The AI and its obsession for food

[alexman]

I just *love* this new debug mode in PtW! Last night I did a little experiment to try and figure out the AI behavior. I created a map with six identical islands, each island made exclusively from bonus grassland and with enough room for just one city.

Then I removed the ability to build any kind of ships and Wonders, turned on accelerated production, turned off civ-specific abilities, cut research costs by a factor of 10, and started all players from the Industrial Age in Democracy.

My goal was to play with the AI build-often preferences and see how they will react in identical environments.

I gave Egypt Happiness, Greece Science, Germany Production, Babylon Culture, and Russia Trade. It didn’t matter. Here’s what they all did:

Build units first (Warrior-Spearman-Spearman). Mine initially, until the city grows so that it needs entertainers (size 3). Then start irrigating to compensate for the food lost from the entertainer. Irrigate more and build a Granary so that the city grows faster. When the city finally grows again, assign another entertainer and start producing Wealth(!). Switch to rifleman when Nationalism is discovered.

Here is a picture of Thebes at this point. Note the lack of a temple, even though happiness is flagged as a preference, and there are obvious happiness problems.

[alexman]

Granted, in an actual game, the failure of the AI to build happiness improvements is less of a problem because in the expansion phase they build settlers and thus reduce population before it becomes a big problem. In the late game they have lots of luxuries so again unhappiness is not a big issue. Still, this experiment revealed the biggest shortcoming of the AI IMHO: AI cities should not try so hard to have a food surplus when they have happiness problems.

This brings me to suggestions for Soren to improve the AI in the next patch:

• When there are happiness problems, stop emphasizing growth and start emphasizing production. Forget the two-food surplus goal and use the increased production to build happiness improvements.
• Make the AI use the luxury slider! Why does the AI fail to make use of this extremely important tool? With entertainers you sacrifice one worked tile for one happy person. With the slider you sacrifice one trade for one happy person. Guess which is better…
• Never build Wealth when there are city improvements to be built

Until the above happens (read: never), here’s a suggestion for modders to improve the AI: Increase the value of specialists. Make an entertainer give two happy faces instead of one. Since the AI uses them instead of the luxury slider, which causes their cities to lose food, which in turn causes them to over-irrigate, he will benefit from having to use less entertainers to get the same happiness result.

Any other suggestions?

[Theseus]

Did you try it?

What about AI city governor settings?

[alexman]

I tried playing with the AI city governor for citizens (food, shields, trade, and any combo), and it made no difference for the above test. Always irrigating like crazy to compensate for the lack of food from assigning entertainers.

I have not tried doubling the value of entertainers to see what happens in the same test. I forgot yesterday, but I will try it tonight. When I proposed this tweak for the AU mod, people thought it would make entertainers too powerful, and they may be right, but now I think that would be a good thing, since the AI uses them so much.

[player1]

Doubled vlaue of entertainers?

Way to powerufull.

That gives you possibilty of totaly ignoring advanced happiness buildings and focus on luxury.

[FrustratedPoet]

Originally posted by player1
That gives you possibilty of totaly ignoring advanced happiness buildings

It also gives the AI a chance to do the same thing, and since the AI doesn't seem too keen to build them (when it should) then it might be worth experimenting with.

[Pyrodrew]

Remove the Growth Option from Rome, Egypt, Russia, America, China, India, & Iroquois. It's sure not helping them.

[Pyrodrew]

...

[alexman]

Originally posted by player1
That gives you possibilty of totaly ignoring advanced happiness buildings and focus on luxury.

As FP said, the AI already does that. If we can't change the AI strategy to become optimal, why not change the optimal strategy to match the AI?

Even if you don't need them, Happiness improvements are not useless because they generally generate lots of culture. Not to mention that you will still need them to get WLTKD in most cases. And don't forget, even with double entertainer value, you still sacrifice precious food for happiness. I'm sure there will be many cases where you want all the food you can get, so you will have to build cathedrals instead of assigning entertainers.

[Theseus]

I hate entertainers... unproductive b-stards.

[lockstep]

Originally posted by alexman
If we can't change the AI strategy to become optimal, why not change the optimal strategy to match the AI?

Because it might dumb down the game to a point where it's not fun anymore?

(BTW: Of the board members who claim that Civ3 isn't as good as Civ2 and/or SMAC, many seem to think that Firaxis has done exactly this: Dumbed down the game so that it matches the AI, but isn't fun.)

Even if you don't need them, Happiness improvements are not useless because they generally generate lots of culture.

I'd rather have buildings that I NEED to build than buildings that 'aren't useless'.

[alexman]

I don't see why this is dumbing down the game. Adding city micromanagement, yes. But dumbing down? I don't think so. It's just a different strategy required to do well than what we are used to.

