Understanding Firaxis's programmers.

[notyoueither]

Originally posted by Coracle
The AI being programmed to build cities - as many cities as possible EVERYWHERE - is NOT an improvement. It is asinine and a big joke.

Seems to have beat you. Pity they get Oil and you don't.

[Sze]

Originally posted by asleepathewheel


I think thats a pretty significant improvement in the game, you do not? For me it changes the AI from a patsy (in Civ2) into something that is more competitive

So then why did they at the same time introduce the crippling corruption model? Isn't that designed to stop ICS? And I'm not arguing that Civ 2 had a good AI, I'm saying that I expected more.

Have you actually played Civ3 (it seems many of people who complain of the game actually have little or no experience with it)? The AI IS better than in Civ2. Whether this is because of the streamlined trade and spying or because the AI expands rapidly and is more able to attack, I don't care, it makes the game more challenging (for me) and thus more enjoyable.

What you call 'streamlining', I call 'simplifying' or rather 'dumbing down'. You do realize that all of these 'simplifications' will make MP less interesting, right?

I will concede that the new trade system is superior, but pretty much every other change made to the game in the name of making the AI seem smarter just made the game less enjoyable overall.

And of course I've played it, though it's been a while because the game sucks. Maybe I should play with the new patch to see how much smarter the AI has become, like when it self-destructs in communism.

[asleepathewheel]

Originally posted by Sze


So then why did they at the same time introduce the crippling corruption model? Isn't that designed to stop ICS? And I'm not arguing that Civ 2 had a good AI, I'm saying that I expected more.

Um, I'm not arguing that there is ICS in Civ3. And I'm not sure that it corruption is "crippling" as you say. Furthermore, I am not one of the programmers or designers, how would I know why they made design decisions?

Originally posted by Sze
What you call 'streamlining', I call 'simplifying' or rather 'dumbing down'. You do realize that all of these 'simplifications' will make MP less interesting, right?

Ah, if you had just said that. I don't care about MP. Never played it before on a Civ game and never will. no time for that.

Originally posted by Sze
I will concede that the new trade system is superior, but pretty much every other change made to the game in the name of making the AI seem smarter just made the game less enjoyable overall.

See, for me, getting rid of crap like caravans and other stuff actually makes the game more enjoyable. But if you like to move hundreds of caravans around the map to build a wonder in one turn or squeeze out a few extra gold, good for you.

Originally posted by Sze

And of course I've played it, though it's been a while because the game sucks. Maybe I should play with the new patch to see how much smarter the AI has become, like when it self-destructs in communism.

Ah, the old, continue to post about a game that you don't play anymore. Fair enough, you paid your money (unlike some others) so you should be allowed to complain, but really, the patches change actions of the AI, so maybe you should play with the new patch to see what is actually going on now, rather than just commenting.

[Coracle]

Last night with that new (crappy) patch the stupid AI as the English started building MOW in a ten tile landlocked lake. My idiot city governor started building battleships in that same lake.

So much for the programmers improving on Civ 2.


As for caravans, in Civ 2 we at least had the ability to INTERDICT and cutoff some overseas trade. In Civ III we have NO WAY to damage an enemy's merchant shipping by getting on his trade routes - thus making subs and privateers pointless. And saving up caravans for a Wonder was a nice piece of advance planning.

As for Espionage, that too was better in Civ 2. Civ III's Espionage stinks even at lowered costs through the Editor. It's best use is getting one to declare war on you. Give me those spies; it was a lot more fun.

[asleepathewheel]

Originally posted by Coracle


As for caravans, in Civ 2 we at least had the ability to INTERDICT and cutoff some overseas trade. In Civ III we have NO WAY to damage an enemy's merchant shipping by getting on his trade routes - thus making subs and privateers pointless. And saving up caravans for a Wonder was a nice piece of advance planning.

um there are ways to disrupt trade. Destroy roads, blockade harbors or sign trade embargos

Originally posted by Coracle
As for Espionage, that too was better in Civ 2. Civ III's Espionage stinks even at lowered costs through the Editor. It's best use is getting one to declare war on you. Give me those spies; it was a lot more fun.

