I will buy an expansion if the AI is improved: IDEAS...

AI improvement suggestions

I posted this on another thread, but wanted to see if I could get another thread focusing on AI enhancement suggestions, post 1.2 beta.

First, let me say I do AI work for a living, so I have VERY high respect, tolerance, and understanding for you who do this. It is clearly the most difficult to implement part of any game, and probably always will be until someone invents the positronic brain.

Just to iterate what I see as necessary core improvements to the AI:

1. Continue to improve the AI's ability to manage improvements on planets:
a. Build associated buildings on proper resources. I see things like "planetary defense" on a "prairie" quite often.
b. Quit building farms "everytime" a planet hits the pop cap, and then some. Use some algorithm like "build farms until the total reaches 10% of the available squares" or something.
c. 30% of the squares having planetary defenses makes no sense. Screw planetary defenses. They don't help much anyway as they thus far have never prevented me from easily taking the AI's planets, it only costs me a few more solders. This would be MUCH better spent on manufacturing and building ships to defend planets with.
d. Where is the manufacturing? In the game I am playing (and most previous games recently), the AI builds WAY too few manufacturing centers.
e. Put a big-time cap on influence buildings letting other important structures lag behind. Influence is quite moot if you can't build your navy fast enough to thwart my attack.
e1. same with any other non-econ, tech, or manuf buildings like moral. You simply need to live with a pop cap, influence cap, and moral cap, then deal with it as you can without sacrificing econ, tech, or manuf.
f. Intelligently manage econ vs. tech vs. manufacturing planets. Too often I see mixes of them on the same planet, which creates major inefficiencies, or at least losses of compounding buildings.
g. Focus WAY more on building infrastructure. I see the AI focusing way to hard on building a navy early, clearly costing money and time to build and maintain.

2. Build killer fleets. I see small, and clearly ineffective fleets by the AI, even when it would "appear" they have maxed out their tech tree, making them easy to defeat. I don't field any fleets that are not as powerful as possible if I intend it to be my main attack force (I do small fast ships for transport interception). Quit fielding many small fleets as they don't even dent my large fleets.

3. Build the best ship possible at all times. The AI builds wimpy ships.

4. Quit over-building transports. The AI builds them willy-nilly, and sends them apparently randomly all over the place. Only build them when you have specific goals/planets you choose to take. Pick a planet that you can attack with fleets, then soon after hit with transports that only leave a "safe area" or a planet when said target planet is ready to take (ie no ships in orbit, or nearby). In late game, put a gazillion engines on transports. Wasting population as I pick off transports for target practice really hurts the AI.

6. Only build and send transports when it is safe. As I knock off the AI's navy, it still keeps building transports and escorts. WHY??? It should be purely focusing on getting strong fleets together to thwart the now stronger opponent before resuming the building of said ships.

5. Guard transports with killer fleets. Screw escorts completely. They are basically useless.

6. Quit building 10 gazillion unused freighters running them all over the universe in random patterns. This is annoying, and clearly inefficient. It is ok to have a few backups, but zillions is ridiculous.

7. Have the AI focused on goals like taking the closest planets one-at-a-time, guarding its space with killer fleets, and focusing on taking out its opponents killer fleets.

8. The AI should be able to switch attack methods once it sees that I have switched countering its defense strategy, and the same with my defense strategy. The AI may actually do this, but once I have obliterated any large fleets, it seems to just build transports and escorts, so I wouldn't see it.

10. The bottom line is, on the highest difficulty level, large galaxy, when the AI is 4x ahead of me in all stats catagories, it should plain and simple, kick my ass every time "in the end game" regardless of any minor cheating I do in the early game to keep up. Just for reference, on my last suididal game that I won, I gave myself 10000 bc, made earth and mars 26 planets, and restarted until I had 1 800% manuf resource on earth. This really isn't much considering I was only able to colonize a total of 8 worlds before the rest were already populated by the AI. I was WAY behind until I was able to build a 125 attack, 27 defense (of whatever the most powerful AI opponent attack method is), and 43 move ship, stack 5 of these together, build about 5 fleets of these, then at that point, I have never seen any game where the AI can over-come this. If the AI happens to still be utra-powerful, like in one game, one AI opponenet had conqued everything else except my 8 planets, they put up an ok fight against me, but in this case, I focused on bringing the most experienced ships into a fleet, and continued to build that fleet's experience until its hit points were so high, it was unstoppable, thus giving me that final guarantee I needed.

11. Now to wrap it up, since the AI planets are poorly built, it just can't build the ships to keep up. Even though the AI has 50 planets to my 8, I still can out-build it, and over-power it with my fleets.
10,180 views 19 replies
Reply #2 Top
Planetary Defense on a prairie is not a problem. Prairies themselves carry no bonuses.