OK, so you won't have to build catherdrals if you have many luxuries, AND you don't have culture-flip dangers, AND you don't care about WLTKD. But you *will* have to build them otherwise. Weakening Cathedrals will also serve to balance the Religious trait, which is too strong IMO.

And what about ADDING the option of using entertainers? Currently I am like Theseus. There is no choice. I hate to use them. It makes more sense to use the luxury slider and waste commerce in other cities rather than lose the food from assigning entertainers.

[Slax]

Maybe I don't completely understand the luxury slider. If you have unhappy people in one city, but many other cities that are happy, isn't the money spent on luxuries wasted making happy cities more happy just to fix one city?
edit -oops, answered in last post by alexman.

[Arrian]

Interesting study you did, Alexman. Like you said, the key here is an AI tweak, via a patch, to get it to use the luxury slider.

-Arrian

[alexman]

Originally posted by Slax
Maybe I don't completely understand the luxury slider. If you have unhappy people in one city, but many other cities that are happy, isn't the money spent on luxuries wasted making happy cities more happy just to fix one city?

Sometimes that is indeed the case. But the commerce lost to entertainment is always a fraction of the commerce produced by each city. So your larger cities that need entertainment usually produce more commerce, which in turn gets turned into more happy faces. With a 10% luxury value, for example, your size-2 cities will usually not get any entertainment.

[lockstep]

Originally posted by alexman
OK, so you won't have to build catherdrals if you have many luxuries, AND you don't have culture-flip dangers, AND you don't care about WLTKD. But you *will* have to build them otherwise.

But what if I won't have to build cathedrals if I have many luxuries, OR I don't have culture-flip dangers, OR I don't care about WLTKD? That would be dumbing down the game IMO. (At the moment, we don't know whether your or my assumption comes closer to the truth, so I appreciate your extensive testing. Still, entertainers that are twice as powerful as temples happiness-wise are a quite radical idea IMO.)

It makes more sense to use the luxury slider and waste commerce in other cities rather than lose the food from assigning entertainers.

I often use a combination of 20%-30% luxuries and one entertainer in border cities that don't have a temple yet.

[lockstep]

Originally posted by Arrian
Interesting study you did, Alexman. Like you said, the key here is an AI tweak, via a patch, to get it to use the luxury slider.

IIRC, some alpha screenshots of Civ3 showed a city screen without a luxury slider. This might well be a hint that there still isn't such a slider for the AI.

[nbarclay]

I think doubling the value of entertainers would add more strategic depth than it takes away. In the standard game, using entertainers is rarely worthwhile until a city gets about as big as you plan to let it get (at least for the time being). Certainly, it's rarely if ever a viable alternative to building happiness improvements or using the luxury slider when you truly want core cities to keep growing.

With doubled costs, entertainers would still hurt both growth and production compared with using city improvements or the luxury slider to boost happiness. The rules change would not even begin to create a situation where adding entertainers is always the right answer. But it would create more situations where using entertainers might be the right answer, and each such situation creates an additional strategic choice.

I do have one possible concern. AI cities already get ridiculously huge in the industrial era (at least if the terrain permits), and with double-value entertainers, they could get even bigger. Still, the prospects of AIs getting some kind of value out of most of the land in a city radius seem a lot better with double-value entertainers.

Oh, and there's one other interesting repercussion, for better or for worse. With entertainers more valuable, captured cities would be easier to control without starving them all the way down to (or at least near) size one. I'm not quite sure to what extent that's good and to what extent its bad, but I personally would probably enjoy the game more under such circumstances.

Nathan

[Fitz]

Flawed experiment. Since the AI only has one city at it's disposal, it must optimize it's income vs it's expense to a level that never exists in a real game. Given that every unit and every improvement costs at least one gold to maintain, and it cannot increase it's income except through growth of it's one city, is it suprising that it will not build excess units/improvements and will attempt growth (and "build" wealth)? Of course not.

Try this again, but give the AI room to expand up to it's optimal number of cities. Put each AI on a much larger island and disable boats so they cannot attack each other or settle other islands. Make sure there are a reasonable amount of rivers & specials for a decent income. You may still see problems, since the lack of contact with other civs means that they will never contact each other, and therefore will always have to research at the worst rate.

[Pyrodrew]

Given that every unit and every improvement costs at least one gold to maintain, and it cannot increase it's income except through growth of it's one city, is it suprising that it will not build excess units/improvements and will attempt growth (and "build" wealth)?

More tests are always good, but the AI still had no good reason to build that granary. That only increased their improvement costs & increased their # of unhappy citizens at a faster pace than necessary. Better structures/units could have been built instead. And it doesn't look like they are building wealth, rather a rifleman (more costs).

[Alex]

Interesting debate. Keep going.

[aaglo]

Originally posted by Theseus
I hate entertainers... unproductive b-stards.

HEEEEY!!!! Watch it!

[Nor Me]

I've already tried doubled value entertainers and it does not alter my style of play much. The A.I. would get as much out of them.

