Why, oh why does the poor AI...

[alexman]

This happens most often in higher difficulty levels where the AI gets free units and doesn't need to build more.

But this tendency of the AI is exploitable even on lower difficulty levels. You attack a ancient UU civ during their expansion phase, making sure you trigger their Golden Age, prompty sue for peace, and you waste their Golden Age because they try to build settlers, not realizing that their boosted production completes them before their cities are large enough. (a.k.a. Sir Ralph's dirty trick)

[alexman]

Or for that matter, why even bother to improve all those tiles when they are not going to be worked for centuries?

[alexman]

Building a harbor here instead of a temple would give the city two happy faces instead of one, and each city in the rest of the empire one extra happy face.

[alexman]

This city has an entertainer and will have to assign another one as soon as it grows. Yet it is building a Granary and using floodplains instead of bonus grassland. Note that there are more cities in this position, so we could ask another question that Soren has already answered: Why, oh why doesn’t the poor AI use the luxury slider?

[alexman]

All screenshots above were taken from a normal tiny-map game in debug mode, with Rome having no build-often preferences and no traits.

Here are some more things I have observed in PtW, but for which I don't have screenshots right now:

• Why does the AI build a Coal Plant when a Hydro Plant can be built instead? And then why does it build a Hydro Plant as soon as it’s done with the Coal Plant?!
• Why does the AI build regular units in newly founded cities of a large empire when veteran defenders can easily be brought from elsewhere? The new city needs to start building city improvements, which cannot be brought from elsewhere.
• Why doesn't the AI switch production in its cities unless it's for a Wonder that has already been built? It should check and possibly reassign production when it learns a new technology, if not every turn. Extra bonus: could the AI be made to use pre-builds like us humans?

[DaveMcW]

The luxury slider is simple: buying luxury resources is more cost-effective than using the slider. Of course your harbor example shows the difficulties in acquiring luxury resources sometimes...

[nbarclay]

Originally posted by alexman
Extra bonus: could the AI be made to use pre-builds like us humans?

The catch with prebuilds, especially for wonders, is that if the AI does it, humans HAVE to do it to be competitive. That would make the game a whole lot less fun for players who aren't as dedicated to learning all the tricks as those of us at Apolyton are, or who don't like using prebuild trickery too much just as a matter of principle. Plus, I'd hate to even try to compete with a properly programmed AI in calculating perfect timing for a prebuild; the AI has the mathematical capabilities of a computer built in.

Edit: On further thought, that type of calculation is feasible only for the player that's driving the prerequisite research for a wonder without expecting to acquire some of it from someone else, and even then, there are some variables involved. So there are some aspects humans might have a significant advantage on too.

"Allow AI prebuilds" might make a worthwhile option for people to turn on if they want it, but at least for wonders, I don't think it would make good default behavior.

Nathan

[nbarclay]

Originally posted by DaveMcW
The luxury slider is simple: buying luxury resources is more cost-effective than using the slider. Of course your harbor example shows the difficulties in acquiring luxury resources sometimes...

If you can't get enough luxuries, though, the luxury slider can be a very powerful tool. The fact that the AI is missing that tool places it at a disadvantage.

Nathan

[Theseus]

I have a proposal:

First off, Soren, you need a vacation. Unbelievably great job on further AI tweaks (you guys are hilarious, btw: "Aw, we didn;t really touch hardly nothin' at all..." Yeah, sure.), but I'm willing to bet the Infogreed / Firaxis battles were a major pain.

Second, alexman has 1) as good an understanding of a multitude of game mechanics as anyone around here, 2) as much of a focus on AI performance and what actually going on as any fan I've seen, and 3) concrete ideas / observations / testing on remaining flaws.

Friggin' hire the guy as a consultant!! Hell, give him a per diem, sit him in front of a computer in lovely Maryland for 4-5 days, Mike B. buys the beer, Dan throws in a sombrero, and it's a done deal!

The marketing potential is HUGE...

(Of course, he already told you in this thread much of what needs to be worked on, but I'm sure there's more where that came from!)

In the alternative, a second proposal, especially 'cause I KNOW you tweaked stuff that you not telling us about...

Give us "Soren's Killer AI Civ PTW Mod" and we'll kiss the ground you walk on!

[THE CROWD GOES WILD!!!]

[Minmaster]

lol theseus

great observation alexman

[Harovan]

Originally posted by nbarclay
The catch with prebuilds, especially for wonders, is that if the AI does it, humans HAVE to do it to be competitive. That would make the game a whole lot less fun for players who aren't as dedicated to learning all the tricks as those of us at Apolyton are, or who don't like using prebuild trickery too much just as a matter of principle. Plus, I'd hate to even try to compete with a properly programmed AI in calculating perfect timing for a prebuild; the AI has the mathematical capabilities of a computer built in.

So what about adding such improved abilities one by one in higher difficulty levels, and dropping them in lower ones, instead of making solely bonuses all the difference between the levels?

