Livonya

Join Date 08/2006

Top 10 problems with AI

I originally started writing this as a post for another thread, but after finishing it I thought I would post it as a new post.

Hopefully this will be of some use.

Some background on my experience...

I have won many Gigantic games on Suicidal.

In a few of those games I personally researched/or stole every single tech. I didn't trade for any techs.

I don't do sneak attacks or wars that start and end in one turn (it simply isn't necessary).

Winning on Suicidal takes time because mico-management is what it takes. Every single turn I go to each and every planet and set the sliders depending on what is going on.

I have won with several different strategies, trying to see what works and what doesn't work.

The key is always micro-management.

In my current game of Dark Avatar I won the game (I still have to finish, but I can already tell I won) by ignoring most of the colony rush.

I stopped the colony rush phase at least 1 year before the AI stopped. While the AI was pumping out colony ships I had less than 10 planets producing ships and had my military slider set to 1% with those 10 planets focusing. I wasn't building war ships or colony ships. I had 8 trade routes up and running at least 1 year before any one of the 9 other nations had even 1 trade route established. I had an economic resource and a moral resource fully build as soon as possible.

I have fully developed planets. I put my social slider at 49% and my research slider at 50% and I developed my relatively small empire.

Anyway on to the point of my post....

Here is my list of the 10 stupidest things that the AI does in order of when they start making the mistakes.

1) They don't build explorer ships to find anomalies.

This doesn't need any explanation. Explorer ships can find anomalies, locate exceptional planets, and locate glactic resources (upgrade the explorer to a constructor).

2) They don't research engiine technology soon enough.

My slowest ships are ususally moving at 6 while the AI is building ships that move at 1 or 2. When I start war my ships are twice as fast as their ships. I would NEVER build a ship without an engine. What a waste of BC.

3) They don't build enough factories on their planets.

The smaller the planet the dumber the AI uses the planet. I am often invading planets with NO factories. The AI has NO idea what it is doing. The AI should completely fill a planet with low end factories BEFORE it starts building some massive building that will take 1 year or more to finish. Duh. Just because a planet is small does NOT change the fact that you will need at least 4 or 5 factories to get a planet full functional. The AI sometimes builds power plants BEFORE it builds factories. Dumb. Dumb. Dumb.

4) The AI continues the colonization period way, way, way, way too long.

Seriously, there are dozens and dozens of unused colony ships all over the place. I would NEVER send a colony ship to the other side of the map. What would be the point? When I get there the planet will have been claimed or if I do get a planet it will just get flipped or invaded if I can't get at least a few planets together in a group. I recently destroyed an enemy colony ship that was upgrading to something else. It was going to take over 100 turns. Duh. It spent money doing this?

5) The AI will send out colony ships with less than 200 people in them.

Colonizing planets with these small populations is just a waste of time. These planets flip instantly and even if they don't flip they take forever to develop into income producers. I have seen some with as few as 25 people.

(Combine point #5 with point #3 and they are just taking up space)

6) The AI does NOT understand the importance of unique buildings.

As far as I can tell the AI never makes any real effort to either make these buildings or to finish them.

Finishing these buildings is one of my main objectives early on. In my current game of DA I have managed to build every single unique building first. The Korx didn't even bother to start the Galactic Bizaar.

7) The AI doesn't specialize planets.

This is even more important in DL with the power planets and the research specializers. The AI builds too many starports and generalizes too much.

8) The AI doesn't seem to realize when a building is completely pointless.

The AI builds planitary defence buildings. The very first thing I do when I invade their planets is upgrading these worthless buildings into something useful...

Worthless buildings: Orbital Fleet Manager, Hyperion Fleet Manager, Secret Police Center...

9) The AI tries to defend each planet rather than defending each sector.

The AI defends each planet as if they are isolated from each other. If there are 5 planets in a sector it will have 1 ship on each planet with 1 or 2 speed. My small fleet can easily destroy them one at a time. If they had engines and left orbit they could be in a fleet which could then actually defend the sector.

A planet deep in their territory does NOT need a defender. 99% of my planets are undefended. I defend by having mobile fleets on the border of my empire.

10) The AI doesn't understand sacrifice, or long term planning.

