Where is the vastly improved AI?

from Sins Forums
1.03 claims to provide us with a challenging AI to fight...but I haven't seen one so far.  If "vastly" improved AI means givng them a resource bonus with the "unfair" difficulty, than that's just sad.  They still run from every engagement, they still are so stupid that they can't win a fleet battle, and they still seem to have no clue what to build (wow...a fleet of 30 percherons, 30 robotics cruisers, and 2 Marzas...great strategy, what are you going to do repair me to death?).  I see no point in playing the single player version of this game anymore, it's just a game of repetative boring crap.  Attack, enemy retreats, nuke planet, colonize, repeat.  The only semi useful mechanic was utilized in 1.02, the one that forces the enemy AI to attack if you begin bombing their homeworld.  The problem is, in late game scenarios, I've spent so much time chasing/destroying fleeing fleets, that the forces they assemble to defend their homeworld are rag-tag at best, and offer no kind of challenge. 

Now that I have gotten used to how the AI plays (attack the place with the least defenses, the only thing the AI can do well), I have played multiple games against four or more unfair AI opponents without losing a single capitol ship.  All you have to do is keep your fleets in key positions, and whenever you move them to any grav well where an AI fleet is they retreat.  I have never seen a different outcome...they always run, and this has been a problem since day one.  The fact that more time was taken to give the AI the ability to calculate whether or not it can win rather than giving it some actual tactical prowess really saddens me.  An unfair AI should be able to use tactics to beat me when it's outnumbered, or outgunned.  An unfair AI should know that letting me nuke all his planets without a fight is a futile effort, or that they should make as stand before they're down to 1 world.  Maybe "vastly improved AI" means they give up so I don't have to waste my time playing phase jump cat and mouse anymore. 

This claim of "vastly improved AI" is a complete falsification, and I am extremely dissapointed with this patch, which I was hoping would fix the issue.  If Stardock cannot do anything to actually improve it, not just giving it a resource bonus, than at least implement the "last stand at the homeworld" mechanic for key planets also.  That way, the computer will actually give a sh!t about important planets as well.   They should do that at the very least.  It makes no sense that they would pour so many resources into a planet and then just abandon it at the first sign of the enemy.

If this can't be adressed at all I'm about ready to give up on single player, and I don't have the time to spend long periods of time online playing.  Please, Stardock, give the AI some balls so I can have fun playing this game. 

And please, don't call me a single player noob... not everyone plays the game online.


49,017 views 29 replies
Reply #1 from Stardock Forums Top
You don't like not fighting 400 disciples? I've only played one game and I already saw massive improvement, resource bonus aside. I was using a level 8 Sova has bomber suppression against an Advent ai and the little bastard ignored it until I clicked the insta build button to replace my then dead fighters. It got almost halfway through them before it was repeatedly disabled till the ability wore off. Is it a good tactical ai, obviously not, but the little shit is a hell of a lot better than it was with 1.02. I got to fight a mixed advent fleet with tolerable capship support and use of support cruisers, the bombers blew the shit out of my kodiaks while they disabled my Sova to prevent me from keeping the fighters continuously built, drained my dunovs of antimatter so I couldn't restore shields on the kodiaks, and all around kicked the shit out of me with their twice as large fleet. It was a twice as large fleet of course, which means if I'd actually been playing instead of just checking it still would have been easy, but being able to lose 2-1 odds fights is a hell of a lot better than not being able to lose 10-1 fights because they're only using light frigates, don't have any of the abilities researched, and run as soon as the single capital ship dies and they're no longer able to break through the shield restoration of a whopping two high level dunovs.
Reply #2 from Sins Forums Top
To be honest I'm pretty disapointed as well, went 2v 2 with a friend of mine agaisnt 2 unfair AI.

Dominated them totally, they had a few HUGE fleets with weird as combonation of units, and they seemed to just constantly patrol their systems, barely attacking. When they did attack, they retreated within 10 seconds, making the game pretty unexciting.