Great observations as always, btw, alexman .

[DaveMcW]

By far the most powerful ability would be correct placement of the FP. If it was possible to code this it would result in +50% production or more for the AI.

They could even use a leader to rush it.

[alexman]

Originally posted by DaveMcW
By far the most powerful ability would be correct placement of the FP.

Yes, you are absolutely right. I forgot that one. If I could pick one improvement to the AI, it would be that one. Better FP placement would help the AI in the late game where it usually struggles.

Extra bonus: teach the AI to use a leader to relocate the Palace after the FP has been built, if the empire still has high corruption. This is arguably the most important function of leaders for the human player, and the AI is currently missing out.

PS. Theseus, thanks for the promo! Actually, I would give anything to be able to help improve Civ3 (especially if it involved coding).

[alexman]

For anyone interested, here is the build queue for the map I was using above, if you give the AI all techs up to the modern age:

Library
University
Colosseum
Temple
Cathedral
Marketplace
Bank
Stock Exchange
Factory
Coal Plant
Hydro Plant

Aqueduct built as needed when city reaches size 6
Harbor built as needed for extra food.
Hospital built as needed when city reaches size 12
Granary built when city has less than 2 food surplus.
Courthouses' position in the list depends on corruption.

The above list is for no half-price buildings and no build-often preferences. If you check production as build-often, you get this list:

Library
University
Colosseum
Temple
Cathedral
Factory
Coal Plant
Hydro Plant
Marketplace
Bank
Stock Exchange

Which is very encouraging, but still I think factories are higher on the list for most good human players.

If you mark trade as build-often you get something else which is encouraging in the screenshot below (because it connects the extra luxury), but it might not be optimal in some cases:

[badman]

Why is the colosseum so important?
It's usually one of my last buildings.

[Dominae]

Library-University seems like an odd decision to me. I would put Temple-Cathedral as the "default" improvements, since smallish cities benefit more from those than science improvements. Personally I never build Libraries in my cities unless it's a core, high-commerce city (unless I'm using the Library for culture, which I don't often do).

alexman, are those build queues for any given civ, or just for one? As your other examples hint at, it seems natural for a "Trade" civ to build Marketplaces before Libraries. To repeat my question, are you sure the queue you listed is default for all types of civs?


Dominae

[alexman]

I removed all civ traits from Rome, and all items from their build-often list. But note that the build queue that I listed was for starting the game with all techs up to modern age. In a real game the AI usually builds temples first, because that's one of the first technologies (with a useful build) that it gets.

As for Colosseums, the AI picks them before temples because they generate two happy people to the temple's one. Again, this doesn't happen in a normal game because civs usually don't know construction before ceremonial burrial.

[Dominae]

So, basically you've demonstrated the preference structure among all improvements for the AI. But my question still remains: how do the "build often" flags affect this structure? Does it work as expected?


Dominae

[alva]

IMO market -places should be much higher on the list!!

But maybe this has to do , with only one lux beeing present ?

[alexman]

Dominae, build preferences work to a certain extent. For production, I listed the effects above. If you flag happiness, the AI builds colosseums before libraries. If you flag wealth, it builds marketplaces before catherdrals. If you flag growth, it builds granaries somewhere in the top half of the list.

The bottom line is that the build preferences work as expected, but the game situations where the AI actually has a choice that would be affected by those preferences are relatively rare.

The best build queue depends on many parameters, so it's not possible to say that one building is always better than another. We should give Firaxis' testers some credit and accept that the AI default priorities work well as they are, most of the time. Still, I think all civs should have production, and perhaps trade for harbors and marketplaces, flagged. No civ should have growth flagged.

alva848, there were three luxuries present, two of which were connected very fast, and one was on that island. That makes one extra happy face for each marketplace. Not really worth it. Marketplaces are important, but when the science slider is at 50% or higher, and there are no happiness problems, libraries are a better deal (cheaper). Also, the AI doesn't overlap its cities too much, so it needs those border expansions from cultural buildings.

[Arrian]

Well, starting from scratch with all the techs and no civ traits is an abnormal case, as I'm sure you realized. If presented with that situation, I would probably build a factory FIRST, using the luxury slider to deal with happiness until it was done. The rest of the builds depend on a number of factors (such as # of luxury resouces - marketplace now or marketplace after temple/cathedral?), but I can say that the library/university/research lab would be... umm... dead last if built at all. What need have you of research if you have all the techs?

-Arrian

[Theseus]

I wonder... do the "build often" preferences influence research at all?

[alexman]

Theseus, I can see where you're going with this!
I will investigate, although it's a bit of a pain because debug mode does not directly give you that information.

Another thing I just discovered:
The AI does not consider the maintenance of a building in its decision of whether to build it or not!
I increased the maintenenace of libraries to 100 gold, and it still built them before anything else.

