Major AI flaw

Hello,

Often, I put 4 farms on each planet.

When I invade one, it has about 3 to 4 (I think).

Would you mind tell me how much farms you put on your planets?

Maybe I can do better!

Generally you want to keep putting farms on 1 by 1 (with a hospktal), until you hit the pop limit, then add another farm, or research a food tech, or similar. You want to.spend as minimum time pop capped as possible. 

You generally want to stop when your farm count reaches  (roughly) the same number of labs/factories/wealth buildings that you have.

You can think of it like being pylon blocked in Starcraft. You don't want to build too many pylons at the start because it's a waste of resources. At the same time, you don't want your unit production to stop because you don't have enough pylons!

In galciv, you don't want to build farms at the start. You should only build them when your population of a planet nears the cap. Then build 1 farm. Then wait for it to get close again, then build another  (or upgrade the first one). Etc.

I like the idea of keeping the same number of farms than the number of other labs/factories/wealth buildings.

So, if the AI keep building less farms, it will fall behind.

Yeah. I did a calculation a while ago - it turns out for a class 16 factory world, the ultimate manufacturing output comes with roughly 8 farms and 8 factories -> as a rule of thumb.

Actually,  you want to maximise and equalise your percentage gain in manufacturing and raw production - but that's a pain to calculate! So I use the rule of thumb above.

But again, a farm with no population being supported by it is useless. Hence why hospitals are so crucial - they increase your population growth substantially meaning you can get high pop faster.

Morale is also really key,  since low Morale hurts both raw production and growth - a double whammy on 2 crucial and related stats.

But the major issue is that the AI builds a few farms and that's it! It spends majority of the game pop capped which causes it to fall behind economically in a big way.

Teach the AI to build farms properly.

Make the AI great again!

So how does this work in practice? If I colonise a class 14 world and decide to make it a manufacturing world, I would

1) build a few factories.

2) build a hospital. 

3) build more factories until the world approaches pop cap. 

4) build first farm next to hospital

5) build more factories until all hexes used. Add solar collector then start upgrading.

6) every time the world gets near pop cap, either upgrade a farm, or destroy a factory and replace with a new farm.

7) when I hit 6 farms surrounding a hospital - and reach pop cap again, destroy the hospital and replace with a food distribution plant thing.

This should  leave me with 7 food buildings and 7 manufacturing buildings - and a lot of manufacturing to produce ships with!

adamb1011 explanations are mostly correct.

- use farms and manufacturin/research buildings about 50% to 50% of your space (I did a similar calculation a short while ago). If you use a lot of starbases or a manufacturing capital, use even more food buildings and maybe you will need an approval building or two.
- if you don't mind the micromanagement, deleting factories in favor of food buildings is also very correct. Even if you build a research world you should get some first and destroy later (or use the world as mixed world).
- I would consider getting the food distribution way earlier (If I have the tech the order is usually hospital - farm - farm - food distribution - more farms). It grants more pop then a basic farm when you hit 20 pop plus it gets growth adjacency bonusses and thus acts similar to a hospital.

What the AI does can be read in the GoverneorDefs.xml file.

<Governor>
    <InternalName>AIManufacturing</InternalName>
    <DisplayName>AIManufacturing_Name</DisplayName>
    <Description>AIManufacturing_Dec</Description>
    <Icon>GC3_Governor_Manufacturing_Spending_Icon.png</Icon>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>ManufacturingHub</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Food</Improvement>
    <Improvement>Food</Improvement>
    <Improvement>ManufacturingHub</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>ManufacturingHub</Improvement>
    <Improvement>Food</Improvement>
    <Improvement>Approval</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Manufacturing</Improvement>
    <Improvement>Influence</Improvement>
    <FillerImprovement>Manufacturing</FillerImprovement>
    <FillerImprovement>Manufacturing</FillerImprovement>
    <FillerImprovement>ManufacturingHub</FillerImprovement>
    <FillerImprovement>Food</FillerImprovement>
    <FillerImprovement>Approval</FillerImprovement>
    <FillerImprovement>Manufacturing</FillerImprovement>
    <ProjectName>ProjectBirthingSubsidies</ProjectName>
    <ProjectName>ProjectMilitary</ProjectName>
    <ProjectName>ProjectEconomicStimulus</ProjectName>
  </Governor>

So we have slots 7 and 8 (capital being slot 1) go to food and then slot 16 goes to a food improvement (next: 25, 31, 37, ...). Class 12 worlds will only have two farms when maxed out. This is waaaay too low and also lacks growth buildings. Manufacturinhubs are solar collectors, anti matter plants etc.