And yeah... if you are cheating in the game, of course the AI will be simple to beat- you're cheating.
Reply #3 Top
c. 30% of the squares having planetary defenses makes no sense. Screw planetary defenses. They don't help much anyway as they thus far have never prevented me from easily taking the AI's planets, it only costs me a few more solders. This would be MUCH better spent on manufacturing and building ships to defend planets with.


I agree with this. While the 20% bonus is good, most times I will simple send an extra transport along so I dont really see the value when compared to other uses for the square.

3. Build the best ship possible at all times. The AI builds wimpy ships.


I think this is being addressed in that the AI only redesigns once a year but under new options may be able to do it more often?

Reply #4 Top
Yeah planetary defense is useless.

It should make a greater difference in my opinion. 50% would be correct in my opinion.

So when a planet have a planetary defense, you have to build lot more transport to invade, so it let the time to the player or AI to build starship to defend the planet before it get invaded.
Reply #5 Top
The devs rarely stop by Game Talk. The place for this thread would be in the main GalCivII forum.
Reply #6 Top
Thanks for opening a thread about the AI, Ivan, it's in the focus of my interest too.

Now I'm not sure if you played with 1.2-Beta already, but in my experience with it most of your points under 1. (planetary management) are no longer an issue: The AI builds a lot of factories, uses special tiles better and has stopped building a gazillion influence buildings, farms or morale boosters. It's not perfect yet, but much better than before.

Regarding the planetary defense building: I share your concerns, because I'm not sure if it actually does anything!
I did a test with 1.1, gifting a planet to the AI, one time with 13 planetary defense buildings (yes, thirteen) and one time without them - conquering it was roughly the same in both cases (except for slight differences in the random roll, of course). The numbers representing the soldiering skill were the same, and I took the same losses approximately. It might be a coincidence, or it might have been changed already, but my guess is that they don't work as they should.
Please have a look at it, in case a dev reads this.

Regarding the problems with AI shipbuilding, grouping of fleets, lonely and too many transports or lack of strategic focus, I agree with most of what you say. I'm really looking forward to the option (I hope it's still planned) to give the AI more time (a) to design ships more often and (b) to do more tactical evaluations.
I realize that these are difficult things to implement, and that the GalCiv AI has gone a long way already (it's the best TBS AI that I know in any case), but enhancing it more could make it truly outstanding and a benchmark for all comparable games.
As of now, I can outmanoeuvre and destroy it rather easily on the tactical level on all difficulty levels, I think there's room for feasible improvements.
Reply #7 Top
I never said the AI was "bad". The AI is actually pretty good compared to many games. I am making suggestions for improvement.

And to answer the question of "cheating", can you beat the game on suicidal with out cheating in some way? It is possible, but likely extremely impossible. Since the AI can send out a gazillion colony ships before you can even produce 1, it gets an early huge leg up on you in terms of numbers of colonies, and thus will likely be able to over-power you early on. I "cheated" to assure that I can get past this early rush that is gained by the AI "cheating" (stated so in the manual) giving it huge economic and military boosts.

This really is all about the AI itself, not how much money you start wirh, or what bonuses you start with on your home world.
Reply #8 Top
Try 1.02.009 Beta and give us a ring. Planet management and aggressiveness has been improved.

Also waiting for the intensive AI give it more CPU time option.

I see the AI getting better every time a new Beta has been released.

One thing though, I see more and more the AI's beating up on each other. One reason you may not see the full wrath of the AI is that they consider your empire (you the real flesh and blood player) as just another race and may be concentrating on some other AI player instead of you. The player is not singled out for special attention, all races are evaluated the same.

Frogboy has stated that the real problem is CPU time to have the AI evaluate things. Some things we see intuatively are just hard to quantify.
Reply #9 Top
I love this game, and I agree it's got the best AI on a computer game so far, or at least one of the best... but still, there're are some things I just can't understand.

Today I decided to play the campaign, and I have to say I'm quite disappointed with Ai performance. I understand the AI should be able to play a scenario a lot better than a random game. In the scenario there are clear goals, and the AI can be programmed to follow a scenario specific strategy.

I just played the three first missions, wich i understand have to be easy, but it's not easyness what I'm cpmplaining about, it's just that the AI doesn't seem to realize it's playing a scenario. They just play like in a normal game.

The first one for instance, it's just a tiny map with total of 7 planets. One Altarian, One Drengin, Two terran, and two on the middle for the first one who grabs them. My goal is to kill the Drengin, and theirs is of course to kill me. I think it's quite obvious that you don't need to research the whole tech tree and get to build stock markets to win that game. All you need is a few lasers and planet invasion technologies, after that you can put research to 0% if you want.

