Theory of AI

[ShadowWarrior]

What is the differences between playing against AI and playing against human players via multiplayer?

The chief difference I believe is that AI is programmed to destroy us human players, while human players' goal is to survive.

Imagine a game where I play against three AI. Since AI is programmed to destroy us, we are basically playing a game where alliance with AI is at best superficial. That makes alliance, as well as most of other instruments of statescraft typically available to statesmen useless to us human players. Therefore the only thing we can possibly do in this game is fight wars. AI will even go to the length of doing ridiculously stupid things to make sure that we are destroyed.

Human players are different. We humans want to survive. Imagine myself playing agains three other human players. The three other human rivals are not bend on seeing to my destructions only. These three other human rivals of mine wants to make sure that the other two will also get destroyed. It is in this environment where killing me is not the only ultimate goal that we can realistically expect sophisticated diplomacy, alliances, and other instruments of statescraft to be useful. I and human player #1 might decided to ally against human player #2 to prevent the latter from dominating the both of us.

If I were playing against AI, then I can't possibly ally AI#1 against #2 and expect some kind of real tangible help from my ally because AI is programmed to destroy me, not another AI.

In another word, only when AI are really programmed to act in the interest of their own survival, not in the interest of destruction of human players that we can expect to be able to use sophisticated diplomacy, and other features of empire building.

However, programming AI to act like humans seem technologically infeasible. Therefore, I suggest that the designers instead shift their attention to making multiplayer MUCH more accessible, convenient, and easier.

 

[warpstorm]

ShadowWarrior, Civ3's AI doesn't gang against the human player (in fact, it doesn't seem to be particularly tryinbg to win the game). I don't see why they would step backward on CIV's AI.

If CIV skimped on the AI in favor of a MP only approach (or even a mainly MP approah) I would skip on it and leave it in the box on my shelf.

 

[Aussie_Lurker]

Hell, Warpstorm, I would probably leave it in its box IN the videogame store I was shopping in at the time!
If for one have always played SP, and have yet to play an MP game yet!! If they were to focus on MP over SP, then I would probably just 'walk on by' and find a game that catered to my SP addiction !

Yours,
Aussie_Lurker.

 

[warpstorm]

Well, I've got friends who give me signed boxes for my shelf.

 

[socralynnek]

What is the differences between playing against AI and playing against human players via multiplayer?

The chief difference I believe is that AI is programmed to destroy us human players, while human players' goal is to survive.

I think, it's the other way round...

Hopefully the AI will also be in Python, so everyone could have its own AI!

 

[warpstorm]

Good point, socralynnek. In the MP games that I've played, the opponents were even more ruthless, cutthroat and less willing to cooperate than the AI ever was.

 

[Volstag]

I agree with Warpstorm: The AI in Civ III is not hell bent on destroying the player. If it was, it would be a totally different story. The problem with the AI in Civ III (or almost another other strategy game, for that matter) is that it lacks the ability to develop long range goals/plans -- it develops purely reactionary, short term goals, on a turn by turn basis (as far as I can tell... I'm sure there's some basic long range planning).

We've all played games where there's been an enemy civ (or three) that could totally steamroll everyone on the planet, if they could only put together a half-way effective long range plan.

-V

 

[Stid]

i think both AI and human players play the same except for one difference and that is the computers predictability such as when they send 20 bowman into your territory and they're not protecting settlers what they gona do?? we both play to survive but if it was a human, they build up an army adequate to defeating opponents depending on their size, tech etc but the computer just rushes straight in.

 

[Jon Shafer]

Well, I've got friends who give me signed boxes for my shelf.

Well you don't have to go and brag about it.

Volstag is right. The problem is that the AI doesn't plan for things like a human player does. It doesn't figure out that they'll declare war on Civ A on turn X and should have Y units ready and in place by then. It doesn't know that wonder M comes with tech L and should prepare city 3 for that purpose.

 

[dh_epic]

I don't even give the AI that much credit. I think the AI is programmed just to prevent the human player from winning, not even to win themselves.

At any rate, there's a LOT of work that needs to be done on the AI front -- but I don't think it's as simple as "make the AI better". The goals for a good AI are important.

Which is why I often emphasize this solution:

- Have 80% of the Civs be there for historical realism, cooperating with those honorable and strong, and attacking those who are dishonorable or weak, and otherwise treading water

- Have the other few Civs be there for competition, who try to stay neck at neck with the human player, cooperates with the ruthless to stay ahead, and pounds on the trusting to get an advantage

 

[warpstorm]

I think the AI is programmed just to prevent the human player from winning, not even to win themselves.

No, they don't seem to even try to prevent you from winning. I never get attacked just because I have a major cultural lead and am going to win in a few turns nor do they try to take me down just because I start building spaceshiup parts.

 

[Dwarven Zerker]

I don't even give the AI that much credit. I think the AI is programmed just to prevent the human player from winning, not even to win themselves.

Actually, the AI is programmed to respond to certain conditions. Those reactions are aimed at preventing the human from winning and resulting in a histographic win for the AI should the human still be alive at 2050.

At any rate, there's a LOT of work that needs to be done on the AI front -- but I don't think it's as simple as "make the AI better". The goals for a good AI are important.

A lot of people keep asking for an AI that can make plans (long or short range.) That is very difficult to impossible to do at this time. Object oriented programming is really good at giving the computer behaviors (reactions) but isn't so good with providing goal making decisions.

