Understanding Firaxis's programmers.

[Coracle]

I read on the CFF that Soren actually admitted they deliberately gave latter-era military units too low values because in case a civ was missing an important resource they would still have "a chance".

This is the most effed up thing I can imagine in game design.

Firaxis creates a mod with ridiculously rare resources. Iron and coal are especially too rare. So, instead of fixing that, they offer us idiotic and non-historical unit values, especially for post-gunpowder units!!

Ridiculous.

[asleepathewheel]

Originally posted by Coracle
I read on the CFF that Soren actually admitted they deliberately gave latter-era military units too low values because in case a civ was missing an important resource they would still have "a chance".

This is the most effed up thing I can imagine in game design.

you must have a poor imagination.

Firaxis creates a mod with ridiculously rare resources. Iron and coal are especially too rare.

one per civ is too rare? thats up to 16. Why even bother with resources if they are much more plentiful?

So, instead of fixing that, they offer us idiotic and non-historical unit values, especially for post-gunpowder units!!

Ridiculous.

as opposed to the earlier units that have historical unit values? I didn't realize that military units had numerical ratings.

and you have a problem with the lack of resources, yet you have units like riflemen and longbowmen which do not require resources (among others, I can't recall the units that recquire which or no resources) that may allow you to capture a certain resource, if you plan carefully. But to you, thats ridiculous? do you just want a screen that proclaims your victory when you build your first armor?

Blah blah blah

[BillChin]

Originally posted by nbarclay
I also don't care much for the idea of attack scripts based on year. For me, the decision process is more along the lines of, "Okay, I just got the technology and resources to launch such-and-such type of attack if I want to. Do I have enough of a potential advantage that I want to focus on units and attack, or would I be better off waiting?" (And another variant: "I'm about to get the technology. Do I want to focus on units I can upgrade to the attackers I want?") That way, if I attack at a given level of technology at all, I maximize the time I have to attack while the attacker still has the advantage (e.g. swordsmen vs. spearmen, knights vs. pikemen, or cavalry vs. musketmen). And for that matter, my timing for what technology I try to get when is sometimes influenced by whether, for example, I would want to launch a cavalry attack in the near future if I had Military Tradition.

The more capable the AI is of basing its tactics on the game position instead of purely following a script, the better off it is. (Assuming, of course, that its analysis is competent.)

I am a big believer in process. The current AI is a closed system approach. The flaw to any closed system AI is that it can not adapt. The key point in my mind is that the AI must adapt if it is to keep up with human players. One of the best ways, is to let the same humans that beat the AI, to script the AI.

A script approach taps the virtually unlimited time and talent of the fan base. I might favor scripts based on time (Timing is everything ). You might favor strategies based on tech and resources. Someone else may favor strategies based on number of cities. That is the real power of scripts. With a rich scripting language, there would be almost as much diversity in scripts as human multiplayer. Different styles are very hard for a closed system AI done by a single design team to similate. By having random selection of scripts for each game, a human faces uncertainty. Is it Bill's script that I am playing against, or Nathan's or someone else's?

Your point about analysis may be an incredible hill to climb. In chess, this was cracked with a database of all grandmaster level games, every move, every board position and whether this led to win, lose or draw. This allowed top chess programs to do high quality analysis on any board position by comparing it to similar board positions and real world results. In a game such as Civ III, evaluating a single "board position" can become a gargantuan task. It may be very tough to figure out whether taking a certain city and probably losing five units is a better board position than not attacking. Even if the analysis is clear, I see predictability as a bad trait. Some other game has an AI that only attacks when the odds of victory are 75% of greater. Solid analysis, but it makes for a boring and predictable game. The more the human is left guessing, the richer the game play, the better the AI's chances.

I believe mimicking the best human players and adaptation are a good way to produce a better AI. Mimickry and adaptation are processes that work for any type of game, any rule set. Victory or defeat in the real world becomes the measuring stick. The best way I see to get mimickry and adaptation is a scripting language. There are others ways to implement an adaptive AI, but I see scripts as one of the most time and money efficient ways to go.

[The Thinker]

Originally posted by BillChin

From a recent poll, only about 20% of players play on Emperor and Diety level. I believe is a similarly small percentage, maybe 20% that really want a better AI. Those 20% have a hard time understanding what the other 80% are thinking

If you think I want a better AI so Diety level will be harder and more challenging, you are wrong. I don't play the hard levels normally, I don't micro manage enough, I don't go back to earlier saves, and I don't like losing.

Right now the AI is very stupid. To make up for this, the AI cheats right and left. This interfers with my having fun because this game is not Pac-Man, where game is doing something different from me but the computer is supposed to be playing the same game I am. Imagine if you bought a Bridge game and the programers had trouble making the AI tough so they let the computer players pass each other cards. Then imagine one day you notice that west ALWAYS has more trump then east. Now when you lose you blame the computer for cheating and when you win you victory is hollow because the computer was dumb and didn't play the cards he had well.