What do you think the Drengin did while I was doing this? Well, they built two wonders and researched soil improvement (or whatever the first terraforming technology is called)?!? I guess they thought they needed more room to keep building wonders...
Reply #10 Top
In respect to the fleet building, 1.02.009 beta is a WORLDS difference with the not waiting each year (52 turns) to upgrade/redesign ships.

Before 1.2.009, on my games (Huge, 9 AIs), I would see M5 and M6 (Mark 5 and 6, which indicate generation) as the max, when I finally reached top-end of armor (zero point, aeron, ultimate invuln) and you are correct, one or two ships could kill entire fleets.

Now, with the intensive AI on, and the latest patch, I've not even reached top-end on missile att/def and very near to just one of the other two, the computer is on M29/M30 versions of their ships. And they're maxing out logistics technology to stack them. I just ran into a 9-ship fleet with 458 combined attack and 150-ish defense in Mass drivers, and 30 defense in missile/beam, though no attack yet.

I have two of the maps three military resources, maxed out on mining, and my starbases fully loaded (110+ defense for each category, with 56+ attack on each) and they will still die to one of their fleets. I've had to put my own fleets on top of starbases to protect them now.

So, in other words, if you have not turned on intensive CPU, DO SO. Explore over to ..\My Documents\My Games\GalCiv2\prefs.ini and change it! "AllowCPUIntensiveAlgorithms=1"!!

-thefinkster
Reply #11 Top
Maybe in that case planetary defenses should be improved- especially if they are so weak.

Perhaps make them add +100% more to defense, or give each defense a 10% shot of taking out a transport before the fight, or something.

Maybe also allow them to aid defense for ships on the planet?

Reply #12 Top
Update on ship build AI:

I just started my second game with the Intensive AI turned on, with 1.2.009. This is the FIRST and ONLY game so far, where I have seen the enemy ships with more armor than attack.

Yes, you heard that right, the fleet I just came across, with two Heavy Fighter M8's, have 30 mass driver attack, 60 mass driver defense! Double the armor versus attack.

As far as I can tell, the AI is working beautifully now. My only complaint was the ship/fleet AI to be tougher, and they have fixed that in spades.

About the only problem (and it's a VERY minor one) I have now, is AI fleets/ships mass suiciding into my starbases. I mean, a ship with 12 attack and 6 defense will take on a starbase with full hit points and more defense than it has attack/defense combined. And that's IMMEDIATELY after their fleet failed to do the job. I mean, what's up with that? Some learning inside a game would be nice. (JK!)

-thefinkster
Reply #13 Top
i noticed this to, but i after 1.2, and setting the difficulty higher i've noticed it doesn't happen anymore, at all, the AI builds the right buildings on the bonus tiles
Reply #14 Top
I was beat by 1.2b last night, which was a great thing. I have never been happier about getting whooped on before. However, it was fairly early game. The Torians declared war on weak little me without asking for cash first. This probably makes it more entertaining, but again, I hadn't had enough time to build up to avoid the "suicidal diff AI mad early rush". We will see what happens when the AI lets me live a while longer.

Yes, sometimes the AI opps beat on each other, and that weakens them, however I have had several games where one AI opp (lets say pre 1.2b) is way dominate, and my statements still apply.

I will be working to counter this new AI, but it appears that I have a new challenge. Already I have to say that the AI is doing much more intelligent things with this new version. I am now confident that it will only get better.


I vote to turn on this "intensive AI CPU" option. I won't mind waiting longer between turns if it means that the AI is better. Options are good since some people don't like a challenge.
Reply #15 Top
I think we ultimately have to accept a gun v. butter philosophy with the AI. Part of the reason that other Civ type games might be a little more "tight" is because the A.I. has fewer variables to deal with. We all (myself included) wanted ship design to be a part of this game. This opens up a new can of worms for the AI as it has to try and figure out how to optimally design a starship to deal with its adversaries. In games where all of the units are predetermined, the AI can spend more of it's time executing grand strategy and using its units more effectively. I remember the days of galactic conquest games where the time inbetween turns would often exceed 1 or 2 minutes while the computer was trying to figure everything out. Ultimately, we are limited by the speed of our processors.

This doesn't mean the A.I. can't be improved -- many players, like Ivan, have made great suggestions that I believe can be implemented without significantly slowing the game down. Perhaps another alternative is to increase the frequency and potency of "random events". In the MOO series, the Antarens would occaisionally send poweful fleets of starships from another dimension to harass you (in Star Trek: Birth of the Federation the Bog would occasionally pop out of nowhere are start attacking). Perhaps an occasional asteroid will collide with one of your planets destroying many of its improvements and reducing it's quality. Maybe a mysterious and evil faction will rise up on one of your worlds and cause it to defect from your empire (as sometimes happened in the first GalCiv). All great civilizations have to cope with the periodic hurricane or tsunami -- the true test of a great leader is how he/she handles these types of random catastrophes. One way a expert player could "turn up the difficulty" is to select an option in the game which increases these types of random challenges. This might take a little pressure off of the A.I. as it currently bares the full weight of making this a challenging game.
Reply #16 Top
So, in other words, if you have not turned on intensive CPU, DO SO. Explore over to ..\My Documents\My Games\GalCiv2\prefs.ini and change it! "AllowCPUIntensiveAlgorithms=1"!!


