Matrix Games Forums

Inspired by Bingeling's excellent AAR which made a number of important observations on improvements that could be made to the AI (a worthwhile read, especially if you are a developer: tm.asp?m=2722742 ) I decided to compile my own brief list of immediately observable issues. In a strategy game, DW being no exception, having a dysfunctional AI can take all the fun out of an otherwise brilliant game - many of you are probably familiar with the absolutely atrociously poor AI of Firaxis' recentmost Civilization V. This is not to say that DW's AI programming is comparatively poor - but there are a number of issues which do hurt the quality of the game a great deal. Fortunately, I believe that most of these issues would be relatively easy to fix for the obviously talented developers behind DW. It is my belief that a solid AI ranks amongst the very highest priorities when it comes to a 4X title and thus I hope the developers give these issues the attention they deserve.

I can think of a number of more advanced fixes I would like to see done to the AI over the long term. However, I will not go forth with those just yet - the basics should come first.

You are very welcome to post your own observations as well. Maybe we can turn this thread into a thorough compilation of AI issues that need to be remedied. Developers are welcomed to issue feedback.

Following is my brief compilation based on half an hour observing my fully automated empire, using Return of the Shakturi v 1501:


* Empires start off with a very high potential research capacity. However, the AI does not utilize this because it doesn't build research labs unless it comes in the form of a research station built at a research bonus location. The effect is that for a very, very long time during the opening stages of the game - especially for empires that have a scarcity of research locations near their starting location - the AI is getting only a fraction of the research it 'should', putting it vastly behind a player who intelligently outfits his initial spaceport with extra research facilities to acommodate the large research potential. Consequently, the AI - even AI technocracies - are hopelessly behind the player in terms of research from the very beginning of the game.


* The AI fails at prioritizing colonization technologies. Empires of species that inherently have the ability to colonize continental planets should make this research subject a very low priority, choosing the marshy swamp equivalent instead. A Human empire has absolutely no use for the continental colonization tech other than as a passageway to accessing desert colonization and consequently researching it should take very low priority for a starting empire.


* The AI is much, much too reluctant to spend its cash. In my game my empire was at war very early on. My empire had a 100,000 cashflow and a huge 500,000 treasury yet invested in very few ships - obviously a terrible decision. I know that Bingeling found the same in his game. The AI needs to be much more aggressive about spending its cash as long as there is a large treasury and/or cashflow.


* The AI builds far too few exploration ships. The consequences of this are far-reaching but the most important issue is not finding many planets to settle. This means the AI's economy and power base builds up at a snail's pace compared to the player's. The AI needs to produce a *lot* more explorers.


* The AI also doesn't colonize fast enough. It will frequently have discovered a valuable planet yet wait several minutes before a colony ship is built and sent. In these situations there is no reason not to immediately build the colony ship and claim the planet for the empire. Please make the colonizing much more aggressive.


* In wars, the AI has an unfortunate tendency to send very weak forces (sometimes just a single ship) against vastly superior enemy fleets. These single ships will only try to escape once their shields hit 20% at which point the large enemy fleet has them surrounded and they get killed before they can power up the hyperdrives. This causes the AI to slowly but surely lose its entire fleet without having ever put up a real fight. There needs to be an algorithm that estimates the relative power levels of ships in a system: If enemy fleets are vastly superior in numbers and firepower, ships should immediately retreat, not fight.


* The AI also prioritizes targets poorly in wars and doesn't not take fuel into account enough. It will often send lot of ships large distances to take out minor targets only to have these ships completely drained of fuel and useless for a very long time until they can limp their way back to a space port. In my game, a 20+ ship fleet moved more than a sector to destroy a research station then flew the same sector back to homeworld to refuel. The AI needs to 'understand' the objectives of its various fleets better: Reserve large fleets like this to assaults on major enemy planets and leave the mop-up of minor targets to small strike force fleets. My AI empire would have lost its only truly valuable target, the homeworld, if the opposing empire had taken opportunity of this blunder.


* Likewise, the AI doesn't consider that it might be wise to [temporarily] remove ships that are out of fuel from a fleet to avoid slowing the entire fleet down. This would be a very nice optimization to see and would greatly help the AI.


* The AI will build expensive projects like space ports on worlds that are not secure - even while the planet they are built on is under attack! This frequently causes severe blunders like spending a good 20K+ cash to try to construct a Large Space Port only to have this space port instantly destroyed. Not only does this severe flaw seriously hurt the AI's economy, it also completely breaks the immersion for the player.