For instance the most aggressive and militaristic nations don't seem to understand that the quickest way to military dominance isn't through military reasearch but through social build up. The most war interested AI races won't sacrifice early military dominance for a true military dominance.

What good is having the best weapons and largest ships if you don't have the social infastructure to build these ships?

You can also see this with regards to what buildings it will build. In DL the AI doesn't even bother to build old tech once it gets new tech.

You can actually destroy an enemy simply by giving them high-end tech. Give them the tech for the biggest factories and that alone with prevent them from competing. This should have changed with DL but I still see tons of enemy planets where they start by building the highest most expensive factory, or they will start by building the most expensive power planet.

Locked Post

Advanced Search

Watch this post

Do not email me updates for this post Email me updates for this post

Successfully updated karma reason!

Honest question: why is the Secret Police Center worthless? An extra 20% to base approval seems pretty good, what with not being diminished by population the way a normal morale building is.

I agree about the fleet manager being worthless though, for the AI's fighting style, if nothing else. It just means I can clear out all the worthless defenders in one go instead of having to hammer the planet over and over and over. And if it's protected by something nasty that would really benefit from it, I'm bringing in an overwhelming fleet to deal with it anyways...

Also:

You can also see this with regards to what buildings it will build. In DL the AI doesn't even bother to build old tech once it gets new tech.

I assume you mean DA?

Successfully updated karma reason!

I did a whole bunch of tests with the Secret Police Center and I don't think there is any difference between the 20% base approval that it advertises and just the normal 20% moral benefit that the 2nd tier moral building gives. Perhaps I am wrong, but in my experience the 30% and 40% from the 3rd and the 4th tier moral buildings is always better than the 20% the Secret Police Center gives... If I am wrong I would love to be corrected. I will test it again just to be sure.

And, yeah I mean to say DA and not DL. Oops.

Successfully updated karma reason!

I can agree on all those points, for a single player game the AI is lacking alot in terms of strategy.

Successfully updated karma reason!

The AI should be scripted to build at least a certain amount of facories on a planet whatever the circumstance.

eg

below13PQ=2 Factories
13-15PQ=3 Factories
16-19PQ=4 Factories
20+PQ=5 Factories

A certain amount of factories is never a bad thing regardless of what the planet is specialising in (The planets still need to build ships and the infrastructure in the first placed).

On the whole livonya I agree with most of your points although I guess your coming from a gigantic map non rare anomaly perspective for point 1.

Successfully updated karma reason!

I would add that one of the weakest point is Spore Ship by Korath AI and AI whiich is unable to defend against it.
In my last game as Torians while i was annihilated by Korath, i was able to destroy a lot of Spore Ships because they were all unescorted.Other AI were also too much liable to lose planets due to Spore Ships they simply didn't know how to defend from them.

Other points i have noticed is that some AI civs are really bad Korx particularly sucks.So some AI civs should get more love.

Successfully updated karma reason!

Hi!

I have won many Gigantic games on Suicidal.

I played only one gigantic game with masochistc AI. My impression after finishing the game was the AI was not at masochistic at all. Compared to other maso games the AI felt way to pasive. So I'll qoute EberKain to get to my point:

for a single player game the AI is lacking alot in terms of strategy.

Gigantic map involves too much strategy over tactics. So if you want real challenge from AI, play with game settings where there needs to be more tactical than strategical decisions. For certain that should not be a gigantic map. Even huge is to big. Start a large all max map with full number of strongest AIs, no tech trade, and you'll fight for survival from the turn #1. They will out-colonize, out-tech, out-build, out-cash you, and at the end of the second year they'll have from a sigle planet 200% more production than your whole 11-planet empire, like it happened to me Link: Balancing masochistic AI.

I'm not defending the AI. It needs some improvements. With time it also gets those improvements. But it was never meant to play in so high-strategic mode huge maps require.

BR, Iztok

Successfully updated karma reason!

The AI should be scripted to build at least a certain amount of facories on a planet whatever the circumstance.

eg

below13PQ=2 Factories
13-15PQ=3 Factories
16-19PQ=4 Factories
20+PQ=5 Factories

errrr wow those numbers are really low. I already see the computor building more then that on some planets all the time. PQ > 12, Factories = 6 + 1 of those quantom converters.