I want to feel that the AI and I are playing the same game.

Further, because the AI is dumb, it can't handle a mod well. Witness it's inablity to handle changes from the patch. I've tried to change things using the editor provided and can change the information on just one unit and it can cause the computers turns to get longer and longer until the game crashes.

Rik

[Inverse Icarus]

Originally posted by Ethelred


Its an inland sea. A lake is all coastal and you get to use it as a freshwater source. An inland sea is saltwater. If you couldn't treat it as such you wouldn't be able to get much use out of it.

I HAVE built destroyers on an inland sea myself. Used it to control the area around it by bombarding the enemies.

i'm not up to battleships in this game: but i have build galleons / frigates in that inner ocean. it helped me take the ring i have now

i AM planning to build 2 sh1t cities to act as "canals" to connect the main ocean to the inner one, but i still built a navy in an inland ocean

[Inverse Icarus]

i dont play diety, simply because the AI cheats rampantly.

i do not see any improvement in the AI's tactics, domestic public works, or trading ability except for vicious cheating and a flagrant anti-human mindset.

[Mark_Everson]

Originally posted by The Thinker
Right now the AI is very stupid. To make up for this, the AI cheats right and left. (snip)

I want to feel that the AI and I are playing the same game.

Further, because the AI is dumb, it can't handle a mod well.

I agree with Rik's points.

Doing AI according to a commercial game schedule is a big part of the problem IMO. Once the rules are nailed down there just isn't much time for good AI...

I have no faith in the commercial houses putting out an AI API that is versatile and deep enough to allow for really good AI. Of course worthwhile improvements at the margins are possible, and should be encouraged, but a breakthrough AI that many of us want won't come from this path IMO.

If anyone would like to check out some admittedly long discussions on the AI for Clash of Civilizations, I'd like to hear what you think. I believe some of our discussions there are directly applicable to the topic here. The one-sentence blurb is that the AI is hierarchical, having different levels that think about issues on different geographic and time scales. For a Much longer description and the discussion, see AI -- the thread . If you want to cut to the chase skip to my summary post about the next steps for our AI starting in my post of 13-02-2002 22:56 in that thread.

We're always looking for input from people with fresh perspectives!

[nbarclay]

Originally posted by BillChin


I am a big believer in process. The current AI is a closed system approach. The flaw to any closed system AI is that it can not adapt. The key point in my mind is that the AI must adapt if it is to keep up with human players. One of the best ways, is to let the same humans that beat the AI, to script the AI.

A script approach taps the virtually unlimited time and talent of the fan base. I might favor scripts based on time (Timing is everything ). You might favor strategies based on tech and resources. Someone else may favor strategies based on number of cities. That is the real power of scripts. With a rich scripting language, there would be almost as much diversity in scripts as human multiplayer. Different styles are very hard for a closed system AI done by a single design team to similate. By having random selection of scripts for each game, a human faces uncertainty. Is it Bill's script that I am playing against, or Nathan's or someone else's?

Your point about analysis may be an incredible hill to climb. In chess, this was cracked with a database of all grandmaster level games, every move, every board position and whether this led to win, lose or draw. This allowed top chess programs to do high quality analysis on any board position by comparing it to similar board positions and real world results. In a game such as Civ III, evaluating a single "board position" can become a gargantuan task. It may be very tough to figure out whether taking a certain city and probably losing five units is a better board position than not attacking. Even if the analysis is clear, I see predictability as a bad trait. Some other game has an AI that only attacks when the odds of victory are 75% of greater. Solid analysis, but it makes for a boring and predictable game. The more the human is left guessing, the richer the game play, the better the AI's chances.

I believe mimicking the best human players and adaptation are a good way to produce a better AI. Mimickry and adaptation are processes that work for any type of game, any rule set. Victory or defeat in the real world becomes the measuring stick. The best way I see to get mimickry and adaptation is a scripting language. There are others ways to implement an adaptive AI, but I see scripts as one of the most time and money efficient ways to go.

I think the key is to use a combination of scripts and analysis of the game position. The AI has to be sophisticated enough that it can choose which of several paths to follow depending on the situation. A scripting language with full access to information about the game status and the ability to change what branch of the script to follow depending on the situation could accomplish that. Ideally, the scripting language would incorporate a full-featured programming language but would allow developing simpler scripts without using all of those features.

An AI based on such an approach could do quite well as long as the only situations it runs into are situations covered by its scripts and its instructions on which script to use when. But it would have no real creativity for recognizing or coping with situations the scriptwriter forgot to cover. That's the area where I think incorporating other AI techniques could be useful: some way of looking ahead and figuring out, "Hey, following this script would probably be suicide."

Nathan