I am surprised to see, that manufacturing uniques (manufacturing capital, singularity plant,...) are not present in the AIs game plan. Has anyone conquered a world with one of these recently?

Only Power Plants and Antimatter buildings, but they are classed as Hubs.

I think there is internal c++ code that starts the initial planet layout, from looking at the xml's, Governors are kicking in after x amount of improvements or x% of planet tiles used (or a mix of both):

    <MinPerceptPlanetFilledGovAssignment>0.20</MinPerceptPlanetFilledGovAssignment>
    <MinNumBuildingCountGovAssignment>2</MinNumBuildingCountGovAssignment>

At least that is what I think these mean in the GalCiv3AIDefs.xml . But then it could be purely related to Player's assigning Governors(I never do so cant be sure).

So it is possible for the internal code to have the ability to build uniques. I never paid it that much attention before modding the Governors to build more stuff

AI lack of farms playing vs Godlike hasn't been my experience but with all the Raw Production the AI can really build things quickly at that level. As it seems to an issue and lower levels of play it certainly is one of many things the devs need to take a look at.

Well, at "higher" levels the ai cheats heavily.

Apparently genius is the highest level of actual AI - on incredible and godlike the AI just gets artificial boosts 

Cheats? What cheats?

Before you answer Handicaps are not cheats, every 4x is a Handicap game, it baffles me people call them cheats for this game, whats the difference between them in this game and the Sid Civ games? Nothing other than they are in a file HandicapInfos.xml's for Civ, but in GlobalDefs.xml for this game.

Ok here's my question then what would you rather play a more challenging game or would rather the ai get more bonuses. I called them cheats not because noone else did them but because i wanted a more challenging fair game with a better player.

Well, if he is refering to cheats at incredible and godlike, he probably means the free military tech chance, free credits chance and free ideology point chance. It's not much of a stretch to call those cheats. Then there is the FOW issue...

I can understand the argument of handicaps (+X% production for example) compensating for player abilities in fleet usage and planet layouts and therefore not exactly being cheats. These are used by all but the normal AI. AI behaviour does hardly change to my knowledge; colony scouting is a bit weird above normal.

Regardless of whether we call them cheats, handicaps or pigeon trainers, the less the AI has to use these across the board boosts, the better. This is essentially why you wrote your tweaked AI mod, right?

I just thought some things were missing from the AI that needed fixing. Better Blueprints, missing Improvements, weak Starbases, better Invasion chance, a more random Technology picking based on personality, are the main reasons I tweaked the AI.

I only play on Gifted or Genius(usualy to get my butt handed to me these days). TBH I have not looked that hard at the difficulty "boosts", but yeah free techs(or anything free for that matter) that is just plain cheating .

Have not really paid it that much mind on what or how my changes effect the handicaps, mainly because I am not that great at "micro management" to be able to test the higher difficulties

I actually  meant genius. I believe this is the last difficulty before the AI moves from getting standard +production bonuses (handicaps),  to outright cheats in the form of fog of war removal and extra starting  techs. Might be wrong about this though!

And yeah zuploed  - the method I stated was a bit of a simplification. Trying to write a repeatable method to perfectly micro a planet in under 10 steps is impossible! So I took a few shortcuts and simplifications.

Anyways I think we can all agree that an AI that manages it's planets better would require less "handicaps" and thus be a better opponent for the player to play against.

I sometimes play on incredible and godlike,  but I feel like the AI play a totally different game to what I am playing. They basically get a 30 turn headstart in tech. No fog of war. Etc. Yes I can win if I get lucky,  and micro well,  but it's not really fun.

And yes, I know the AI in civilisation cheats heavily. That's beside the point.

Galciv series has a (well deserved) reputation for the best AI in the genre. The bad news for the devs is that means players set the bar higher

That's not all of if by far. The AI has several flaws that cripple its effectiveness even at high difficulty levels:

1. Lack of farms: Limits raw production, granted.

2. Not enough economic starbases: While farms should take care of population and thus raw production, starbases are necessary to supply the manufacturing/research/economy multipliers for competitiveness. The AI just doesn't use these nearly enough

3. Insufficient engines on ships: The AI doesn't weigh ship speed nearly enough in its designs, which lets the players generally outmanoeuvre even a nominally stronger AI. Also, the AI (in my opinion) over garrisons its planets and doesn't put enough fleets into its attacks. if I actively deploy, say, 70-80% of my forces and the AI only 25%, I can defeat even a stronger opponent by attacking with multiple fleets and wearing down their defenders.

4. During warfare, the AI is hugely handicapped by the one attack per turn rule, while a player fleet can attack as long as it has movement points. I recommend changing movement rules for the player so any leftover movement points are forfeited after attacking.