The computor i think does a good job at randomizing the planets on speciality. Lower PQ planets for research and cash high pq for manufacturoing.

The one thing i think needs the most work is the colonization. It is abit stupid to colonize a single planet in the middle of another persons base. I use this to my advantage all the time. There is always that nice pq4 planet next to your starting planet... i never colonize i let them and then i let it switch to me.

Successfully updated karma reason!

Anybody else notice that during ship battle, friendly ships will fire on each other and enemy ships will fire on themselves? At least that is what it looked like during the battle.

Successfully updated karma reason!

for a single player game the AI is lacking alot in terms of strategy.
Gigantic map involves too much strategy over tactics. So if you want real challenge from AI, play with game settings where there needs to be more tactical than strategical decisions. For certain that should not be a gigantic map. Even huge is to big. Start a large all max map with full number of strongest AIs, no tech trade, and you'll fight for survival from the turn #1.

.

In my opinion the smaller maps are even easier. Then you only need to develope 1 to 3 planets to win the game. I started on smaller maps and they were simply too easy, and most of the tech and buildings didn't even come into play.

90% of my complaints involve simple fixes, that would help on all maps.

For instance, the AI needs a better build order.

Every planet needs at least 2 to 5 factories or more, some planets should have as many as 10 factories, and then convert them to something else once you have enough of the final tier factories.

I still enjoy Galciv 2. And this post was not meant as general criticism. I just constantly see small things that the AI could do to give me a run for the money.

And clearly, the AI strategy should change depending on the map size, as my own strategy would.

Successfully updated karma reason!

Alstein

Join Date 07/2004

+55

I also think the AI handles peace incorrectly/getting allies. An AI in trouble should offer the moon to get someone else to jump in, before it's too late. Even if it means giving up a planet. They should offer planet for war as an option.

As for the small planets, I often don't build factories on them that much, I just use them as research collectors- spend up a research lab and let them go. My good planets are my econ/military planets

Successfully updated karma reason!

I'm glad somebody finally agrees that the AI isn't the cat meow that it could be. It certainly isn't a skilled human player.

Unfortunately I don't see real drastic improvements occurring in the near term but I could be wrong. What would be nice would be a multiplayer option, so you could at least struggle with some human companionship.

Play at a level beyond the AI. Maybe that'll eventually happen one day too.

Successfully updated karma reason!

I agree with 2 and 3, on the others I either disagree with what you want the AI to do or diagree that the AI's not already doing that. For example, the AI does specialize planets, at least in DA. I've captured several worlds that were clearly specced as research or manufacturing worlds.

Successfully updated karma reason!

Also, I got the Galactic Achievement of Galactic Privateer which makes your freighters immune to attack. The AI should know that if you have that Achievement don't even bother to try and attack the freighters. The Drengin had about 15 stacks of 6 corvettes and kept following my freighters around, never attacked my freighters but slowed my system down extremely and I am running a 3.6GIG processor, 2GIG RAM and 256Geforce and I mean it was 5 minutes between turns.

Successfully updated karma reason!

Hi!

In my opinion the smaller maps are even easier. Then you only need to develope 1 to 3 planets to win the game

O yeah!?! Even easier, when at the end of the first year Drengin from the next door come after your planets with 5 troop transports and 15 heavy fighters, each with 3 plasmas, and your beast weapons tech is lasers-3, and you have only 10 tiny hulls with single cheapest weapon to fake you're strong, an not even a hint of a planetary invasion, because you spent most your money for buying colonizers to overtake those AI colonizers (with 30% more space and 2 more speed you could possibly have) they were all sending at the turn 2 towards your space to colonize "your" planets, and you're still buying the tech for Trade centers to finaly get out if red income numbers?

That's even easier than growing for the whole year in a gigantic galaxy, before even meeting the first AI scout?

Well... OK. Other places, other habits. Strange sometimes IMO.

BR, Iztok

Successfully updated karma reason!

Voqar

Join Date 03/2003

+21

I would agree on a lot of the points, particularly plantet development (which if done better, would lead to an overall stronger opponent) - stuff like building research/manu/tech capitals on tiny worlds, or 2-3 of them on the same tiny world, and everything else mentioned.

