Rant about stupid AI

You've discovered that all AI's, even GalCiv2, have their limitations. A good commander tries to learn the weaknesses of his/her enemy and expoit it. You have done this in your situation. The AI, being simply a set of programmed instructions for the computer player, is unable to learn and adjust to your tactics in this case. It's possible the AI programmers for this game might come up with a new counter for your strategy, but even if they do, you'll likely find a new strategy the AI doesn't counter.

From what I can tell, the AI's "flaw" is not protecting transports. It probably does this because in fleet-fleet combat, you have no control over what gets destroyed, therefore it's pointless to "protect" transports by melding them into a powerful fleet with escorts. Since you have ships that are much faster than the AI's (because it chose not to build super-fast ships), it's exposed transports are easily destroyed by your fleet. Even if the AI built 5 transports for each one of their warships, you'd still likely kill them all (and thus kill of large chunks of the AI's population).

In this case, the "solution" isn't to make the AI build ships that are faster than yours, instead the game should allow for better protection methods. Currently I do not like the fact you cannot defensively protect your vulnerable ships. You have to "intercept" the enemy, meaning if they field faster ships, you have to counter that or die. Instead, I think transports and other non-combat ships should be allowed part of a fleet, but do not get targetted until after the combat ships get killed. Yes a real commander would target the transports first, but this at least would allow real escorting of ships, vs having to intercept attackers before they get to your ships. Plus, using valuable logistic points for your transport(s) would make the fleet weaker, offering a weakness to this strategy.

Well that is true but this particular one AI can counter if programmed so. Algorithm is simple:

-Check opponent fastest combat ship speed
-Build transport with that speed +4
-When deploying them check for enemy ship presence within their combat range - if there are move transports out of the way). AI can do that better than human since humans are too lazy to calculate exact amount of squares between fleets ( I lose quite a few ships due to failure to do that) .
-Increase AI prioritization to conquer worlds in wars
-Make a new algorithm to deal with interception .- This is probably hardest thing to do, since you will have to make AI make choices (what region to protect, how to patrol areas etc).



This all is relatively trivial to code .Especially considering AI is already pretty good in quite a few areas. I bet those things are well within reach of stardock's AI programmers.

I think the biggest problem now is just the lack of speed in the AI. I personally don't understand how the AI designs such crappy ships.

For all the changes in ship engines, the AI isn't really much better off. The solution that needs to happen is the AI needs to use more engines... (and as you said don't bother with weapons on transports)

As always, the other dimension of the problem is that developers seem totally focused on small maps. On a small map, a slow ship isn't so bad. On a huge map, a slow ship is totally worthless. Especially if the transports can't follow up an attack fleet without getting destroyed. I just don't understand the hesitation to allow the AI to build faster ships.

AI came some way since original DL where he was designing ships with 2 and 3 speed in an end game (basically engine-less junk).

Now days he does decent individual ships design (7 and even 9 and occasional 11 speed).Problem is he has also to be mindful to what and how he puts them in fleets because his fleets are somehow always 3 to 5 speed - e.g. he puts slow junk in and slows whole fleet altogether.



Good point. Devs seem to not take into account balancing on maps larger than medium -that could be the explanation.

I also think they do not have any algorithm in place to counter speed. - e.g to check opponent 'ships speed and build their faster or at least as fast. They do this with weapons why not do that for speed also?

Perhaps because it is faster to see what happens in such maps ... How many hours are needed before reaching the stage where the AI flaws exposed in this thread are evident? Perharps the dev need savegames of that stage. Who knows?

One main issue with AI programming is the computing power need to number crunch the AI options. Basicly at Turn time the AI scans 8 move slots around each ship/object/planet to assess threats / opportunities / next moves etc. That involves Huge - utterly Huge - numbers of threads and calculations, to extend that to use more than 8 slots around its objects is not a practical possibility, the computing power for number crunching is far too great.

Brad recently added the option in the control panel to utilise more cpu power with the advent of dual-core processors, and that has helped in the AI responses. The AI will get better as the entry level PC gets more powerful, as more number crunching can be done.

Brad has done a great job with the AI since GalCiv II first rolled out, and its response is far far better than V1, but he does have practical limitations due to the number crunching limitations. He has refused to take the scripted route, rightly so in my view as in the long term that makes for a better AI, and considering its quality compared to others, and the plaudits the AI receives in the industry, he does a good job with the limitations he has.

Regards
Zy

Hi!

And an avid player will deploy a large ship with most engines and single gun. AI will build ships with only engines.
A player will deploy some tiny disposable ships in vicinity of the front. AI will not move troop transports close.
Like you see it's not so easy to counter the opponent that can adapt.


This one is mandatory.


Those are n-squared problems. Brute force doesn't work well here, because there's always one additional variable that can be included in calculations. You need to use heuristical approach. If that means some scripting, to "prune" less possible variants, then I vote for using it, because I'm using it all the time.

BR, Iztok

Well some sanity checks on "min guns" would solve this Of course player who is hell bent on exploiting AI will find a flaw anyways, but I am talking about some basic stuff non exploiting player would do and AI can counter that if programmed to do so.




Well I frankly dont see a lot of number crunching necessary. - I mean to calculate range around ship/planets and determine whether object is in there or not does not take much at all. I mean the grid is an array and all you have to do is check cells in there, 10x10 square is only 100 cells. Doing a for cycle on it for whole huge map takes nothing.

That is a brute force approach I am sure there is a lot more efficient algorithms for calculating range overlap on a grid between objects.

