) will become dominant in 20 turns, it may set about a plan to deal with it.

This of course says nothing about how well the plan can be executed, as this depends on still other aspects of the AI (like how well it wages battles, how well its diplomacy algorithm is, and the way it employs a strategy). So ultimately, Soren can program an AI that performsn competently on all these levels and we end up with a more human AI (acting on expectations) but one that may not neccessarily be more difficult. And I will happily take it.

 

[Volstag]

I think CS is a poor example of "good" strategy AI. I'll freely admit, however, it's good FPS/bot AI. As mentioned by others, the amount of information the enemy bot has to deal with is exceedingly less than what's required in Civ.

This is somewhat hard to articulate, but I'll give it a shot: in an FPS the bot AI has a clearly defined goal: deliver more "damage" to the human player, than the human player can deliver to the bot. In Civ, the (ideal) goal of the AI is much more nebulous: outplay the human. The former lends itself to clear/logical heuristics, while the latter doesn't.

Or, to put it another way: they could easily write bot AI in an FPS that would either win, or draw, against a human opponent every single time. Lightning quick "reflexes" coupled with infinitely precise "aim" is a tough, if not impossible, nut to crack. In Civ, there are so many different ways to win, and so many different ways to approach each victory condition, that it would be a momumental task to account/react to all of it.

If the only victory condition in Civ was Conquest, it might be a radically different story.

-V

 

[Jon Shafer]

I don't think that's possible. You see, setting the goals IS the skill, watching the score, declaring war, attacking the weak, all quite skilled. The player has an advantage in this respect -- they don't pay attention to reputation or honor. The human player simply WILL engage in more conquest than the AI, and in serial to boot. The AI will give up a conquest for a peace treaty, whereas a player will play ruthlessly.

You're right about one thing, though, the AI makes up for it in skill. But since there are very few alternate strategies or paths to victory for the AI, it boosts its skill artificially, with accelerated production and research.

Show me a super-skilled player who has even one shred of mercy, and I'll show you a loser.

This is only true for a certain ruleset. The ruleset can always be changed so that if a human player does nothing but play ruthlessly, conquering all in range of him, that he should be penalized.

We're not working within a static set of boundaries here - it's not like real life and there's nothing to prevent or suggest certain activities unless YOU have the ability to enforce your will.

It could be something as simple as Civ 2's solution - as the human player gets stronger and ticks off more AI civs, he ends up being ganged up on until the entire world has united to put him in his place. In response to the human trying to break the rules of the system, the system adapts and tries to force the human player to conform. Unless the human is exceptionally skilled and powerful, the combined might of the rest of the world will bring him to his knees. And he will learn to play nice.

That is only one example. There are other ways to approach this problem, including the more abstract issues like "a human player will always fight a war to the death whereas an AI player will make peace with only 2 cities left." There are solutions to these sorts of problems, but since I'm not coding the game, I don't feel the need to have to come up with any for everything.

 

[dexters]

IMO, I think the ganging up effect on Civ2 was a preprogrammed effect to keep the challenge there in the late game. It is essentially AI vs human stategy that many find distasteful and equate to cheating.

Like many others, I quite like the every civ for himself attitude in Civ3. Although there is one AI running all the Civs outside human control, each of the Civs behave as if they are independent of one another. It's essentially an AI with split personalities. I've already discussed how to improve the AI (in terms of its leader abilities) to plan more strategically. Ganging up may be an outcome of this decision making. But I would strongly suggest against doing such a thing in Civ4 as a predetermined outcome to be expected by a powerful Civ.

 

[dh_epic]

Hey trip, this is the problem I encountered in the second assumption, to quote myself:

"2: Therefore, the AI should not play at the level of the player."

"The game is either non competitive, or you're playing against a one dimensional boring player. Or you end up with the Civ 2 "everybody attack the player!!" solution -- you still can't lose, but at least they can stop you from winning. Boring."

The side effect of this, too, is that multiplayer Civ and single player Civ are two very different games. Not different like playing against different styles, but different as in the goal and constraints are different. In one game, any player can win. In the other game, you can't lose, but they can stop you from winning.

It's not real competition. Imagine playing a basketball game where you always end up 30 points ahead, so your opponents just focus on stopping you from making 100 points. I guess we can agree to disagree, if you think that's a better challenge.

 

[Jon Shafer]

IMO, I think the ganging up effect on Civ2 was a preprogrammed effect to keep the challenge there in the late game. It is essentially AI vs human stategy that many find distasteful and equate to cheating.

Like many others, I quite like the every civ for himself attitude in Civ3. Although there is one AI running all the Civs outside human control, each of the Civs behave as if they are independent of one another. It's essentially an AI with split personalities. I've already discussed how to improve the AI (in terms of its leader abilities) to plan more strategically. Ganging up may be an outcome of this decision making. But I would strongly suggest against doing such a thing in Civ4 as a predetermined outcome to be expected by a powerful Civ.

And how is this behavior different from human behavior? Why is it cheating?

If you're playing with all humans and you have the strongest army on the planet and manage to conquer half of it, do you think that the rest of the players will completely ignore you?

HELL NO

They'll scheme and plot and ally and take you down. Why should the AI behave any differently? Do players think that if the AI actually tries to BEAT them it's unfair? That if the AI civs decide that you're too strong and think that the only way they can survive is to team up against you that it's CHEATING?