Making the AI a fellow competitor

Mewtarthio

Emperor
Joined
Apr 14, 2004
Messages
1,930
One thing has always annoyed me about cIV: The AI is nothing more than a computerized enemy meant to be crushed by the player. I'm not saying that I actually believe a sapient, human-level AI is possible with current technology, but I am annoyed at the obvious gap between human and AI. It's not that the human is inherently smarter than an AI, but rather that the human plays a completely different game. The AI runs around, tallying up things like State Religions and recent tributes and the like, while the human is capable of ignoring such pointless matters. The human cares about one thing: winning. The AI cares about a dozen different things, few of which get it closer to victory, and all of which can be used to manipulate it relatively easily.

I believe that, in principle, the AI should be another competitor, not just an obstacle for the player to overcome. In this respect, religion is poorly implemented: No human would care if another player had the same state religion under the current system. It would be better if you had benefits for sharing SR with friendly civs and penalties for sharing SR with unfriendly civs (or being friendly with heathen civs), such as (as a random, totally balance-free suggestion) increased happiness per trade route with a foreign brother-in-faith, increased unhappiness if your people trade with heathens (not necessarily to say that they'll be particularly miffed by the contact, but that unhappiness represents a certain extent of political instabilty, caused in part by exposure to foreign ideas). In this way, both human and AI act out historical roles, spreading their religions throughout the world and attempting to share religions with their neighbors whether by spreading their faith with the sword (We offer: Peace Treaty; We want: Convert to Hinduism), convincing their neighbors to follow them (We offer: Drama; We want: Convert to Hinduism), or just going with the majority (Everyone else is Hindu; we'd better follow suit). The difference is that these are no longer done to acheive aribitrary modifiers but rather to pacify the citizens.

My belief is that changes such as this must be effected if the AI is to be competitive. Pull up any high-level OCC victory strategy to see what I'm talking about: It will almost certainly revolve around finding an AI and manipulating it (using a system so transparent that the numbers actually appear on-screen) into signing a Permanent Alliance. What human player would ally with someone who has but a single city? What sort of intelligence at all, real or artificial, would throw in its lot with such a weakling? It's not as if the unlimited National Wonders give any benefit to the city's ally.

I often see topics (there's a pretty big one on this forum right now) asking about how to improve the AI so it can compete with a human. I claim that, with Civ's current abuse of the AI, that is not possible. The AI is transparent and foolish, and it's aribitrarily handicapped by taking into account items with no true impact on gameplay. I really believe that the game rules should be retooled, possibly not even until Civ V, such that history can be replayed and realistic relationships formed with an AI just as ruthless, if not as cunning, as the player. Until then, the AI will be nothing more than a fancy video game boss, no more a true opponent to the player than Bowser is to Mario.
 
I'm not sure if you're dancing around it or if you have it in mind... but you touched on the main problem.

People request a competitive AI. That means opportunistic wars, backstabbing, picking on the weak, and trying to delay fights with the strong until it's on your own terms.

But people also request a realistic AI. That means wars based on politics and religion, loyalty to friends, and getting angry at backstabbers.

Throughout the history of the entire Civilization franchise is you quickly realize that realistic diplomacy is NOT profitable. Yet without realistic diplomacy, the game becomes almost like an RTS, but slower. Why build a wonder when you can take it by conquest? Why found a religion? Why keep your friend? Why pursue peaceful spaceship victory when you can beat the snot out of someone, take their land, and have a huge economy for an even faster spacesihp victory?

The problem isn't the AI, it's the ruleset. The Civilization franchise is built on a contradiction: that you can conquer the world by yourself and indeed that's the best way to win, but the AI is supposed to pretend it's not a game and that you guys can be good buddies.

Firaxis needs to resolve that contradiction. Because a truly competitive AI in Civilization 4 would take us into an "always war" game, and thus ruin half of Civ 4's features.
 
The writer of this thread looks like a C-evo fan in the making, imo. lol. Seriously though, I'd say the best solution to that problem is to stick to multiplayer games. (Although few have the endurance for a multiplayer Marathon game. lol.) Until AI's are developed that remember and rationalize, you won't have an AI that can be cunning without being visibly cutthroat. A cunning AI eludes game designers, although I've seen my share of cutthroat AI's in my day. (Hence the C-evo reference)
 
Multiplayer is great. But it instantly cuts the BS out -- which is, unfortunately, a lot of the game.