This might be a blessing in disguise for modders that want to strengthen the AI, because they can bring the human optimal strategy closer to the AI strategy by tweaking building maintenance costs. For example, I think that a very good change to make would be to increase the maintenance of cathedrals to 3 gold. (Temple=1 happy face, 1 maintenance, Colosseum=2hf, 2m, Cathedral=3hf, 3m). This would weaken Cathedrals (and, btw the strong religious trait), in favor of Colosseums, so the human and AI strategies would no longer be that different.

[Edit: Arrian, good point, but the AI still had the modern age techs to research]

[alexman]

Edit: Deleted some information about build preferences that I am not so sure about any more. Still testing...

[Theseus]

Wow...

1) We have to re-assess any mods.

2) Soren should help.

3) Firaxis, can I say "PER DIEM" again?

[Zachriel]

The "AI" is actually an "SI," that is, Simulated Intelligence. You're not supposed to look at an AI town close enough to see that it is wasting shields. That's like looking behind the town store in a Western movie and seeing the braces holding up the fake building. You know they're there, but you pretend you don't.

Basically, it just doesn't matter how efficient AI towns are; everything is adjusted by selecting the difficulty level. If the AI is 50% efficient, then give it a 50% boost. This is much simpler than actually trying to produce a truly intelligent AI -- and it works just as well in many respects. This is one of the compromises that makes Civ possible on a "primitive" 20th century computing machine.

And not that many previous suggestions in this thread wouldn't be good ideas and in furtherance of the phantasmagoria. Just remember, the goal of the gamemaker is not to actually design intelligence, but simply to create a successful illusion.

Simulating Intelligent combat is another thing. . . .

[alexman]

Zachriel, wise comments as usual.

BUT, there is a small problem with the illusion that Civ3 SI presents to us right now. The AI bonuses give the human a good (often too good) challenge early-on in the game, but are not hardly enough in the late game. Our goal here is to make the AI behave more like the human at the city-level so that it can use its bonuses more effectively in the latter part of the game.

[Theseus]

alexman, my turn:

You just NAILED it... my quest for killer AI civs, the AU Mod, and even Vel's thoughts about extending the (don;t remember his words) period of 'balance' are all about getting away from the benefits given to the AI civs via difficulty level, and getting them to perform better AS THE GAME GOES ON.

Zachriel, your comments are 'correct' but in a sense not relevant (nooooo insult intended )... I just want to see a better challenge, whether or not we acknowledge that it's SI versus AI. Do we care that the chess programs are not true AIs? Yes and no... If the AU Mod can pass the equivalent of a Turing Test with the bunch of us, that's good enough for me.

[Dominae]

I think alexman's distinction is extremely important. Civ3 is only challenging in the early game (and sometimes into the midgame, admittedly) because of the bonuses the SI gets. Once you realise this, the endgame becomes quite boring IMO. Ideally I'm looking for an AI/SI that doesn't need the bonuses it gets at the beginning of the game but is challenging due to algorithms that play the game smartly. These algorithms don't need to be cutting-edge AI, but should at least incorporate some of what alexman and other players know to be strategically sound.


Dominae

[Pyrodrew]

Very informative.

Library
University
Colosseum
Temple
Cathedral
Factory
Coal Plant
Hydro Plant
Marketplace
Bank
Stock Exchange

Eek!

As for Colosseums, the AI picks them before temples because they generate two happy people to the temple's one.
I think that a very good change to make would be to increase the maintenance of cathedrals to 3 gold. (Temple=1 happy face, 1 maintenance, Colosseum=2hf, 2m, Cathedral=3hf, 3m). This would weaken Cathedrals (and, btw the strong religious trait), in favor of Colosseums, so the human and AI strategies would no longer be that different.

Increasing Cathedral maintenance to 3 will still make the AI build the Colosseum too soon. IMO, making the Temple a requirement before you can build the Colosseum would help the AI more. Then logically & similar to human strategy, the AI would build a Cathedral, once available, before the Colosseum as well.

all civs should have production, and perhaps trade for harbors and marketplaces, flagged. No civ should have growth flagged.

1.I strongly agree all civs should have production flagged (if not then factories should be weakened). Factories + Cavalry = dead AI.

2.I agree most AI Civs should have Trade flagged. Not only because of the marketplace, but because the AI is horrible at setting up its road network with other Civs, harbors are crucial for them. In my current unmodded game several Civs have luxuries, but because they refused to build harbors they could not trade with other Civs giving me a huge advantage. Harbors offer trade, veteran ships, & more food at a cheap cost too good to pass up.

3.I agree most AI Civs should not have Growth flagged.

The only exception to points 2 & 3 might be for a little AI variety so AI Civs do not all have the exact same things flagged/unflagged.

I would probably build a factory FIRST

This is why it is important to keep the rule that no more than 4 aspects are flagged for the AI Civs... the more that are checked potentially drops the priority of Factories.