Either there's something wrong with my installation (got also 1.2.009b) or it doesn't work yet. I turned it on, but in game it didn't change anything. I went back out and saw that the AllowCPUIntensiveAlgorithms variable was automatically set to 0 again. Just in case you wonder: yes, I saved & tried it a few times to avoid an error on my side.

If I'm correct then your observations are based on the new version only, but not on the new cpu time option.

Reply #17 Top
Either there's something wrong with my installation (got also 1.2.009b) or it doesn't work yet. I turned it on, but in game it didn't change anything. I went back out and saw that the AllowCPUIntensiveAlgorithms variable was automatically set to 0 again. Just in case you wonder: yes, I saved & tried it a few times to avoid an error on my side.

If I'm correct then your observations are based on the new version only, but not on the new cpu time option.


Actually, I noticed this only after updating to another version. I just checked, and mine is still set to =1. Two games so far, and this second one, I am going to lose, unless I can churn out some constructors.

Hm.

I saw frogboy post that this was in 1.11, and then searched to find that it was in prefs.ini.

Can we get a clarification... did the allow intensive cpu allow for more ship upgrades for the AI, or did 1.02.008/9? That was probably the most marked change I noticed. Generation 30 ships!! (For those that don't know, when you see a ship named Drengin Heavy Fighter M5-152, the M5 indicates which generation this fighter is in.)

-thefinkster
Reply #18 Top
I try it again and set the file on writeprotect... hope that doesnt screw up the game
I also did no intensive testing, just one game - near the end of the 2nd year there were still M1 class escorts around. Maybe there really is something wrong with my setup. I think I better do a reinstall.

Edit: It is true the layout for the function was implemented in V1.11, but back then it wasn't put to use yet / there were no additional algorithms available. Frogboy said that in V1.2 it would be 'filled' up..
Reply #19 Top
And to answer the question of "cheating", can you beat the game on suicidal with out cheating in some way? It is possible, but likely extremely impossible


It is not easy to beat the AI on Suicidal, but it is certainly something you can do as long as you don't start in an untenable position. Try it without buffing up your initial quota of goodies, it will give you more satisfacion.

Intelligent trading will definitely help, but forget about weapon techs as the AI are really really obtuse about trading them. The largest ai's always always head up the missile tree, so as soon as you are able pd right to aeron defence and the 20% reaserch bonus trade good: hyper computers.

The initial rush is definitely not something you need to worry about. The key is to do a decent spread of the galaxy don't worry about leaving poor planets (PQ8 and less) in your neighbourhood. You'll soon aquire them if you're clever with your influence. Restaurant of eternity is a must if this happens.

Money: As long as you do not overspend initially you should be able to keep your spend slider on 100%. There are times when you seem to loose 100+ a turn in the early phase, but I usually tend to sell tech to keep myself there. Even with the slow tech option it is frightening how quickly the tech list is chewed up. the drengin seem to be great for the financial techs and the ship upgrades. Torians are good with miniaturization and logistics. Eartlings are good for weapons initially, but the drath always seem to get ahead of them somehow. for some strange reason the thalans were phenomenal with weapons in my last game, but usually they are mediocre. As for the rest - i've noticed no real trends but you do get an even spread of techs off of them

Planets. High pop worlds would be nice but if you are desperate for cash and don't want too many farms, high tax rate will see you through. Specialist worlds shoud be a priority, I usually have two ship building worlds all factories and one entertainment, one of which holds the manufacturing capital. One research world maybe two if i'm lucky. All the rest are finance worlds. To get the most out of your research and production worlds: ring them with economic starbases, the bonuses are cumulative and you can have 4 starbases per sector.

Ships: I do not build anything except constructors and scouts until planetary invasion has been researched. Then its small and medium ships with armaments but no defences until dreadnaughts hit the scene. I use to build a lot of military starbases, but they're a waste of time for anything but the dreadlords.

Trade goods - Diplomatic translators, Aphrodisiac, hyper computers. Tech them, build them and never let ayone even get a sniff of them.

Wonders - Restaurant of eternity, galactic showcase, tir quan and eyes of the universe. With these the odds are getting better.

This is for 1.1, with 1.2 things will probably change.