The AI cheats! (Possible spoiler alert)

[Bhruic]

What AI could learn from Deity players is how to pursue etreme tactics to achive a finite goal. For example: The Liberialism race is close, human players could switch to caste system, starve all ther cities for a few turns and run max scientiststs, just to win the race. I don't think the AI will even consider doing that, when it should, as the ways of researching faster for a while are very specific and limited in this game.

Another example is culture wins, do AI's switch to Universal surfage, just to ruch buy cathedrals for a few turns and then switch back to representation for the faster teching? This is a simple finite technique and they should be able to calculate it, if the programmers told them to consider it.

That is why deity players are needed to help programming the AI, because the programmers have never though of that themselves and consequently have not programed the AI to even consider it.

Sorry, but those aren't really good examples. AI programming is designed to be extremely generic. There are very few examples of specific case profiling - and with good reason. There are generally hundreds, if not thousands, if not many times more than that cases that could be considered. Specifically coding for each instance is futile, and a waste of time. Especially when you consider that Civ is designed to be modded.

Good game AI is going to deal with the simplest ruleset and work within that to achieve whatever desired goal. The more specific you try and make it, the more time consuming it becomes, with increasingly limited effect.

Bh

 

[Indiansmoke]

Sorry, but those aren't really good examples. AI programming is designed to be extremely generic. There are very few examples of specific case profiling - and with good reason. There are generally hundreds, if not thousands, if not many times more than that cases that could be considered. Specifically coding for each instance is futile, and a waste of time. Especially when you consider that Civ is designed to be modded.

Good game AI is going to deal with the simplest ruleset and work within that to achieve whatever desired goal. The more specific you try and make it, the more time consuming it becomes, with increasingly limited effect.

Bh

In chess computers use databases with human games to better get a "feeling" of the position and to help them evaluate their possible moves.

I don't see why something similar would not help in civilization.

I disagree that there are thousands of cases like the Liberialism one, maybe I don't know the game good enough though.

However I agree that if you manage to make it work with a simple rule set and without getting into specific situations that is optimal...but if it could play better just by adding a few "ideas" for the AI to evaluate under certain circumstances then why not?

It has already happened to some extend and you did it yourself as well. Things like build more defenders early, knows how to steal techs, spend less on espionage, spread your corporations etc are not based on generic rules, but are specific things.

The modding thing is an issue, but then you could have a separete AI engine, more generic, just for modding.

 

[Gooblah]

You can't use Chess as an example. Couple things:
Chess has a finite set of moves possible, and only one way to score victory.
Chess has 6 unit types, and 20 units per team.

These are a few major differences. These differences enable an AI to be programmed for a Chess game, like Deep Blue.

A Civ game has 5 Victory options, near, if not infinite possibilities, and around 100 units, if not more. Programming an AI to investigate ALL possibilities would be horrendously difficult, time consuming, and would take up so much memory, it would be insane...

 

[Breunor]

And remember my other point about chess -- Chess AI was very, very poor for about 10 - 15 years. I remember playing the 'strongest computer inthe world' in the mid-70's; we were about even (I won). I was an 1800 chess player at the time, probably the equivalent of a Monarch player in Civ. This was 10 years after the David Levy bet that started the chess AI craze.

Best wishes,

Breunor

 

[TeraHammer]

Chess is fundementally different from civ because chess only involves military.

 

[r_rolo1]

You can't use Chess as an example. Couple things:
Chess has a finite set of moves possible, and only one way to score victory.
Chess has 6 unit types, and 20 units per team.

These are a few major differences. These differences enable an AI to be programmed for a Chess game, like Deep Blue.

A Civ game has 5 Victory options, near, if not infinite possibilities, and around 100 units, if not more. Programming an AI to investigate ALL possibilities would be horrendously difficult, time consuming, and would take up so much memory, it would be insane...

And the civ board changes in every game ( and during game: chops, improvements, razings, events )... and in Civ we have ( normally... ) more than 2 players, that scrambles things even more ( instead of 2^something we have 7 or 34^something )...

The fact is that ( like it said a post that I saw about civ III a long time ago... ) we can't transpose brute force methods of chess ( simply calculate all the variants until x move and choose the best ) or use the grand masters plays in memory ( first because there aren't Civ IV grand masters ( there are very good players, but I doubt that any Civ IV player in the world has the kind of preparation for the game as a chess grand master has for chess ( with his personal team trying to discover new/forgotten moves, the trining since youth ,1/2 decades of play and all the rest ) ), second because civ IV players do not take notes of all their moves ( including queue zapping, citizen management and even the path of the units in terrain ) and without that a computer doesn't understand a bit of what we did, third because even if we had the previous two, because the map changes every game nothing is directly transposable for other game... ). Without that, all that we can teach the AI is some tricks, like bombarding the cities before attacking ( and even there we have to say to it that probably it is not a good idea to bombard until 0% a city with a warrior as a garrison when we have +20 cavalry, 30 cannons and 15 rifles ) or that if we are being attacked by a single unit type force, is better to build his counter..... I.e. tricks, instead of real creative ( and learning ) inteligence, a thing that would take atleast some decades