  • Religion becomes something you grab individually, if at all.
  • Diplomacy is VERY every man for himself... and, in fact, diplomacy is usually turned off because there's the risk that two players gang up on another player. What seemed fair in SP all of the sudden becomes an exploit in MP: everyone dogpiles on the leader.
  • How many MP games involve launching the space ship?
  • How many MPers would vote someone in for a diplomatic victory? "You're a swell guy. Here, I'll let you win."
  • Is there ever an MP game where two people start out as peaceful friends, only to declare war later? Is there ever an MP game where one player DOESN'T try to at least choke another player from the start?
  • What do you think the ideal AI would look like... Gandhi, or Genghis Khan?

You quickly realize that 90% of the game gets cut out by making the AI 100% ruthless. You're left with an RTS in slo-mo.

A lot of Civ players want to have it both ways: an AI that keeps realistic relationships, but also a competitive AI. That can't be resolved with the AI alone -- the AI is chasing two different things then. The solution is to strike right at the root contradiction: the game has to start rewarding realistic behavior, rather than having the AI chase the two rabbits running in opposite directions (realism and competitiveness) and lose both.
 
The AI wouldn't have to be so cut throat and ruthless to be competitive if winning by peaceful means was a real option for the player. A realistic AI would make many alliances and would cooperate with the other civs. A competitive AI would singularly focus on destroying everyone else. With the way the Civ franchise greatly favors conquest the AI can't really be programmed to act both ways.

Side Note: Also it seems most of Civ4's problems are related in some way. It's hard make a really good AI with the lack of peaceful and cooperative options.
 
Multiplayer is great. But it instantly cuts the BS out -- which is, unfortunately, a lot of the game.

  • Religion becomes something you grab individually, if at all.
  • Diplomacy is VERY every man for himself... and, in fact, diplomacy is usually turned off because there's the risk that two players gang up on another player. What seemed fair in SP all of the sudden becomes an exploit in MP: everyone dogpiles on the leader.
  • How many MP games involve launching the space ship?
  • How many MPers would vote someone in for a diplomatic victory? "You're a swell guy. Here, I'll let you win."
  • Is there ever an MP game where two people start out as peaceful friends, only to declare war later? Is there ever an MP game where one player DOESN'T try to at least choke another player from the start?
  • What do you think the ideal AI would look like... Gandhi, or Genghis Khan?

You quickly realize that 90% of the game gets cut out by making the AI 100% ruthless. You're left with an RTS in slo-mo.

A lot of Civ players want to have it both ways: an AI that keeps realistic relationships, but also a competitive AI. That can't be resolved with the AI alone -- the AI is chasing two different things then. The solution is to strike right at the root contradiction: the game has to start rewarding realistic behavior, rather than having the AI chase the two rabbits running in opposite directions (realism and competitiveness) and lose both.

Thank you, dh_epic. You've said that a lot better than I could. I was sort of getting at that with the religion example (having the same religion as everyone else should give you more than an arbitrarily benevolent AI).
 
A lot of single player folks can't enjoy multiplayer, while a lot of multiplayer folks will tear through the single player AI. It's all because of that fundamental contradiction: the AI is designed to keep some kinds of friendships. And group A might want competition, but group B gets mad when the AI does an "unrealistic" backstab -- and indeed, there's some people who are in both groups.

When I think about Civ, I often ask myself what the game would look like if it rewarded realistic behavior -- especially realistic diplomacy (instead of gamey diplomacy). I think it would change the game all the way to the victory conditions.
 
In my opinion, rewarding realistic diplomacy would be a desirable thing for Civ V (I don't think we will have another expansion to IV). This would make it more difficult to, for instance, make a civ as friendly as possible just to get them to declare war on your enemies, so that while their forces are away you can backstab them. However, some people would not like these penalties. Some people will probably think it makes the game "unfun" (those people who routinely backstab).

Here's what I am thinking:

1) For civs that you have +6 or more relations with: If you declare war on them, you will get a -1 "you treated your friend badly" modifier with everyone which stays for the remainder of the game (on top of possible "you declared war on our friend" modifiers. You will also get a "refuses to talk" modifier with non-friendly civs for 5-10 turns. This is not terrible, but you weren't that friendly with the civ anyway.

2) For civs that you have +10 or more relations with: If you declare war on them, you will get a permanent -2 modifier with all civs and a refuses to talk from non-friendly civs for the duration of the war + 5 turns. This penalty would be fairly significant, perhaps enough to deter backstabbers.

I think these measures would be enough to force people to make a strategic choice about whether or not to backstab, perhaps making one aspect of diplomacy more realistic. People would still be able to declare on their closest friends, but it would be a hard descision, not an easy one.
 