* AI needs to be better at estimating whether war is really worth continuing. My observed empire now has 'rampant' war weariness with no sign of wanting to end the war (which is going nowhere). Of course, these unnecessarily prolonged wars are to a great extent due to the AI's poor warring abilities (see above).

* Unit AI: When told to engage a moving target (such as an enemy fleet), ships will only move to whichever position this enemy fleet was at when the order was given - not to the fleet itself, because they don't update its position. This effectively means it is not possible to order an attack on any moving enemy unit and as such can be considered quite a major bug.

good compilation, maybe we should stick this one as it is to big for the wishlist...maybe stick it as a thread dealing with balance issues?

I would enjoy to have this stickied as a topic dealing specifically with AI issues.

I only have DW vanilla but some of these issues are kind of different, depending greatly on race and government type:

* Empires start off with a very high potential research capacity. However, the AI does not utilize this because it doesn't build research labs unless it comes in the form of a research station built at a research bonus location. The effect is that for a very, very long time during the opening stages of the game - especially for empires that have a scarcity of research locations near their starting location - the AI is getting only a fraction of the research it 'should', putting it vastly behind a player who intelligently outfits his initial spaceport with extra research facilities to acommodate the large research potential. Consequently, the AI - even AI technocracies - are hopelessly behind the player in terms of research from the very beginning of the game.

-> I agree, the AI really hates research.

* The AI is much, much too reluctant to spend its cash. In my game my empire was at war very early on. My empire had a 100,000 cashflow and a huge 500,000 treasury yet invested in very few ships - obviously a terrible decision. I know that Bingeling found the same in his game. The AI needs to be much more aggressive about spending its cash as long as there is a large treasury and/or cashflow.

-> well I've had like 2 games now in which the AI constantly builds ships, both civie and military, bringing my cash flow in the red from like 100k to -80k. The most aggressive AI seems to be for monarchy and reptile setting. In the pirate AAR, the AI builds defense bases, cruiser and destroyer en-masse even when at war with multiple empires. I'm absolutely thrilled by the efficiency of the AI

* The AI builds far too few exploration ships. The consequences of this are far-reaching but the most important issue is not finding many planets to settle. This means the AI's economy and power base builds up at a snail's pace compared to the player's. The AI needs to produce a *lot* more explorers.

-> As before, depends GREATLY on government type and race. In the pirates AAR, the AI CONSTANTLY maintained at least 8 exploration ships and explored faster than I could cope with. However, in my previous games as a human empire, the AI barely had 3-4 E-ships, and those just stood there doing nothing most of the time, when there were unknown systems with Korrabian spice a couple of sectors away.

* The AI also doesn't colonize fast enough. It will frequently have discovered a valuable planet yet wait several minutes before a colony ship is built and sent. In these situations there is no reason not to immediately build the colony ship and claim the planet for the empire. Please make the colonizing much more aggressive.

-> Again, depends on the race here 99%. The atuuk are simply OBSESSED with colonization. In all my experiences with this race as an AI empire, their greatest strength was not their fleet, but their HUGE ability to expand. At war, for every one of their colonies taken by the enemy they colonized 2 others. As long as you don't take out their infrastructure and capital, they're UNSTOPPABLE!!!!! They can get to >200 colonies by mid-game if you're not careful with them, they colonize everything and anything. Seems to me the AI is consistent enough here. As for the humans, the AI is a lazy son of a jack, constantly debating whether it's better to colonize that 90% continental planet with no resources 5000 sectors away or the one in the capital system 80% but with like 5 luxury resources.

* In wars, the AI has an unfortunate tendency to send very weak forces (sometimes just a single ship) against vastly superior enemy fleets. These single ships will only try to escape once their shields hit 20% at which point the large enemy fleet has them surrounded and they get killed before they can power up the hyperdrives. This causes the AI to slowly but surely lose its entire fleet without having ever put up a real fight. There needs to be an algorithm that estimates the relative power levels of ships in a system: If enemy fleets are vastly superior in numbers and firepower, ships should immediately retreat, not fight.

-> Hmm, the AI in DW vanilla does seem to act differently sometimes. As in my case where the AI empire send 27 ships against my large spaceport and almost took it out were it not for my reinforcement fleet. However, in direct combat with my ships they lost even though they had superior firepower and numbers, just because they just seem to float in circles...