Parts of the espionage were better, but comeon, buying a civ city by city? ridiculous

[nbarclay]

Re: Re: Understanding Firaxis's programmers.

Originally posted by BillChin

I agree and disagree. I believe Civ III is a relatively simple game especially when compared to chess. I am a novice player (only three months in). I taught my 14 year-old nephew how to play a half-decent game of Civ III in a day. He was struggling on Warlord difficulty and now can crush Regent difficulty consistently. There is little chance that a three month chess player can play a decent game, much less teach someone else. Even a great chess coach would have a hard time teaching a novice player how to play a half-decent game of chess in a day.

The big difference between "simple game" AI and "complex game" AI is in the ability to rely on "lookahead" to consider the outcomes of each possible sequence of moves. Tic-tac-toe is essentially a trivial case. Even an old Commodore Vic-20 (predecessor of the Commodore 64) could play out every possible sequence of moves and find moves it can make and be absolutely guaranteed not to lose.

Chess is a few levels up from that in complexity. There are twenty first moves white can make. For each of those, there are twenty first moves black can make, resulting in 400 possible positions after two moves. And so on. (Of course the number of possibilities in a given turn varies depending on board position.) The complexity of the game thus grows exponentially with how far you look ahead, and the simplest exhaustive lookahead-based techniques would require about four hundred times the computing power (give or take, depending on the position) to look just one move by each player farther ahead.

That requires a more complex level of AI programming. Any decent modern chess AI will almost inevitably have a book of standard openings that human chess players have developed over the centuries. It will also have ways of identifying which possibilities are worth looking farther ahead on and which are not. And it will have some way of figuring out what positions should be considered "better" based on human experience, since it can't play all the possibilities out to the end to see whether a position will result in a win or a loss if no one makes any mistakes.

Still, a chess AI based mostly on lookahead can be pretty decent if not downright good. The real key for an AI to beat most human players is not to let the human set a trap so far in advance that the AI won't see it in time to escape. If the AI can pretty reliably see the significant possibilities about four moves by each side into the future, most human players will have a hard time setting up an elaborate enough trap to catch it. And if the AI can avoid making a significant mistake before the human does, it can probably win.

In that respect, Civ 3 is as far beyond chess as chess is beyond tic-tac-toe. In Civ 3, you have to plan tens of turns into the future in spite of having vastly more things that can happen on each turn. That makes reliance on and implementation of lookahead-based techniques a whole lot more complicated, and dramatically increases the importance of scripted elements in the AI strategy.

Civ 3 might not look all that much more complex than chess from a human perspective, because a huge amount of filtering goes on at the subconscious level to zoom us in on the factors we consider important. But the kinds of filtering techniques that you can take for granted in a 14-year-old boy's mind when you teach him the game are next to impossible to get across to an AI.

Looking at the number of pieces and the size of the map frames the problem in a difficult way. I prefer to look at what the computer can do easily and build upon that.

I agree with the Rock, Scissors, Paper element of Civ III. Good point.

The Civ III AI is decent. It can beat 20% of casual players. Look at the many threads from players wanting basic help. I believe with a doubling or tripling of AI resources, Firaxis could get the AI up another 20% to beat about 40% of players. Unfortunately, marketing would likely veto such a budgeting decision. Many humans do not want a better AI. They find it frustrating and makes them want to toss the CD in the garbage.

No publisher is going to sign up for higher costs and fewer customers. I believe the AI problem is primarily a matter of marketing and budgeting. With enough resources, I am sure Firaxis could come up with a good AI. Good defined as an AI that can beat 80% of casual players on an even playing field. However, the overall budget might be 50% higher (maybe 10x or 20x more for AI) and the customer base 20% smaller.