Looking at the example, if entatianers were doubled, Thebes would still need 2 of them. Only when it grows to size 7 would it make a difference.

A temple has the effect of allowing a city to work an extra square if it were 1 bigger whatever the value of entertainers. Only if up against a population limit of 6 or 12 can it be devalued

If Thebes were 1 square away from fresh water, A temple or rifleman would not help at the moment but would allow Thebes to work 5 squares if entertainers produced 2 entertainment.

Since entertainers are automatically content themselves, they would only really be 50% better.

Having said that they are much better at getting We Love The Haircut days than I first thought.

[alexman]

Originally posted by Fitz
Flawed experiment.

As I said, maybe it doesn't represent a typical game situation, but it does reveal some areas where the AI needs some help.

Given that every unit and every improvement costs at least one gold to maintain, and it cannot increase it's income except through growth of it's one city, is it suprising that it will not build excess units/improvements and will attempt growth (and "build" wealth)?

As Pyrodrew said, the does AI build units, which also cost support. As for having only one city, you might have less income, but you also have less expenses (you have only one of each building to support). Actually, because of corruption, your income/expense ratio is greater when you have one city!

...therefore will always have to research at the worst rate.

As I mentioned in the first post, I reduced the tech cost, so everyone was discovering new techs after 5-10 turns. I also tried reducing the optimal number of cities to 1, and got the same results.

[alexman]

Update: I tried doubling the happiness produced by entertainers and the AI seemed to do better for most cases. In some cases it still irrigated too much, but in others it actually mined instead. I think this is the best the AI can do without using the luxury slider. Here's a picture of Babylon from approximately the same time:

[Dominae]

I would really love to see an improvement in the AI, especially concerning the "Irrigation problem".

However, I'm not sure I like the double effect of Entertainers. Am I the only one that thinks this creates an entirely new game for the human players, which revolves around never building Cathedrals, Sistine Chapel, heck even Temples, instead focusing on the Luxury slider and population? Regardless of how the AI plays, one of the main aspects of Civ3 is having to become the builder eventually to deal with problems in your cities (unhappiness, corruption, etc.). With the proposed change, all you need is Workers to keep everything in order. Furthermore, I'm still not convinced that the Cultural value of happiness improvements is terribly important, even less so with the proposed change; Temples would only be built for their culture value, which doesn't seem worth it to me (might as well build Libraries instead).

Again, I hope the Civ3 AI improves sometime in the future, but not at the expense of the "stock" Civ3 feeling. The proposed change is a huge leap in the other direction, IMO. If only we could actually tell the stupid AI to build Mines more often...what a simple solution that would be!


Dominae

[alexman]

I don't think the double entertainer value changes strategy that much for the human. I really don't. At the same time as the AI reached the point of the screenshot, my single city was size 7, full of improvements, and much more productive - mostly because of my use of the luxury slider. I never used a single entertainer. As Nathan pointed out, the food lost from assigning entertainers is a steep price to pay.

As for telling the AI to mine more, I don't think it would solve the problem. The real cause of the problem is the fact that the AI doesn't use the luxury slider, so it has to assign entertainers, which reduce the food surplus, so it HAS to irrigate to grow.

[Dominae]

alexman, maybe you could try irrigating everything yourself in the test run. Your city would grow extremely quickly (especially with a Granary), and you could just use entertainers to keep happiness in check until you reach size 12. At that point you could "go back" and build all the improvements you like, slowly replace the irrigation with mines...basically transform your now huge city into a more productive one. This unnatural growth spurt is made entirely possible by making entertainers doubly powerful.


Dominae

[lockstep]

Originally posted by Dominae
I'm not sure I like the double effect of Entertainers. Am I the only one that thinks this creates an entirely new game for the human players, which revolves around never building Cathedrals, Sistine Chapel, heck even Temples, instead focusing on the Luxury slider and population?

No, you aren't. Making entertainers more powerful is very likely to make building happiness improvements an option instead of a necessity, and I wouldn't like that at all. Seems like a violation of the core game design to me.

[alexman]

Maybe I could have done better by irrigating. I didn't really try to optimize my growth. But one thing is for sure: In this test, I am ALWAYS better off using all my citizens as laborers and using the luxury slider for entertainment than I am by using entertainers.

Why? In this test, 1 entertainer gives 2 happy people and nothing else. 1 laborer gives you 2-3 food, 2-3 shields, and 2 commerce. Using the luxury slider, the 2 extra commerce from the laborer has exaclty the same effect as a double-value entertainer, but you get to keep the extra food and shields.

This is not the case in corrupt cities, and this is where the human strategy will change. Corrupt cities will become easier to manage if you double the entertainer value and irrigate everything, but this is what is already optimal in totally corrupt cities. What will change is the percentage of corruption where this strategy becomes optimal.