No AI, no matter how much it will be improved (and I hope a lot), can be expected to beat an experienced player by fair means. So let it be by foul means...

quote: Originally posted by Daimyo on 03-21-2001 11:07 AM I wonder how many civer's are that **good enough**to beat the AI easily on the toughest level..

Several dozen at least in the strategy forum on this site. Those who can beat the game easily probably only make up a tiny fraction of the people playing the game, but we're a vocal minority...

Well, if you use a rule based AI, it doesn't take long for even the least experienced civer to work out that the AI always reacts in the same way .. and unfortunatly, nearly all AI's are rule based.

What I have proposed for eons is the partial use of neural nets .. basically, keep the rule base as nature, and allow the AI to learn through nurture .. how does it learn ?? from YOU !!

eventually, a sucessful nuture can be passed to nature .. and failing nature can be removed ..

This evolutionary approach will ensure that each game is unique ..

If AI cheating is to be considered, it should be user controlled .. option to turn it on, and even select the kind of cheating level you want .. so you really can make it impossible for yourself, if you so desire.

You see the AI toggling cheat mode. you see the AI eliminating all of your cities. You see the AI add all advances to himsellf. You see the AI build the space ship in one turn.

Is that how you want it to cheat?

If there is any cheating (to circumvent AI bloat code and analysis) then it must be of the form that does not affect player to player interaction.

For example if the AI is cheating and not letting the city go into revolt and I send in a spy to cause civil unrest than it is affecting my game play and wasting my time.

The AI needs to learn! Something that I have been pondering for a long time is a slight expansion on the rules+neural net idea above. A rules+neural net+"apolyton for AI" where AI's can turn in good rules, and check new good rules out.

Another way is to do it like crisium is planning to do it for stars supernova. (www.crisium.com) They plan to let you set a time limit for AI pondering of problems: (Like, spend the night on it boy.) And their AI is going to learn from actual playing. (They are trying it out in the beta to see if it learns "the right stuff" or not.)

quote: Originally posted by The Viceroy on 03-22-2001 10:22 AM What I have proposed for eons is the partial use of neural nets .. basically, keep the rule base as nature, and allow the AI to learn through nurture .. how does it learn ?? from YOU !!

An AI that remembers how you played when you defeated it, and tries to use the same tactics next time ?

If that could work, it would be absolutely wonderful !

We could even make contests to see who could make the smartest AI this way.

I know if I get really pissed off with a computer Civ, I'll send a land sea air invasion to teach him a lesson (i.e. don't mess with me). But will the computer know how to do the same in return/advance?

AI should not announce when its about to finish a wonder.
The AI should always get the latest improvements in all cities and not bother to build them. (in later stages of the game)

I want an additional field in the rules.txt for "keep this unit type inside cities forever" ie pikemen or warriors, i can't stand when they run around with useless units.

The AI should work together to attack you from multipul points. The AI should work together to impose trade embargo's.

There has to be an option that you can set that says, "never recieve tech from capturing a AI city" and another option for "never give tribute"

The AI as a whole is as previously noted..predictable and simple in its methods..I cant wait until an AI becomes a smarter more competitive source of entertainment!

I hereby cast my vote..CHEAT CHEAT..You wont ever Beat!
LOL

Yours in Civin

Troll

------------------
Hebrews 11:1

But a pretty darn good AI could still be made. Especially if they can turn in strategies to a central repository and check out new strategies. (Like we do here at Apolyton.)

It's like the code breaking competitions. No supercomputer in the world can break codes as fast as the distributed programs do. (Screensavers that work on the fully parallellizable task of factoring huge prime numbers, ie 56 bit keys.) Or the SETI screensaver. No one computer does alot of work, but it adds up.

Especially if the AI has some rudimentary skills at analyzing human moves. (One might include an "omniscient learning AI", ie the AI that you play against only sees what it should see, but another deamon runs in the background and tries to analyze your moves. When it learns a city blockade technique it adds it to the list to send to the repository. When the neural net AI that is your opponent comes up with something new it adds it to the list. The list is sent in, and a new list is checked out from the repository.)

Seriously though, cheating is not the way to go... unless you mean trying to simulate human intuition, where you allow the AI to know a few facts about the map/opponets/army positions that it can't see at random, therby allowing the AI to "Guess" things (like humans do), the higher the diff. lvl., the better they guess and the more often they do so.

quote: Originally posted by Trachmir on 04-03-2001 12:08 AM For all of you who really want the AI to give you a run for your money, I suggest you let the AI do one thing: LOAD AUTOSAVES!

ROTFL!