I'll certainly agree that with more effort and money, Firaxis could have come up with a significantly better AI. But that is a management and budgeting issue, not an issue of programmer competence. What gets me upset is seeing people cast aspersions on the skill of Firaxis's programmers for the "crimes" of merely doing their best given the time and resources they have and of not being perfect or superhuman.

I don't think a better AI would actually hurt Civ 3's sales because (1) they could have the AI not use all its skills on lower levels and (2) it is hard to imagine people minding getting beat by a good AI on higher levels more than they mind getting beat by a bad AI that cheats every which way. But AI enhancements that drive up costs more than they drive up sales don't make economic sense.

It occurs to me that this may be an area where an open source Civ-type game could have huge long-term advantages. If AI researchers and students could be persuaded to take an interest, an AI for such a game might be able to attract interest far beyond what the simple economics of game sales would warrant, especially over the long term. Imagine the AI quality we could have for Civ 2 if its AI were open source and people had been working on improving it ever since the game came out.

Nathan

[nbarclay]

Re: Re: Re: Re: Understanding Firaxis's programmers.

Originally posted by BillChin
My favored approach is a scripting language that lets the fan base direct the AI. Have the game randomly choose from available scripts at the start of the game. Due to the Rock, Paper, Scissors nature of Civ III, a player will lose every now and again. Good scripts will be copied and improved upon. Bad scripts will be weeded out very quickly. This Darwinism will produce a good AI. Again, good defined as one that can defeat 80% of casual players on a level playing field.

As I understand it, the Call to Power AI has a number of files that can be modified to change various scripted elements of AI behavior, and some mod designers have used that to make improvements to the AI. I don't know whether Civ 3 has anything along those lines or not.

I like the basic idea of letting users modify scripted elements of the AI's behavior, but there is a lot more to a good AI than just following scripts. Scripting languages can't look ten or twenty turns ahead and try to figure out what the game situation is likely to look like if the AI makes one choice compared with if it makes another. Human players can. So if AI designers want an AI that can truly compete with good human players, they have to weigh what's best in terms of modifiability against what's best in terms of maximizing the AI's analytical capabilities.

Nathan

[nbarclay]

Originally posted by Spectator

Hey Nathan, I hope you dont mind I used your post from Civgame, I thought it was really good and that I could use it here to make my point.

Again thanks.

Spec.

Actually, I'm a bit flattered, and I like the thread it got going here.

Nathan

[Ethelred]

Originally posted by Coracle
Last night with that new (crappy) patch the stupid AI as the English started building MOW in a ten tile landlocked lake. My idiot city governor started building battleships in that same lake.

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.

[Zachriel]

Putting a warship on an inland body of water can change the local balance of power, such as in the story of the African Queen.

[BillChin]

A few comments:
Look ahead in the same way Chess programs look ahead is a bad way to go for Civ type games. Many people have cited the number of pieces and size of the map. I believe the keys to a good AI are the ability to adapt, to weed out the weak AIs, and to mimick the most successful human players. Adaptation and mimicking can be introduced by letting fans modify the scripts. Darwinism is a powerful element to weed out weak AIs. Look ahead is not really needed and a poor way to approach this particular problem. The human experience of what happens next is often good enough, if the humans can put it in the scripts.

For example, in my April game thread, (http://apolyton.net/forums/showthrea...threadid=46271 )
I outline several strategies with attack times and number of units. This is from experience. I believe crunching the numbers to look ahead that many moves in detail is a bad way to go. An AI that makes a rough estimate, such as 700 B. C. attack, have about 10 units to attack, 10 A. D. attack, good to have 20 units to attack, is easier to code. With a bit of randomness thrown it, it is probably more effective as well. Again these numbers can be changed in the scripts as human players adapt to the AI.

