AI and the game

[Roman]

Many people say that some features should not be included in the game because they would cause trouble for the AI.
I am of a different opinion. Obviously the simpler the game the easier it is for the AI, but I do not think that should be the overiding factor in considering which features to include, though it is wise to take it into account. After all, as much as I enjoy chess, Civ III ought to be different.
Lets face it for a relatively complex game like Civ 3, we will never have an ultra- challenging human-like AI, so lets not base decisions on what features to include solely on this aspect of the game.

[Roman]

I also think that for the AI to be challenging in this type of game it needs to cheat, so this should be done. Lets say the AI should always have double production or something like that. Any suggestions on how to implement this without it being too obvious that the AI is cheating?

[tniem]

Double production? Seems to be going to far.

One of the biggest problems with the AI in Civ II, SMAC, etc. is that the AI was not able to handle diplomacy. You have two nations at war, one asks you to join and as soon as you do they make peace. Diplomacy really needs an overhaul.

[Roman]

Well, yes, diplomacy AI does need an overhaul, but my point is we will never have a challenging AI unless it cheats. While double production might be a bit excessive, I think the AI needs some similar type of benefits (at least on Deity) to be challenging.

[Changmai Beagle]

I think it would be helpful to use the structure from Railroad Tycoon II, where you can modify the amount of production advantage (income) the AI gets, among other variables, with a %challenge outcome. This would be better than an automatic fixed amount 'cheat' built into the programming - IMHO.

[wittlich]

Well how about not having the world map blind to the AI. I really believe that this already the case since no matter where the AI's start out at, they always seem to make a bee-line for the human civs...almost as if they're following a homing beacon!

[DrFell]

Double production would IMO be too obvious a cheat. I dont belive the ai should get some of the bonuses it does in civ2 (defence vs. barbs bonus). I think production bonuses (ie making units cost less, etc.) are the way to go. I still want a difficult level where the ai doesnt cheat at all (even thought it might be easy) or an option to turn off all ai cheating so i get the advantages of better ai on the harder levels but no ai cheating.

[Straybow]

I agree it would be nice to separate the AI level from the AI production advantage.

[Roman]

I think separating production advanteges from the AI difficulty level would be difficult, as they are the main difference between the difficulty levels. I don't think you would find the AI too challenging if it did not cheat (though the amount of cheating it does seems to be insuficient at best...).

[Adm.Naismith]

Nah, I must dissent!

Wild cheating is not a way to make a game more interesting. I can accept some help to AI, like having some "awareness" of my units position, but only as a general point, or as a view range a little larger than supposed line-of-sight.

I don't like when my units are hidden in fungus (SMAC game) and enemy start bombard me from two squares away: at least a scout unit sent in my direction before bombarding will let me accept the cheat as a "lucky enemy patrol find me".

Really, the best way to go is: put in game mainly features the "AI" programming code can manage, and try to separate as possible AI code (variables and parameters, at least).
This way we can hope Firaxis (or expert players, if customization is quite open as in CTPII) can upload different "AI themes"

Just for example, a good AI navy tactic can be different from a ground one, as from an Air superiority one; it's quite difficult for AI to chose the right to use in open game, but if you specialize them and load them as included in dedicated scenario it could make things a bit more interesting.
In the same way, after a month from the game release, the more dedicated players surely can suggest "hands down" winning tactics but, debating them openly on forum like this, they can also suggest countermeasures that Firaxis can try to include in a dedicated AI upgrade or "theme".

Look, I know a dedicated programmer is an expensive resource, but Firaxis can consider her/him as a great marketing move to keep the game fresh enough for a year at least. About code testing, the AI upgrade should be released AS IS, while lot of people here will be more than happy to try and give right feedback without unnecessary ranting.

About features AI can manage, I'm not suggesting to reduce CIV III to a Tic-Tac-Toe game (remember the movie "Wargames"?) that AI can win or draw every time, but more about game "template", as best city positions precalculated for AI at map generation: not a complete map view, only a knowledge of "interesting points" that AI must use as reference while moving settlers.

BTW, there are a lot of good not so old threads around here on the same topic.

Massive AI cheats will spoil the game, as humans playing full games with "cheat code=on" do, while a quick help here and there can reduce frustration in long adventure or make AI mid/end game more interesting.

------------------
Admiral Naismith AKA mcostant

[The Viceroy]

I agree with Adm.Naismith .. As id already posted in a previous thread, cheating is a good way of speeding up the AI process, but it leads to obvious flaws in the AI logic, which the human player quickly catches on to .. and exploits. Things like double production are good idea's for hard levels, but without a reasonable intellegence guiding this force, its worthless.. In CTP2, the computer generates almost 2 times more units that I do , but because it refuses to attack .. I never loose.

It is possible to create a damn good AI without such short cuts, and with a bit of pre-planning the AI can even learn from your tactics .. making it harder to play against each time .. all it needs is a decent knowledge base ... >

Im sure Firaxis will use whatever resources they have to get the job done.. if they have some AI experience, they may well use Neural nets .. if they don't.. they'll stick to the rule based AI, which can't learn .. but big enough, can be quite a challenge.



------------------
"Wherever wood floats, you will find the British" . Napoleon

[Roman]

I found this interview with Maranthea, a designer for the 3DO Heroes game series. Heroes 3 is a turn based strategy game of approximatelly equal (if not higher) level of complexity as Civ 3.

"I wasn't part of any real debate like that for Heroes III, but I do know that input from the testing department was able to keep the game from being too hard. Our AI programmer was so good that no one could beat the game without cheating, so after some discussions the AI was brought down a notch or two"

Whoa, maybe there is some hope after all!! I have played heroes 3 numerous times and when playing together with my sister, we can beat 6 computers allied against us. On hard though, we almost always lose. On hardest ... I haven't even tried. In Civ 2 though beating deity alone is no problem at all. Maybe good AI is possible after all.

[Roman]

On second thoughts, maybe the quality of the Heroes III AI can be attributed partially to the fact that no new cities can be founded or the land improved. This is partially what kills the AI in Civ 2.

[Grumbold]

I would have to dispute the quality of the AI in Heroes 3. It may perform well in some areas, but there are ploys that it falls for every single time, no exceptions. An enemy hero with enough troops to stomp all yours combined will spend its entire life marching back and forth recapturing two locations that two of your heroes tag-team capture every time he marches away. Many games make this mistake. Leave a prize unprotected deep in your territory and suddenly an enemy army will be striding blithely into your ambushes because the priority to capture the prize (that it should know nothing about) is more important than the immediate goal of keeping itself alive.

The first challenge should be to provide an AI which competes acceptably against the player without any artificial enhancement whatsoever. A competent, if not sparkling, method of attack combined with a diplomatic model that allows AI countries to form defence pacts and exchange technology if the player is seen as becoming the primary threat is a good start. The player may be able to outperform two AI civs but will have to be wary if three or more are invading. Different difficulty levels need only change the size of threat the player poses before the alliance begins. For Deity level it could even be permanent! Only if this level of competence cannot be achieved should the AI begin to cheat. Mind you, if it can't be achieved I believe the AI is going to suck no matter how many cheats you bolster it with. The cowardly CtP2 AI is a prime example.

[Cyclotron]

quote: Originally posted by ChrisShaffer on 02-08-2001 12:39 PM This is wrong. It's entirely possible to have a competent AI. Today's desktop computers are certainly up to the job. The question is whether Firaxis will put the resources into the project.

I agree completely! Cheating may make the game harder, but not more fun or more challenging. I know I would get bored of it fast. There is just no substitute for a good AI, and I hope Firaxis knows this!