For some of those items, you have to remember this is AI - it's not you, it's not a human. The AI has a huge amount of things to keep track of, and coding AI to sacrifice or do other things that are serious human judgement calls, isn't easy. The AI also has to be fast, so that the time between turns isn't ridiculous. It's surely hard to make the AI know when to patrol sectors with fleets, identify threats or real enemies, etc.

As a human, we can scan the game and make scores of decisions involving tons of ready information and we can combine that with all previous experience in all games and we can plan ahead, we can make mental notes, we can do comprehensive empire-wide planing - all in the blink of an eye. It's not so easy to code that.

As a human we can get a feel for things - I start a game, take a guess at how many worlds I'll be able to colonize, save my manufacturing capital til I get a good spot, maybe at some point decide it's now or never and stop saving it, build it, or decide I'll wait til I take a juicy planet via flip or conquest from a nearb foe and save it for then. It sometimes not so clear when to do things - you go by experience or gut feelings - something no AI will ever have, or for it to have it would require more processing that you'd want to sit thru every turn in a game that spans hundreds of turns.

The GalCiv2 AI is pretty damn good as far as single player game AI goes. There are definitely some things it could do better without crushing the game performance.

Successfully updated karma reason!

I agree, Vogar, I am not critizing but just trying to put points out there for the programmers if they are listening for the next patch if mine and other items are to be fixed. It is still a great game and I am not sorry I purchased it for $30, but if we can make it better, why not.

Successfully updated karma reason!

Frogboy

Join Date 03/2001

+2412

Good feedback.

Each time people post strategies and such, I'm able to take some of it and make the AI better and better.

Often times, it's not so much the AI is bad but rather that a certain % of players are simply better at the game than I am.

In multiplayer RTS's different strategies come and go and good players adapt. But whn playing against an AI, it can't adapt to new full game strategies.

For example, I had always felt that lots of research centers was more important than havintg lots of factories. But as I've played it, I've come to realize that that's really not the case normally.

Hence, why there will be AI updates and such for a long time to come. I love updating it and you guys help come up with new strategies.

Successfully updated karma reason!

Alstein

Join Date 07/2004

+55

There needs to be some randomness in the AI though, so that it isn't predictable, or as easy to counter...

The biggest problem to me though is diplomacy, the AI could make better use of it- even if it looks like ganging up, it's realistic.

Successfully updated karma reason!

I agree that the AI will never be human. But it can always be improved.

I spent a few more hours spreading my power through the galaxy in my current game. I still love the game, even if I have some complaints.

Here are some things I saw.

There are some things that are just sad and some that are weird.

I found an enemy planet that has two bonus tiles (the 700% research and the 100% research) but the enemy AI built factories on there and is using the planet as a ship producer.

I am just about done with the 3rd year. I have 15 spies. I have never had an enemy spy placed on any of my worlds.

I am currently crushing one enemy ai. I was looking at what the enemy was doing with it's economy. It appears to be spending twice as much resources on espionage than on military, and yet it isn't attacking me with spies.

Everyone always complains that they are plagued with enemy spies. I am currently at war with 4 nations (when I attacked one nation 2 others joined in so that was a step in the right direction), but none of them are using their spies on me.

Is it possible that since I have such a surplus of spies they are ignoring me?

I stopped using the spies as I was figuring that they would start pestering me soon, but it just isn't happening.

Anyway, I think the AI is pretty good, but there are some areas where it just doesn't perform.

It really needs to build factories on all worlds. It appears to either fill a planet with factories, or it builds just one and goes research. But it really needs factories to build up a powerful research planet. Even an economic planet needs some factories.

And Voqar has a good point. The AI often builds things like the Economic Capital, or Manufacturing Capital on tiny planets. I often save the Research Capital for a planet with some bonus tiles or a general bonus.

And building a power plant before a factory is a very bad move.

Anyway, I am going back to the salt mines. Now that the Mind Control Center stops enemy planets from flipping, and with the additions of tons of tiny planets, it is going to take a long, long, long time to invade every rinky dinky backwater planet.