Again, having randomly selected scripts at the start keeps the human player guessing. Will the AI attack in 2000 B.C. with three units, 700 B. C. with ten units, or 400 B. C. with 15, attack in 500 A. D. with Knights, or is it a pure expansionist AI that looks to settle every spot of land (the current model)? Rock, paper, scissors, anyone? Having scripts for each similates a level of intelligence that the AI really does not have. A human faces uncertainty and has to plan for the possibility of each making for much more interesting game play.

As for my main point that a better AI translates into fewer customers, this is my conclusion from my discussions on the early MOO3 development boards. Several fans stomped their feet and threw a tantrum when I suggested such things as an adaptive AI, a scripted AI be put in MOO3. I am sure the publishers were paying close attention to these responses.

I believe that a better AI will translate into fewer customers. Other may differ on this point. However, everyone has seen threads where players get upset if the AI plays halfway intelligently. Everyone has seen threads from frustrated players that are crushing the game, but want it made easier still. Yes, the publisher can make the AI dumber or give it huge production disadvantages on the lower difficulty levels, but I think many people underestimate the frustration potential of a good AI and the potential for alienating many customers.

I've posted many times about patch 1.17f and fun factor. Games are supposed to be fun. No one wants to buy a game to be frustrated. 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

[nbarclay]

The key to making lookahead useful for anything beyond immediate tactics in a game like Civ 3 would be to base it on abstractions rather than on exact positions. Fuzz things together so that small differences in numbers and positions of units are treated as the same position. And, for strategic planning, don't worry about unit positions at all, focusing rather on "big picture" issues and quite possibly working with blocks of turns rather than individual turns. I think it could be done, although it wouldn't be easy.

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.)

Nathan

[ALPHA WOLF 64]

US Military analysis on Dec 6, 1941:
the japanese are not strong enough to attack the US.

US Military analysis on Dec 7, 1941:
OOOPS.

[Trevman]

Re: Re: Re: Understanding Firaxis's programmers.

Originally posted by WarpStorm


Ah, but could your 14 year old nephew beat one of the players who regularly plays on Emperor or Diety? Compare apples to apples. The Civ AI is fairly easily exploited at all difficulty levels due to the complexity of system.

No one in their right mind can conceive that Civ3 is more complex than chess. Sure there is more options and units, but in chess there is no way to overpower your opponent. You cannot simply gain a tech advantage. Even a pawn can kill a queen. To beat your opponent on an even playing field requires very well thought out strategy and tactics, not neccesarily "complex" though. What I mean is that you have to consider all possibilites and remember that any attack would immediatly destroy your piece. I would like to see any Civ3 player to win under these rules.

[nbarclay]

Re: Re: Re: Re: Understanding Firaxis's programmers.

Originally posted by Trevman

No one in their right mind can conceive that Civ3 is more complex than chess. Sure there is more options and units, but in chess there is no way to overpower your opponent. You cannot simply gain a tech advantage. Even a pawn can kill a queen. To beat your opponent on an even playing field requires very well thought out strategy and tactics, not neccesarily "complex" though. What I mean is that you have to consider all possibilites and remember that any attack would immediatly destroy your piece. I would like to see any Civ3 player to win under these rules.

Chess is difficult, but it really isn't all that complex. For most players, the difficulty of chess comes mainly from the extreme level of thoroughness required to avoid making mistakes or falling into traps. I'll grant that among better players, a good deal of complexity goes into designing traps covering so many options that the opponent won't see them in time to get out, but even that complexity is ultimately a product of thoroughly examining a huge number of possibilities within a relatively simple set of game mechanics.

In contrast, Civ 3 is more about strategy than about thoroughness. If your strategy is better than your opponent's, all else being equal, you can afford a few careless mistakes here and there because anything you lose (except a wonder in a razed city) can be recaptured or rebuilt.

Computers are good at thoroughness. Crunching lots of numbers in a fraction of a second is their stock in trade. But developing genuine strategies based on a particular situation at hand is another matter entirely.

Nathan