But they already had something on this line: do you know when the game suddendly crash and you think it's a bug?
Nah, it's true "HAL 9000 like" AI behavior. Thanks God it doesn't try to electrocute you!

1) Civ2 doesn't have an AI, it has a pseudo-random number generator. Virtually everything it does is chosen at random. The Smac AI just about qualifies as a 'player', though, so there's been some improvement (not much!)

2) Comparisons with chess AIs are really misleading. Believe me, I've written one. You should be able write an AI which would frequently win on Deity level vs the normal AI without cheats, though the best human players would still kick it's arse pretty much whatever you did; that's the benefit of creativity.

3) Writing an AI which learns from it's experiences is very hard in a game like Civ. It could learn a number of cities to build and which units work better, but that's about it, and it could be set more easily and effectively by a human editing a user-accesible script (hint-hint.) Writing a Civ AI which learns effectively from human players is pretty much AI-complete. Firaxis probably isn't planning to develop a human-level AI, so I'd forget about that one.

4) Because of 2 and 3, there need to be cheats. Otherwise, some people (even only 100 or so) are going to be severely bored of the single-player game after a week.

I'll leave you with some words from the Civ2 manual....

"[In Civ 1], we predicted that Emperor level could be beaten, but not consistently."

quote: Originally posted by Insigna on 04-05-2001 06:38 PM I think that AI cheets should be an option. Have the regular Levals Prince, King etc. But then have the ability to put it on AI cheeting with these levals if they want an extra chlange in King or Emporer or whatever. A bad AI was my main dislike of CTP. They should concentrait on a good AI.

Yeah, except it would make no business sense. The AI always seems intimidating for the first week, so all reviews will be praising the revolutionary new AI, when in fact there is no such beast. Almost everybody who bought Civ II shelved it before they even started a game on Diety level. (A second note on reviews, the big companies always get good reviews, because they cancel all advertising after one bad review. The secret behind the scores of Westwood games in gaming magazines.)

All it would get them is the respect of a few old farts in the fan community, and that generally doesn't amount to much cash-flow.

Good AI won't bring in money, as they can get the same money by lying about having a good AI, because no one will spot it in time for the reviews anyway. And most sales of a game is made in the first month or two of it's lifetime anyway. So it's cheaper to have Sid or whoever just stare right into the camera and say that sure the AI will be groundbreaking, when it in reality will be just warmed up code from the original Civ. Compare this with hiring someone with the competence to write a good AI. (That person has atleast a Ph.D and will want a good salary. Probably a fair bit over 100k...)

Now take on the role as marketing dept, and look the developer in the eye and tell him: "Go out there and lie, they will never see it in time."

quote: Originally posted by Zanzin on 03-21-2001 11:37 PM Cheating AI!! No!! I'm sick of AI's that cheat!

Me too! What's the use of providing an amoral AI that uses every trick in the book to give just a few expert players a good run for their money, while the much larger majority of players - like me! - are getting utterly frustrated with it?

If, as many people seem to be saying here, you can't have a tough AI without allowing it to cheat, at least make cheating an option that can be switched on or off by the player. Then those of us normal mortals who don't "easily" beat 6 AI's at deity level (!) can at least build up gradually to the point where we can allow the AI to cheat and still win...

quote: Originally posted by Lord Maxwell on 04-07-2001 04:36 PMCompare this with hiring someone with the competence to write a good AI. (That person has atleast a Ph.D and will want a good salary. Probably a fair bit over 100k...)

I really doubt that many Ph.Ds in AI could write a decent Civ AI. The problem is way, way off the mainstream of AI research, for a start. The person needs to be a damn good strategy gamer, as well as a damn good coder who can practically invent the field of tbs AI as they go along. The only PhD strategy god I know of is Zsozso, and he's in biochemistry.

quote: Originally posted by Simpson II on 04-07-2001 08:16 PM I really doubt that many Ph.Ds in AI could write a decent Civ AI. The problem is way, way off the mainstream of AI research, for a start. The person needs to be a damn good strategy gamer, as well as a damn good coder who can practically invent the field of tbs AI as they go along. The only PhD strategy god I know of is Zsozso, and he's in biochemistry. Naturally, I could do a great AI for them. Cost a lot less than $100K, too...

You can code a good AI in less than a year with a yearly salary under 100k. Hmm, are you unemployed and willing to move?

Maybe it's ok to have a few cheats. Certaintly that would make things a little bit more harder for the human player.

You sned 40 vet spies at a city to reduce some improvements, all are captured without completing their mission.

12 AI dips come into one of your cities, stealing tech. Suddenly they have all your techs, including nuclear power, which you just discovered. They build Smokey da nuke project the same turn (sorry, Manhattan). you get nuked the next turn.