I love this game, but I really hope that some changes are made to the AI or something.
Curtains - Customize Windows with additional styles
Reply #3 from Sins Forums Top
I have noticed some improvement in the AI, but it still seems unwilling to really ever participate in a pitched battle if the odds are not heavily in it's favor. If this were in neutral territory, or even on the offensive, it would be forgivable, but the fact remains that it is unwilling to protect key worlds that it needs to win. It will run from a Dead Asteroid just as fast as it will a fully developed Terran or Desert.

I noticed the patrol thing too, that Tamsta17 mentioned (thank you, Forwarning). I was watching it just jump between it's systems with decent sized clumps of units, just back and forth, for some reason. I wasn't sure if this was some weird pathfinding issue, or a way to deter raids (no way am I sending a siege frigate raid through an area being constantly crossed by a dozen Light Frigates).

Overall though, the AI is improved, I would say. Not vastly as was said, and with the new market being just so easy to get resources from, the AI's advantage on Unfair seems to be totally negated. But it's a bit better.
Reply #4 from Sins Forums Top
After 4-5 games vs hard I am not impressed with the AI changes. There are some little things it does to make things a bit harder, like when the AI has the advantage it waits to engage until you are deep into the well making it harder to just phase jump away ships taking damage, but for the most part it is just as easy to beat as before.

It still just spams light frig, still jumps siege into then immediatly out of your systems, still runs like hell nine times out of ten, still falls apart when rushed, still just plain sucks. Not a fan of the new market either, but i'll hold that in for later.

KP
Reply #5 from Sins Forums Top
Somewhat indicative - first game I played Vs the 'unfair' AI, I was kind of impressed when I saw it throw 18 cobalts at me right from the start. I ended up pulling my cap back from the fight, and it arrives at a dead asteroid... where the computer has decided to send both an akkan and a colony frigate and have them circle the thing aimlessly.



Reply #6 from Sins Forums Top
Most of the time, when an AI is calculating whether it can win a fight, there should be some kind of 'as good as it gets'-type flag. If we're not likely to get a better match-up against this group of units in the near future, we fight regardless of the odds and devote all resources to replacing the units we're about to lose. For Sins, I suppose it would work to only use this logic when determining odds in high-resource systems, that would result in behavior very much like how human players act in RTS games.

If you want to get even more detailed, you can incorporate the estimated unit cost of retreat into the odds of winning, i.e. how many units can you and the enemy build in the estimated X number of seconds before the next battle? The result is scaled by (my distance from shipyard - enemy distance from shipyard), so I get this outcome:

If I am attacking and my production is stronger, my odds are stronger.
If I am attacking and my production is weaker, my odds are weaker.
If I am defending and my production is stronger, my odds are weaker.
If I am defending and my production is weaker, my odds are stronger.

Both of the above mechanics can incorporate a small random component that will make the AI very unpredictable.
Reply #7 from Sins Forums Top
i dont know where the OP got the idea that 1.03 would provide (to quote the OP) a « vastly improved AI » ! ...

let me quote the official changelog for 1.03, in what concerns the A.I. changes :

-AI now knows how to surrender (based on time elapsed, relative economy, relative fleet value, resource reserves, and allied strength). Also, an AI player will not surrender if he has a human ally.
-AI is now smarter about retreating and is willing to commit local suicide or accept local Phyrric victories if the global gain is higher and will also factor in local allied forces.
-Fixed various AI researching and building stall outs for the various trees (military in particular).
-Rewrote AI siege building and attack logic.
-Improved AI use of Black Market.
-Autocast for many disabling abilities improved.
-New AI Difficulty "Unfair" that gives the AI a tax income bonus and resource income bonus from resource asteroids (there is no other AI cheating in the game even at the Unfair difficulty level).
-Wide variety of AI upgrades.
-Improve AI taunting.

elsewhere, on the forum, i believe i read that the A.I. would receive a more significant overhaul in version 1.04 ... but i dont recall that it was ever promised that the A.I. would be « vastly improved » (or something like that) in 1.03 ...
Reply #8 from Sins Forums Top
@Sorceress

-AI is now smarter about retreating and is willing to commit local suicide or accept local Phyrric victories if the global gain is higher and will also factor in local allied forces.
End of quote




Reply #9 from Stardock Forums Top
I don't think anyone could make a reasonable argument that the AI was NOT improved. I'm pretty certain it was improved. However, the use of the ambiguous term 'vastly' and the straw-man argument resulting from claiming anyone said that are overboard.