Allowing some form of Fedaration style alliance system would help because as it stands you can at best get one perma buddy to share a win with so if the AI is programmed to win or die trying with a mass Alliance system it would not be impossable for the AIs to consider becomeing freinds if thay can share a win, the question really becomes why would player A choose to be player Bs freind is he powerful? will he be able to carry player A to a victory or will he just drag him down. There needs to be some factor for chooseing who will and will not be freinds, there also need to be some why of reeling the AI players in from time to time so thay dont go berzerk and try to zerg the map or all just get lovey dovey all join one happy alliance and end the game as soon as all players have met, this is where you would bring in the factor of populace thay should have a large effect on how the AI plays as by default thay should be mostly peaceful and scared or aggresive and hateful most important would be getting the right mix so that when an AI or human goes on a bender thay get wiplash or if thay turtle to long thay get bum rushed.
I think iv rattled on long enough.
 
I don't think penalties are enough. Especially diplomatic penalties. You've just pulled the backstab that let you nearly double your power: you took down a competitor, and expanded your control. Do you really care that the rest of the world thinks lowly of you now? They tried to induce a heavy diplomatic penalty for the infamous ROP-rape in Civilization 3, but at the end of the day it didn't discourage effectively the most powerful exploit in the entire franchise.

You can jack up those penalties. But then you're left with a Civilization 2 type of game: with every ally that you conquer, you are at risk of provoking more enemies. But the game is still balanced in such a way that you can beat the entire world by yourself. This is too gamey to be realistic.

A solution has to go right down to the victory conditions, IMO.
 
If I remember correctly, the penalty in civ III was that no one would trade with you ever again. I also seem to remember that this did greatly discourage people. People would say things like "make sure to cancel all deals and get all our units out before declaring, or else we're screwed for the rest of the game." This could be wrong, but I remember people thinking that penalty was too much.

On the other point: say you double my diplomatic penalties. In that case no one who backstabbed a friend would have a pretty tough time getting people to trade with them in the future. Maybe if enough people were already at bad relations with you they might form a military alliance against you. I think that that would be a sufficient penalty. You seem to think that whatever the penalty is, it should completely stop people from backstabbing. I think that everyone should have a choice. This is supposed to be realistic, not just "we're at +10 with them, we can't attack."

I think it would be better to have diplomatic penalties than no penalty at all. I'm not sure what you're suggesting as a penalty/reward. It sounds like you think there is no solution.
 
I'm not saying we should stop people from backstabbing. I'm saying that there's no such thing as a penalty that works. Let's say your best friend is about to win the game by space race. You have no choice but to backstab them, no matter what the penalty is. If you don't backstab them, you lose. Period.

I think the solution needs to go deeper, right into the fundamentals of the game. You keep thinking penalties, but I'm thinking rewards. In both cases, it goes right down to the victory conditions.

Galactic Civilization does it with the alliance victory. While it's not perfect, I think it offers an interesting starting point.

I've also heard that MOO (the original) did it. But that's because there's no "race" victory similar to the future tech in GalCiv or the spaceship or culture in Civilization 4. There are only two victories in MOO: diplomatic, and conquest. You play until someone truly comes out on top.

The space race can't be in. Or if it's in, then it can't trigger victory. That's the first step, to me.
 
dh epic said:
I'm not saying we should stop people from backstabbing. I'm saying that there's no such thing as a penalty that works. Let's say your best friend is about to win the game by space race. You have no choice but to backstab them, no matter what the penalty is. If you don't backstab them, you lose. Period.

So...you backstab them. I don't understand. That is a situation where backstabbing is necessary. Are you saying that that kind of situation should not occur?
 
I'm saying all backstabs are necessary. It's a game where there's only one winner. If you're not backstabbing, you're not winning, or you need to jack up the difficulty level.

No reasonable penalty will suffice, because it's almost always worth it to backstab -- the spoils of war are enough to make upsetting everyone irrelevant. You could add an unreasonable penalty for a back stab that's really huge, but then you'd just be obstructing the player and pissing them off.

You have to restructure the victory conditions. That's the answer. Tacking more penalties onto Civ 4 or Civ 3 will do more harm than good.
 
True, but the very idea of "winning" history is ludicrous in and of itself. This isn't a Maxis game, so somebody has to win somehow. I'm willing to accept a bit of unrealism in the victory conditions provided realism is incorporated elsewhere. Besides, what sort of victory condition would promote realism? No matter what, you can always rest assured that people will want to kill those who are near victory.