The problem with 1 attack per turn is that it can lead to some extreme abuse with regards to 1 ship tiny fleets. 

I think the "closeness" or "size" of the battle should determine how many movement points it uses. 

Ie a giant fleet vs a single scout should use just 1 point. Whereby a massive fleet vs massive fleet battle should probably end the attackers turn outright if they win. 

I would rather have the AI check whether it wins or loses its battle and then wait to issue another order to all winning fleets. In recursion this would lead to multiple attacks per turn. Shouldn't be too hard to do (for the devs).

I have been playing on Godlike for a while, but just recently I ratcheted back to Incredible.  Either the AI has gotten generally better or it is better able to take advantage of its obnoxious level of bonuses and free stuff at Godlike.  Either way, I offer this as evidence that the AI is already under continuous improvement and development.  The farm-to-specialty ratio sounds like a productive area for progress, but it will be a subtle point to implement without making things worse.  Other AI work continues, the devs read this stuff and take notes, and we all will have to see what happens, I guess.

Just to climb on my soapbox, I have lowered respect for gamers who discuss handicaps as cheats.  It is lazy and deliberately provocative vocabulary.  People cheat, the AI just does what it s told.    Yeah, the bonuses for Godlike are so heavy as to be painfully apparent.  I'm really glad to be backing down to Incredible.   I'm not really a good enough player to be at top difficulties and don't really enjoy that environment.  But that is what you get if you insist on that level of challenge and cannot arrange multiplayer.  There is really no reason to use value judgements and perjoratives to discuss software behaviors.  Unless it's about the Drengin, of course.

It's more about whether the AI plays within the rules of the game. There's a big difference in terms of playability between an artifical production boost and turn 1 map wide visibility. "Cheating" is just a word to describe that.

And yeah- the devs are generally good at Reading posts and the AI has come a long long way over the past year. A year ago the AI just randomly built improvements, now it properly specialises planets and seems to attempt to make use of adjacency bonuses - which as you've noticed has made it a lot tougher to beat.

The purpose of this post was to provide the devs with a fairly easy "quick win" AI improvement so the AI can get better still

Wether we like it or not, Stardock has established it's operating model as selling a Ferrari that only goes 30 mph.  It looks great, and technically still counts as a functioning vehicle in all important aspects. It's fun to drive for a while,  but it's obviously flawed in some fairly important ways. Over years, the dealer will upgrade it so maybe someday it will drive its top speed.  So, it's a patience game for us buyers I guess.

Whatever we call it, what we all want is the same rules for everyone - human and Ai.  In the end that will make the game more challenging and fun because it will be balanced.  None of this the Ai can do stuff the human cannot and vice versa crap.

Also, if all the Ai flaws that are discussed all over these forums were addressed, there probably wouldn't be a need for Ai cheats/handicaps/buffs whatever you want to call them.  They could all be left in as optional tick boxes or sliders in game options screen so if people found stuff abusive, they could restrict it themselves.

Hfx, this is basically the new model  of game production. I'd get used to it.

Personally I like it. I'd go further and say that if stardock committed to having a full development team continuously improve the game,  I'd be willing to pay a monthly fee for it. 

Or I'd be happy to make some donations if they were to start selling ship parts aka dota2 style model.

One of the best wars I had in GC3 was way back (1.2 or so) against a godlike Yor AI which was swallowing up other empires. I can't really remember if it was them or me to stop ignoring the other, but when war broke out it looked like david vs. goliath with them having like 10x the zone of influence of me and 3+x the number of worlds. My civ were peaceful researchers for the most part of the game and at the start of the war I was really struggling, not being sure I could turn out enough ships in the long run to combat the tide of Yor ships. I eventually got the situation under control and started invading their worlds and not vice versa... the game ended earlier due to my research victory. (ok... enough of the memory lane)

So, I don't think the actual absolute balance of AI vs. player is what I want. But what I do want is the illusion of it. I want to be able to roleplay my tech civ to be highly advanced but vastly outnumbered. The current AI difficulties can't give me that. either I am not that much ahead in tech or the AI doesn't build hordes of ships that corner me despite my technological advantage.

Maybe we can put it like this: If you ahve to apply lots of handicap to an AI, then the AIs weakness is lacking human behaviour (suboptimal builds, low micromanagement, ...). If you don't need to apply a lot of handycaps to the AI, then it has game-inherent weaknesses, e.g. being easy to conquer when playing tech heavy. And It hink this is what it comes down to, when we (or I at least) say we want an AI that plays the same game. We don't want it not to cheat or not to be handicapped, but we want that it's playstyle has drawbacks, not that it is an AI.