* The AI also prioritizes targets poorly in wars and doesn't not take fuel into account enough. It will often send lot of ships large distances to take out minor targets only to have these ships completely drained of fuel and useless for a very long time until they can limp their way back to a space port. In my game, a 20+ ship fleet moved more than a sector to destroy a research station then flew the same sector back to homeworld to refuel. The AI needs to 'understand' the objectives of its various fleets better: Reserve large fleets like this to assaults on major enemy planets and leave the mop-up of minor targets to small strike force fleets. My AI empire would have lost its only truly valuable target, the homeworld, if the opposing empire had taken opportunity of this blunder.

-> As said before, sometimes the AI really does pick a good target, it just doesn't do it often enough in my opinion. In the pirate game, the AI empires hitting the boskara tyrants are systematically taking the outer systems before closing in on the heavily fortified central systems, which in my opinion is pretty good strategy. But in some cases, they head straight for the jugular, like when the Argos attacked me and subsequently lost the war.

* Likewise, the AI doesn't consider that it might be wise to [temporarily] remove ships that are out of fuel from a fleet to avoid slowing the entire fleet down. This would be a very nice optimization to see and would greatly help the AI.

-> I haven't seen a problem here, the ships that need fuel simply disobey the fleet orders and go refuel on their own, while the rest of the fleet does its job.

* The AI will build expensive projects like space ports on worlds that are not secure - even while the planet they are built on is under attack! This frequently causes severe blunders like spending a good 20K+ cash to try to construct a Large Space Port only to have this space port instantly destroyed. Not only does this severe flaw seriously hurt the AI's economy, it also completely breaks the immersion for the player.

-> Haven't seen this happening to my empire AI, but the AI empires have a serious issue with this.

* AI needs to be better at estimating whether war is really worth continuing. My observed empire now has 'rampant' war weariness with no sign of wanting to end the war (which is going nowhere). Of course, these unnecessarily prolonged wars are to a great extent due to the AI's poor warring abilities (see above).

->I used to think this as well, until I noticed that after a while of harassment the AI I'm at war with will accept being a subjugated dominion if it has lost some colonies and a good part of the infrastructure.

* Unit AI: When told to engage a moving target (such as an enemy fleet), ships will only move to whichever position this enemy fleet was at when the order was given - not to the fleet itself, because they don't update its position. This effectively means it is not possible to order an attack on any moving enemy unit and as such can be considered quite a major bug.

-> yep, this indeed is annoying, I fix it by simply selecting the indented target of the enemy fleet.

* In wars, the AI has an unfortunate tendency to send very weak forces (sometimes just a single ship) against vastly superior enemy fleets. These single ships will only try to escape once their shields hit 20% at which point the large enemy fleet has them surrounded and they get killed before they can power up the hyperdrives. This causes the AI to slowly but surely lose its entire fleet without having ever put up a real fight. There needs to be an algorithm that estimates the relative power levels of ships in a system: If enemy fleets are vastly superior in numbers and firepower, ships should immediately retreat,

->I totally agree with this.

The A.I is too slow to get trade agreements with friendly empires.

The A.I sometimes uses fleets to defend worthless recources.

I also have vanilla only.  I've been playing the Dhayut a bunch, and can certainly confirm Apheirox's points.  At least wrt them.

Just today I was trying semi-hands off, letting the AI do all the warfighting.  My fleet started out with a load of armies, and it captured one world with them.  Then it never, ever, picked up a single new army.  Even as my enemy -- Ikkoros -- seemed to have no problem doing so.  So they gradually took over my domain.  (I quit in disgust after they'd captured 10 of my planets.)

They had a fleet, and I didn't really; just a ragtag collection of my starting ships.  I had a large income, but the AI built very few ships for me.  So my fleet never could touch theirs, even though I had superior designs.

I did get the idea to see if I could force a fleet to do planetary assaults, by building a fleet that was purely transports with not a single weapon.  So: by my explicit orders, a fleet of 8 transports built.  They just sit there.  Just like the other fleet, they never pick up a single army.  However, in spite of the fact that they have zero offensive strength (which the program does know; it shows this), they start to prompt me for "attack this gas mine?".  "Attack that gas mine?"  With what?  The troops you don't have and can't attack a mine with anyway, or the weapons you don't have?  Maybe I should have said yes to see what they'd do.  But it can't have been good.  Probably just jump there then sit right on the mine forever.