- Livonya

PS: One last, thing. I tend to build lots, and lots and lots of factories. Every planet will have at least 4 or 5 minimum. I keep these at the 3rd tier factories. Once I have finished the planet then I will convert them to something else, or upgrade them to 5th tier factories. Building a top end moral building takes a LONG time if you don't have a lot of factories. So yes, this seems crucial to me. And I agree that this is just my style, and isn't what the AI should do necessarily, but factories are the building block of planets. Factories are essential for everything. Without them nothing gets done.

Successfully updated karma reason!

errrr wow those numbers are really low.

I did say at least... The AI in my games has a tendency to build zero factories on a planet or nothing but factories on a planet. It's very debatable how many factories should be built.

Using a split production method like the AI uses more factories are probably needed. Using a 100% focus method less factories are probably needed. But the fact is every planet needs a certain amount of factories built.

One problem with the AI is it doesn't reavaluate what a planet should be used for I've never seen the AI build over an existing building eg replace a factory with a research building.

Successfully updated karma reason!

One thing about building a power plant before a factory: it can be a good move IF you can't use lower-tech stuff. A Quantum Power Plant is only slightly more expensive than an Industrial Sector, but gives just as much production due to what the initial colony pumps out while having both a lower maintenance cost and (I believe) a lower production cost due to it being a bonus rather than raw IUs. So it looks to me like building the power plant first is possibly a symptom of the AI's refusal to use lower-tech stuff rather than a problem in and of itself: get the AI to start using Manufacturing Centers or Enhanced Factories instead of Industrial Sectors, and it might quit building a Quantum Power Plant first.

Successfully updated karma reason!

there is definitely a reason to colonize as much as you can. You have more of a say during the galactic meetings. having more planets is sometimes reason enough for other races to fear you and thus be nice to you. more planets equals more money too to fund whatever you need to win.

In DA, I once took over a planet and it had indeed (Jagged Knife) followed the idea of building mucho factories and a starport and that is why it was able to produce so many fighters.

Successfully updated karma reason!

Good feedback.

Each time people post strategies and such, I'm able to take some of it and make the AI better and better.

Often times, it's not so much the AI is bad but rather that a certain % of players are simply better at the game than I am.

In multiplayer RTS's different strategies come and go and good players adapt. But whn playing against an AI, it can't adapt to new full game strategies.

For example, I had always felt that lots of research centers was more important than havintg lots of factories. But as I've played it, I've come to realize that that's really not the case normally.

Hence, why there will be AI updates and such for a long time to come. I love updating it and you guys help come up with new strategies.

awesome

Successfully updated karma reason!

Often times, it's not so much the AI is bad but rather that a certain % of players are simply better at the game than I am.

In multiplayer RTS's different strategies come and go and good players adapt. But whn playing against an AI, it can't adapt to new full game strategies.

But some of these are really basic stuff. Like having a fleet of ships in fleets defending a sector rather than spread out among X number of planets. But I dunno, maybe thats too hard of a thing to code?

Also the AI doesnt seem to understand that you have Galactic Privateer, it still keeps chasing around your freighters and just tag along indefinetely. Ive had empires that I was at war with who had a large percentage of his fleet chasing around my freighters instead of attacking/defending. That may be a good tactic if I didnt have Galactic Privateer but if I have then it's just plain dumb.

Successfully updated karma reason!

I agree that planet development seems to be among the AI's weakest points. That and war, but I don't know if AI will ever really be good at war. I wonder if it wouldn't be reasonable for the AI to have a fairly rigid build strategy for its planets. Planet specialization may ultimately be the best strategy, but having each planet be as well rounded as possible works well too. A PQ 10 that has 3 factories, two labs, starport, 1 farm, 1 entertainment and 1 econ makes a lot of sense and doesn't take strategic planning. Another thing that I think would make sense and be easy to code is to make the AI build only the appropriate building type on bonus tiles, with the exception being food tiles which it can and should ignore. Manufacturing, research capitals and economic capitals are different and in that case it makes sense to build only that type of improvement, with the exception of a few factories that are needed to actually build the other improvements and a farm and entertainment on econ capitals.

Successfully updated karma reason!