Which is why I often emphasize this solution:

- Have 80% of the Civs be there for historical realism, cooperating with those honorable and strong, and attacking those who are dishonorable or weak, and otherwise treading water

- Have the other few Civs be there for competition, who try to stay neck at neck with the human player, cooperates with the ruthless to stay ahead, and pounds on the trusting to get an advantage

The best way to do what you're looking for is to program a few "personalities" that an AI is capable of having. Each personality would emphasize a different aspect of the game such as warmonger, peaceful spaceship builder or loyal ally.

 

[dh_epic]

Yeah I guess you're right, the Civ 3 AI doesn't do much except progress at the speed the difficulty level permits. And the only reactions it will make are to attack a civilization that is either excessively weak or excessively untrustworthy. Otherwise, they'll leave you alone.

I do think that having AI personalities might be a good situation. Have 90% of them play with realistic personality -- with a sense of loyalty and betrayal, but not necessarily conscious of the "game" element. Then have 3 opponents who are playing with a "self conscious" personality -- that has a sense that it IS a game, and that time is running out from the first minute on the way to a huge competition.

I guess the first one would build on the Civ 3 AI.
And the second one would build on the Civ 2 AI.

 

[Dwarven Zerker]

I started designed a low tech RPG game in which NPCs would have "personalities" consisting of different mindsets. Each personality was comprised of a level of aggression, greed and "intelligence". Each of these components had three settings - low, med and high if you will. Highly aggressive NPCs would attack sooner than a low or mid level aggression. Greeder NPCs would pay less for items it bought and would cut off negotiations sooner if the price it would pay for an item fell too low. Intelligence reflected how it would train it's skills - low intelligence would train the same skills where high intelligence would diversify it's training.

With a variable assignment of each of those three components each NPC would seem to be an individual despite the simplicity of the behavior model. I think something similar could be done with the Civ AI. Make the personality components for aggression, loyalty and "greed" - similar to my RPG greed dynamic. Each of these could be randomly generated as a part of world generation at the start of each game and make each leader different from game to game (who says Ghandi always has to be passive? Isn't this a game of alternate possibilities??)

 

[Red Ant]

Actually, the AI is programmed to respond to certain conditions. Those reactions are aimed at preventing the human from winning and resulting in a histographic win for the AI should the human still be alive at 2050.

Seriously?? I think that sucks big ass. The AI should NEVER think in game terms IMO (i.e "Only 10 turns left and the player is leading ... smash him NOW or he'll win the game!"). Much rather they should solely be interested in .... well, just in doing well. Yanno, having a good economy, being powerful and alla that.

 

[dh_epic]

Red Ant, you raise a good point, but AI is kind of in a catch 22 that way. If AIs just "do their best", they fail to account for how vicious and tricky the player can be, let alone other AIs. You can see this in Civ 3, where you can pretty much run away with the game and you can still have every AI love you or at least respect you.

But on the other hand, if the AIs are always concerned with "be better than the player, don't let them win" you end up with a Civ 2 thing. It makes for a more challenging game, but it makes for a really one dimensional, predictable experience. Imagine if God came down and announced 2050 was the end of the world, and the people who would get the nicest place in heaven would be the people in the largest empire.

- Switzerland would be pissed
- China would kick themselves for not occupying America in 1412
- The age of discovery would have happened a lot sooner than the 1500s
- WW2 would have ended in genocide -- by the Allies OR the Axis
- The Incas would have re-evaluated the benefits of expansionism
- The Near East would have conquered all of Europe while it was vulnerable, in the dark ages
- The Iroquois and the Sioux wouldn't have yielded an inch to the Europeans
- There would be no EU, no NATO, no NAFTA, no WTO, probably not even a UN
- France would have been conquered a long time ago

As you can see, you lose a lot of MEANING to the game when you make it, well, a *game*. Because the competition towards one specific goal (controlling 50% of the world) ruins any meaningful interaction between Nations and Civilizations.

But when you have the AI ignore the fact that it's a game, that's when the player exploits that weakness and dominates the game every time in the exact same boring way.

Catch-22.

I know I've been pushing this idea a lot -- have you ever heard me distinguish between "flavor AI" and "competitive AI" and how it can help Civ 4?

 

[Dwarven Zerker]

Seriously?? I think that sucks big ass. The AI should NEVER think in game terms IMO (i.e "Only 10 turns left and the player is leading ... smash him NOW or he'll win the game!"). Much rather they should solely be interested in .... well, just in doing well. Yanno, having a good economy, being powerful and alla that.

That's not exactly the way the AI is programmed. It doesn't have any concept of the 2050 deadline whatsoever. That would show that the AI is capabale of advanced planning. Currently, its only concern is reacting to the current state of affairs for the next turn, whether that turn is be 3950bc or 2049ad. The AI reacts the exact same way in 2000bc as it does 2060ad with the only difference the units it has available to utilize and other game produced differences (empire sizes chief among them.)

 

[general_kill]

I thought the AI situation was already dealt with. Just make an editor in civ 4 that allow players to create and add AI scripts that is simple enough so the majority of players can easily tweek the AI.

 

[Red Ant]

I've heard some say that there will indeed by such an option in Civ IV, essentially making it possible to guide the AI using Python scripts hand-rolled by the player. Anyone know if that's true?

 

[dh_epic]

I know people talked about it being a logical plan, but I haven't heard it being confirmed.

If it's the case that the AI can be modified with python scripts, can we handicap some of the AIs and not all of the AIs? Give some a cheating advantage, while others play fair?