It does take a lot more number crunching than you might think. 100 squares, per ship, per race, possibly each race also checking the ranges of opposing ships they know of... multiply that with several hundred (or thousand) ships for each player, and you'll have a lot of time for coffee breaks.

Brute force simply doesn't work on a game this scale. That means you need a very efficient algorithm to solve the problem, which means it is not 'trivial' as you put it. If it was, it would already be included in the AI (and the AI already is one of the best I have seen in TBS games).

I wish you could make the AI use different strategies. Like in Age of Mythology. You could have blitzer's and boomers. Not a big selection, but it's there. I want to see AI that has incredibly fast fleets that uses strike teams, like DarkMadMax's strategy.

in my experience, if you give the AI enough time to mobilize it will send the bulk of its fleet at you gradually. but i agree that it's so slow as to be totally ineffective in offense.



bingo. the AI can only take so much into account. human minds benefit from being very good at 'soft' reasoning. we can analyze, sythesize and jump between different types of information for answers that are 'good enough' yet lack precision for the most part. we can glace at 'the big picture' and get a pretty good sense of what that big picture is. it seems to me that creating a set of algoirthms to parallel this would be very difficult, to say the least. how do you teach it to account for the indeterminate number of general strategies plus all the variations on them? i certainly don't know, but i can say that the AI in GC2 does a lot better with resources equal to the human player than do most game AIs. that doesn't mean there isn't room for improvement, and these discussions can be quite productive towards those ends.

* * *

one thing no one's mentioned is that the AI virtually never builds any sensors of any type - lone sensor ships, armed or not; sensor-equiped warships in fleets; or sensor-equiped starbases. it doesn't even research sensor tech, and i've never seen it build the EotU even when i've gifted it Sensors IV. part of being able to intercept fast enemy ships is being able to see them before they can get to you.

there are other ways to combat fast ships, it's just that they're not economically competative and, obviously, the AI hasn't been programmed to execute them.

i usually build enough SBs around my home turf to keep the FoW fully illuminated, and like many i usually defend my home turf with interceptor forces. fleets of medium and small ships seem to be optimum for this.

sometimes by late game, when i feel like letting the game drag on for a while, i'll build a network of military SBs (4 per sector placed in the 4th diagonal square from each sector intersection, or the nearest available square if that one's occupied).

while this is a total WASTE of production power, it definately makes defending your home turf a totally different task. it most squares will be affected by 4 SBs in this pattern (with a spread of 3-6, depending on irregularities in the dispersuion due to planets and stuff in the way), which typically means +8 to movement. add another +3 from tech boosts and +1 from grav accelerators, plus another 1 or 2 if you buy the speed boost (which i occasionally do). this means a ship with no engine components will still move 12-14 sct/wk through your home turf. a warship with two Hyperdrive IIIs moves 14-16 sct/wk without any help - in other words, comprable movement rates.

with the new warp inhibitors, you can also reduce the speed of invading enemey forces by -12 (to a minimum of 1), virtually immobilizing them once they start a turn in your territory. even my own fast ships wouldn't be much of a match against this.

for me the boosts to offense and defense aren't the big attraction of military bases; the movement bonuses (and enemey penalties) are. the weapons/defense boosts are definately a perk, of course. they can help a few squadrons of tiny fighters become a powerful, mobile interceptor force. they can get where they need to go pretty quickly; i can build a lot of them because they're cheap; being tiny fighters, they also gain levels faster; and since i also typically play with the repair modules in SD's example mod, i can repair them very quickly.

of course, i only ever build such a network by the time it's clear that i've already won the game, and even then only occasionally. i don't even remember how many modules military bases have in the default game because i typically play with mods on, but i'm pretty sure it's 50+. even if you only equip one type of weapon-boosting module you're still looking at 40ish constructors per base. i can only manage building one of these networks in the end game when i can have most of my planets churning out 'structors every turn or two; and in that case, usually only on a map with abundant planets so that i have enough shipyards to product many stuctors per week.

i hope some of the new features in 1.7 will live up to making it easier to manage bases and making them a more robust part of the game, but the fact remains that having to sacrifice a constructor for every single SB module is cost inefficient.

one very superficial solution could be to add upgraded versions of the construction component making it cheapers, but like most i think it shouldn't consume an entire 'structor per module. if nothing else i'd be content to see some modules use 'structors, such as the battle stations modules. or perhaps some modules should be upgradable without need for a new constructor (from Armor I to Armor II, for example).

the games not perfect by any means, but there aren't really any other games in this genre i feel drawn to presently. so i try to enjoy what i have as much as possible.

Speaking of constructors, using the mining-ship principal (where the ship survives the upgrade, but takes "time" to do the upgrade) would make things less uneconomical.

Currently, you could build a whole fleet of warships with the production points necessary to churn out a dozen constructors to make a single viable starbase. This makes them very uneconomical. The only starbases that are worthwhile (IMO), are economic starbases (near a cluster of high-production planets), mining starbases, and maybe influence starbases (where you're not trying to fight the enemy anyway, so lost warship production redirected to expendable constructors isn't an issue).

Make starbase upgrades take time, allow multiple constructors to work an area at the same time (to give a benefit to building more constructors), and maybe require expending a constructor for the initial base. But right now, starbases cost way too much both economically (to fund the bc's to make constructors), and lost production (planets building constructors instead of warships or infrastructure building).

Transporter (fleets) can be easily protecte by stacking them with warship fleets. (AFAIK stronger fleets defend first)



There isn't much point in researching or using sensors. On lower difficulty levels the AI is tough enough, on higher difficulty levels it gets significant sensor bonus.