Suggestions for AI improvements

from Stardock Forums
Sorry if this is discussed in the nearly 900-post long next-patch thread. I read slow, and this is probably stuff for later patches anyway.
 
Ok, the AI has a few basic flaws that really hurt, IMO. (boring "credential" stuff follows, skip it: Please note that this is all from playing against hard AIs, and in my last game I beat 4 lock-teamed hards with a single medium on my side. Easily. Flogged them. *Without* pirates (it was Gateway - and no, I didn't colonise the other system, just blockaded it off for my ally, who was still almost entirely useless). I play single-player a lot more than multi-play (sadly), so I'm keen for an AI who is more of a challenge, or at least who play a little more like humans...)
 
Here are some flaws in the AI that are horribly easy to exploit:
 
1. The AI deals very poorly with system defences.
 
That is, it blunders directly into them. Some changes that'd really help here:
 
First, the AI should target builder frigates as a matter of priority, the moment they start building something. Perhaps bump this priority up for faster ships to help prevent exploit...? As it stand the player can build turrets just behind the ones being targeted, and the AI won't worry about it. You can keep this going almost indefinitely, with a slow snake of turrets moving across the grav well. It will beat all but the biggest of fleets, as long as sieges don't kill the planet to prevent building. Let them kill the builders as soon as they start building.
 
Second, there should be a way for the attacking siege frigates to consider, for a moment at least, the defending turret range circles "impassable". Can that be done? At least see if there's a spot within range of the planet, outside of turret range. At LEAST make players cover the whole planet. Better yet, if there's no spot uncovered, let the AI work out where the spot of least resistance is... Maybe, to reduce load, just test, say, 20 spots evenly distributed around the planet. Pick the one with the fewest turrets in range X.
 
Third, increase the "scare factor" of defences. AI fleets seem perfectly content to sit under the guns pretty much to destruction, long after they have any chance of winning. This might be lower for siege frigates, who can validly charge into the guns...
 
Fourth, and this is a bit of an ask, can the AI fleets just decide to raid? If the "scare factor" of the defences are too high for the fleet, can it, instead of running away, instead go into "raid" mode? Can it do a concentration of fire consideration similar to the siege frigates above? Basically, can it find turrets with less than X other turrets in range, and - importantly - civilian structures with less than X turrets in range and then kill them? Not perfect, I know, but that'd knock out those structures you usually have to leave exposed, and if that (and inevitable fighter/bomber activity) knocked out enough turrets, that might bring the scare factor back into the black, and they could continue the destruction as usual... alternatively (or, perhaps, as well), let a "retreating" fleet take a second retreat test based only on the enemy mobile units (that is, ignoring defences). If it doesn't need to retreat from just the enemy fleet, have a chance to retreat *forward*, towards an enemy system.
 
2. People hate chicken AI fleets.
 
First, make sure they count allied fleets and defences as part of their force. It doesn't seem they do, and so they retreat even when they have the numbers.
 
Second, I suggest a "desperation factor". Each time the AI initiates a retreat, raise its retreat threshold for next time. Bring this threshold back down to normal over time - slowly. Also increase the threshold temporarily at planets with high development: say, full civilian infrastructure and at least, say, 8 orbital structures. That's in addition to thinking of the defences as part of the fleet.
 
Just on the fleet theme, it seems to me that a comparison should be made to the enemy fleet, and more anti-that should be built. I think the fleet compositions are generally fairly good (within the usually quite poor tech level of the AI), but still a little random. My carrier fleets rarely encounter much in the way of enemy flak or fighter force. Although that might just be an issue with slow replacements, which players tend to suffer from, too.
 
3. The AI doesn't seem to build stuff.
 
Every game, every single one, the "Structures Built" line shows I built at least 4 times what my closest competitor built, usually much more. Even before I start trashing them it's way, way higher. This is not just turrets, either - the only thing the AI seems to build consistently is extractors.
 
I'd suggest that every AI type should have a "optimal end state" (slightly different for each AI type). This should include a trade port on every planet as a matter of priority for ALL AI types, and a refinery for most types. That's EVERY planet and capable asteroid. Players almost always do. It should also include some global stuff, like *only* a couple (say, total planets divided by 6) of frigate factories, and only, say, 2 capital ship factories. It's just a waste of a slot otherwise. The AI should be using "left over" slots for labs until it has 8 of each - it's amazing how long AIs take to get to the end of the tech trees, if they ever do - with those extra slots going to culture and stuff only once the labs are built. Again, in my experience this is how players do it. Obviously some personalities would add, for example, cultural buildings earlier on. not all AIs would act the same.
 
I realise the AI probably has difficulty juggling military and economic spending - obviously players do too. I'm not sure how you do it now, but one method might be to split the AI (sorry if this is the bleeding obvious, I'm just saying how I'd do it...). That is, give the AI a military and an economic "personality", and give a proportion of  income to each. So the economic side might get 50% of the resources, with military getting 50%. If one side gets beyond a certain threshold (say, 3000 of each) that it can't spend (researching one tech at a time, obviously) all income goes to the other side until the first spends some. Let the thresholds change as the game progresses (perhaps by time? Or increase the military proportion - to a maximum - if fleet size drops too far below others'? etc.) Basically, let some amount of resources be earmarked for economy - the AI seems fine at building big fleets, but awful at developing its worlds and researching. I assume it's because it has too high a priority set to ships.
 
In fact, it might be worth splitting it three ways: military, economy, and tech.
 