I think the biggest problem is that, in civ, bigger is always better. Sure, you can backstab a friend, because taking new cities is almost always worth it. As I've said before, cIV took a step in the right direction by adding maintanence that can make those extra cities drain your treasury, but you can still win by sheer size if you expand enough (the only difference is that you have to go a little more slowly now).
 
I agree that Civ 4 has made positive steps in "bigger is always better". Sometimes I think they should go even further in nerfing sheer size. But you have to wonder what would happen if they made a game where having a big empire started to hurt you. Yes, it would be realistic -- how many big empires have lasted all that long? But what would happen if the best way to play was "okay, i'm going to play to get big... but not too big"? Also, how would this be compatible with the fact that there's a victory condition based entirely on being the biggest dog in the yard?

True, but the very idea of "winning" history is ludicrous in and of itself. This isn't a Maxis game, so somebody has to win somehow. I'm willing to accept a bit of unrealism in the victory conditions provided realism is incorporated elsewhere. Besides, what sort of victory condition would promote realism? No matter what, you can always rest assured that people will want to kill those who are near victory.

I think one answer is to pull victory apart into several different broad goals. 5 points for being voted secretary general, 5 points for having X% of the world's land -- with an additional point for every Y% more. 5 points for each legendary city. 5 points for winning the space race.

This would actually encourage diplomacy. I'll vote for you for secretary general if you pass a resolution to enforce my war against Spain. Let's make an alliance: you help me out towards my conquest victory, and I'll let you build the space ship.

In the closest games, you still have big backstabs, yes. If you're about to win the space race and nudge me out 15 points to 10 points, I might just have to nuke you. But a new avenue would open up. Again, if you're about to win the space race, I might be able to negotiate a shared space race with the third place civilization and score myself half the points: 2.5 between me and my ally. I would have stopped you from winning the space race without backstabbing you! And moreover, it wouldn't be cheap, since I'd have scored fewer points than if I had done it myself.
 
Yes, dh_epic, something like that can help :)

It sound like a "realistic Mastery victory" -
we may call it Historic victory :) ... or Civilized victory :D

EDIT:

how about starting a thread on this idea, dh_epic?
Or bring it up in the "devil's advocate" topic?
 
Good topic. Count me in the camp that's simultaneously frustrated by the fact that the AI can't compete without cheating (in the form of bonuses) and the fact that the AI doesn't behave enough like a real leader. I'd like the AI to play smarter, but I emphatically don't want to see a game where the AI plays like I currently do, abandoning all else in favor of securing one of a handful of predetermined victory conditions.

I recall reading a comment in (I think) the Devil's Advocate thread complaining that the AI won't be effective as long as it cares about arbitrary things like state religion and favored civic and thinking that the complant was exactly the opposite of the problem. The AI won't be a satisfactory competitor as long as the player cares about arbitrary things like controlling 65% of the world's land or amassing 150,000 purple musical notes in three different cities.

I like the idea of spreading victory out over a number of conditions to eliminate the "race" factor of the game as it currently stands. Races can be fun, and could continue to exist for certain sub-goals (similar to the Liberalism race in the game today), but a first past the post victory condition encourages the sort of arbitrary goal setting that a player will always enagge in when necesary and that I'd prefer the AI to avoid. Rather than having each condition add a certain number of points to a score, however (really, I'd like to see scoring removed altogether), it might be more intersting if a player could win by completing (not necessarily by being the first to complete) some number of goals, which could vary based on civ, leaderhead, or something else, and even change over time, based on events in the game or player actions.

For example, maybe some leaders have Found a Religion as a starting goal. Maybe actually founding a religion automatically creates a new goal for the founder to spread that religion to X number of cities or Y% of the world's population. Perhaps if you lose a shrine city, you automatically have Reclaim Our Holy Land added to your list of goals, and if the shrine city is razed, you have Eliminate the Heathens Who Defiled Our Holy Land as a goal. If you are unwilling or unable to pursue an existing goal (if, for example, you were supposed to Found a Religion, but all 7 relgions were founded and you couldn't get any of them), perhaps you could go trade one goal for another, perhaps at the cost of something similar to anarchy (there should probably be a significant penalty here, to prevent players from always switching to whatever goals look easiest to accomplish).

It might be best if every civ has at least one goal that is unknown to other civs, to prevent a player from backstabbing anyone who is close to accomplishing all their goals. Perhaps each civ has 1 or 2 starting goals, each leaderhead has 1 or 2 more, and the player can select an additional goal at the start of play (random for AI's). The Civ and leader specific goals would be publicly known, but the player selected goal would be private.
 
Top Bottom