You can't really compare findings for the vanilla game and RotS. They made some pretty significant changes - changes that greatly impact what the AI does well and does less well. Research, for example, is completely different, and this in turn affects many other areas such as colonization and ship design. I think for clarity's sake we should keep such compilation lists separate for each title.

The AI seems to really have a tough time managing fuel.  It seems like 90% of the time I spot an enemy fleet they're out of gas and limping somewhere to refuel.  Then you can fly right past them to their destination, blow up the undefended space port or gas station, and munch on the 0 energy ships when they land.

There are a couple of things that contribute to this:

• The AI also puts far too little fuel on its ship designs.  Escorts with 80 fuel are useless.
• The AI doesn't seem to put energy collectors on every ship, so they waste a lot of fuel while idling.
• The ships usually idle at a planet, which burns fuel as they chase the thing around.  And I think the movement to chase the planet might also keep the energy collectors from working, but I'm not sure.
• The AI doesn't seem to top up idle ships often enough.  It'll often start a voyage with half tanks.

Some comments to various.

* The AI is much, much too reluctant to spend its cash. In my game my empire was at war very early on. My empire had a 100,000 cashflow and a huge 500,000 treasury yet invested in very few ships - obviously a terrible decision. I know that Bingeling found the same in his game. The AI needs to be much more aggressive about spending its cash as long as there is a large treasury and/or cashflow.

I think ship building is depending on factors like cashflow and number of colonies. The AI should be less reluctant to spend money on infrastructure that is beneficial in the long run. In my AAR the AI wanted to build too many ships early on.

* The AI builds far too few exploration ships. The consequences of this are far-reaching but the most important issue is not finding many planets to settle. This means the AI's economy and power base builds up at a snail's pace compared to the player's. The AI needs to produce a *lot* more explorers.

I think the AI tries to keep 10 explorers up at least until the empire gets much larger than startup sizes. However, once they are in a war, the explorers are spent scouting enemy system, and are often killed off. Add do this the current reluctance to build when cashflow is negative, you soon enough have an AI with no exploration ships.

The AI should not forget exploring even if it is in a war. Especially as the AI tends to always have a war running.

The A.I is too slow to get trade agreements with friendly empires.

I think this depends on the racial modifiers, probably the "friendliness". Different races require different relations before they agree to agreements. For instance, the Atuuk are "extremely friendly" and wants free trade agreements even with empires they are "cautious" to.

The AI seems to really have a tough time managing fuel.

Indeed. But I figure the lack of energy collectors are racial features, stupid races should not have really smart designs. It makes them weaker, though.

Good fuel management is most likely really hard to obtain on the AI. For instance deciding whether to head far off to refuel far away, or wait for the local sources to catch up.

Really good fuel management is potentially quite hard.  But there's a big gap between where DW is (at least vanilla), and really good.  It doesn't take heavy programming to make these relatively minor changes:

• know when a fleet is "idle" (basically, has no attack order)
• when fleet is idle, and sufficient fuel in the same system, refuel any ships in system if they're less than 75% full
• when idle, and sufficient fuel in "nearby" system, refuel any ships in nearby system if they're less than 50% full
• when idle, and sufficient fuel anywhere, refuel any ships in system if they're less than 25% full
• when idle, do not "patrol" a planet (or base or whatever); instead "move to the planet" (so you ride with it for free, and can use energy collectors)
• when fleet is not idle, and sufficient fuel in the same system, refuel any ships in system if they're less than 25% full
• when not idle, and sufficient fuel in a "nearby" system, refuel any ships in that system if they're less than 10% full

(Exact levels and definition of "idle" and "nearby" should hopefully be configurable on some UI.)  These are all completely local decisions requiring no intelligence at all. 

Hmm, I have not noticed this fuel problem in any of my DW vanilla games yet... Perhaps this is only ROTS related, because I haven't seen AI empire ships limping back home. My ships occasionally run out of fuel, but so far I've seldom left my fleets to AI control, so it's just my forgetfulness, and for my empire I generally have the fuel mines built myself, simply because I'm CRAZY about building the infrastructure. It's so FUN!!!! seeing all those little freighters run around picking up stuff from my mines hehehhe*evil insane laughter*

Can anyone make this comparison? I want to know if this is a problem ROTS only related.

Cheers

I do not have ROTS.  I see ships in fuel distress all the time, both the AIs and my own if I let the AI control them.

I can honestly say I have yet to have fuel problems or see the AI empires stall because of it...or maybe I'm just not observant enough, I dunno...