Finally, most AIs should have a tech-order. I see the value in some fairly random builds out there (so keep the option in), but most decent players will have a fair idea of what they want to research, and when. Each AI type (military, economic, researcher, etc) should have picked, at the start of the game, from a couple of tech paths, pre-selected. Put some conditionals if you want to get fancy (eg. enemy has over X carriers, research flak frigates now now now. Get Ice / Molten tech if we've just killed the defences of one of those, especially if we don't have anything else to colonise...) Well maybe they already do, but they never seem to build enough labs or put enough into research to matter.
 
---
Anyway, please add any suggestions you might have, and / or comment on these.
17,034 views 7 replies
Reply #1 from Sins Forums Top
Nice breakdown and I fully agree on it except for that awfull blue text u used lol. Had to highlight it to read it ;)
Reply #2 from Stardock Forums Top
Better? :)

Thanks for pointing that out - I'm in the TGN version, so it's a white background.
ObjectDock - Access shortcuts and frequently used applications from an animated dock.
Reply #3 from Sins Forums Top

great OP work...i havent played vs. hard comps yet...would u tell us what makes them « hard » (harder than the normal ones) ?

after reading your OP, i dont get the impression that they are that hard !!
Reply #4 from Sins Forums Top
I think "hard" is too easy just now and I hate cheating AIs, so the suggestings in the first post are very good to improove the AI in my eyes.
Reply #5 from Sins Forums Top
While I mildly disagree with #2, I'll mostly agree with the others and add:

The AI doesn't seem to bother with exploiting neutral wells! Those wells can be really useful (not to mention fun to fight over) but the AI seems to pretty much ignore them.
Reply #6 from Sins Forums Top
I'd like to see the AI produce more balanced fleets in the mid to end game. It's definitely slanted towards spamming as many light and LRM frigates as it can, and then it tends to mass-produce capital ships to go on top of that. While properly-upgraded capital ships can be very worthwhile, mass-producing them will seriously hurt your early economy due to their sheer cost, the cost of researching the tech to allow the crews, and researching the tech that provides the considerable amount of supply for them (which imposes a permanent reduction in resource-gathering rates), can leave the AI gimped later on. Early-level capital ships usually are not worth their "weight" (resources/supply/tech requirements) in regular frigates and cruisers and need to be chosen carefully and shepherded into being high-level bruisers. As it is, I run through lv1 AI capital ships that usually outnumber my own capital ships 2 to 1 (on HARD mode, no less!)

I don't think I'm alone in wanting to see the AI no longer produce the "lol your planet's dead" fleets of 8-20 siege frigates, either. It's simply impossible to prevent these fleets (in various portions of the game) from knocking out a planet's population and health, but then dying to defenses/the rest of your defending fleet. Certainly it costs the player some resources to recolonize the planet and rebuild planet developments. But the cost in siege frigates to the AI is far worse. It just ends up that the AI is slowly but surely throwing resources away in a manner that is extremely aggravating to the player. If the AI came in with a fleet to tie up/maybe even chase away your current fleet and defenses in order to COLONIZE the defeated planet, that would be good strategy. But the AI doesn't do that.

The AI needs more of a carrier focus as well. It's not uncommon for me to find the AI with few to no carriers in their fleets and little flak defense on top of that. If I roll in with 20+ fighter craft, that AI fleet is in for a real pounding. Furthermore, it makes AI fleets taking on planets with defense hangers (which should be the majority of your defenses for just about any frontline planet) particularly weak against those types of defenses.

It's been mentioned above me, but the AI really needs to go after neutral resource extractors. Those things are pretty amazing, and nowadays against an AI I consider them pretty much invincible.

Other than these addendums, I am in agreement with the OP on other needs for AI changes.
Reply #7 from Stardock Forums Top
Yeah, good additions - resource extractors and better fleet makeup. Perhaps some "best case" fleet makeups, preselected? Again, I know this is cheesy, and some random / experience variation could be added, along with some really random builds, but it seems a good way to make sure computer players have fleets that make sense.

So, for example, a "carrier" fleet builder goes for something like 40% carriers, 15% cruisers, 10% lrms, 10% capitals and 5% siege frigates, 20% anti-whatever-the-enemy-has-lots-of. A "bruiser" fleet personality might go with no carriers at all, some flaks and tons of close range stuff. A "stand off" fleet will have a mix of carriers and LRMs, and so on. Whatever the numbers, the AI gets a favoured fleet mix assigned at the start of the game, perhaps randomly selected from a list, and tied to the research tree (as above) and/or overall personality.

Perhaps allow an AI to completely change favoured fleet makeup if it loses, say, 60% of its maximum-built popcap...

One extra thing for capital ships - the AI has little regard for its own. Capital ships should really check to see how long they'll last taking the fire they are, compared with how long it'll take to jump out. When that ratio starts getting close to 1:1 it should leave the fleetand run like hell - while the rest of the fleet fights on! Players do that. As it stands AI capitals only retreat when the whole fleet does.

Tsed: Actually, I for one don't really mind the chicken fleet thing - as a player it can be a smart thing to do. But it is true that, especially in a game against several AIs, the dying AI is better off at least doing some damage to you early, rather than continually running and then only fighting when you've wiped out most of their planets, and when they can't possibly construct a second fleet.

...actually, amendment to #2: With the desperation factor, consider also how close the fleet is to the popcap. If the fleet is within, say, 10% of maximum, it *should* get a lot braver, regardless of the size of the foe. At that point its best chance, even if outnumbered, is to try to whittle you down by throwing multiple waves at you, if it possibly can. It's got little to gain waiting, in any case. So an AI's is fairly close to being at its popcap, the AI should just get a big boost to its bravery, whether that's used to fight your fleet or attack your systems. Drop that bravery off steadily until, say, 60% of maximum. And of course, even at maximum it wouldn't be completely suicidal...