1.03 was a very tiny patch in comparison to what 1.04 will be (and that IS what Ironclad have said). No major updates were made, just some smaller hacks to fix a few immediate problems (and a rather odd black market change, but that's another topic). So if you're looking for more improvement, just provide well-reasoned constructive feedback for 1.04.
Reply #10 from Stardock Forums Top
i dont know where the OP got the idea that 1.03 would provide (to quote the OP) a « vastly improved AI » ! ...let me quote the official changelog for 1.03, in what concerns the A.I. changes : -AI now knows how to surrender (based on time elapsed, relative economy, relative fleet value, resource reserves, and allied strength). Also, an AI player will not surrender if he has a human ally.-AI is now smarter about retreating and is willing to commit local suicide or accept local Phyrric victories if the global gain is higher and will also factor in local allied forces.-Fixed various AI researching and building stall outs for the various trees (military in particular).-Rewrote AI siege building and attack logic.-Improved AI use of Black Market.-Autocast for many disabling abilities improved.-New AI Difficulty "Unfair" that gives the AI a tax income bonus and resource income bonus from resource asteroids (there is no other AI cheating in the game even at the Unfair difficulty level).-Wide variety of AI upgrades.-Improve AI taunting.elsewhere, on the forum, i believe i read that the A.I. would receive a more significant overhaul in version 1.04 ... but i dont recall that it was ever promised that the A.I. would be « vastly improved » (or something like that) in 1.03 ...
End of quote


On the homepage of sinsofasolarempire.com:

The new version adds dozens of new features into the game including:

• Many new game setup options such as fleet size, pirate options, income rate speeds, build speeds and more
• Players can use the + and – keys in single player to control the game speed
• Improved black market provides a more volatile market
• Players who drop can now be placed by cunning computer AI players
• Vastly improved computer AI
• New difficulty level “Unfair” for really hard core players
• Computer players will surrender
• Improvements to Ironclad Online
• Lots of tweaking and balancing based on feedback
• MUCH more…

That's where I got the idea, right on the front page, where even a person with impaired vision can see it easily. It was the first article on the page after the release of 1.03.

I don't think anyone could make a reasonable argument that the AI was NOT improved. I'm pretty certain it was improved. However, the use of the ambiguous term 'vastly' and the straw-man argument resulting from claiming anyone said that are overboard.1.03 was a very tiny patch in comparison to what 1.04 will be (and that IS what Ironclad have said). No major updates were made, just some smaller hacks to fix a few immediate problems (and a rather odd black market change, but that's another topic). So if you're looking for more improvement, just provide well-reasoned constructive feedback for 1.04.
End of quote


This is my feedback for 1.04, though it is presented in perhaps a little more angry fashion than a well reasoned feedback should be. That is only because I have been clamoring on the forums since the release day to fix the "coward AI" issue, but obviously my input (and the input of many others who are being frustrated by the very same issue) is being ignored, or at least has not been adressed or commented on by the sins staff up to this point. I also find it ironic that people are complaining that they are tunning the game for single player sins owners, when really the single player version of the game is the most annoying and frustrating portion. If I had more time, I would rather play online and lose all my games than play single player and win over such pathetic and stupid AI opponents.
Reply #11 from Stardock Forums Top
Well, it is. It still blows, but that doesn't negate reality. Building 400 light frigates and a couple capital ships nearly every game, never researching any abilities, and getting creamed even with 10-1 odds is a vastly inferior ai when compared to the still pathetic, but vastly superior ai we have now. It sucks, but it's still better than the previous even more sucky version.
Reply #12 from Stardock Forums Top
Well, it is. It still blows, but that doesn't negate reality. Building 400 light frigates and a couple capital ships nearly every game, never researching any abilities, and getting creamed even with 10-1 odds is a vastly inferior ai when compared to the still pathetic, but vastly superior ai we have now. It sucks, but it's still better than the previous even more sucky version.
End of quote


As far as I have seen, the AI still does this. In fact, not only do they build an unnecessarily large amount of light frigates, they also build an equally unnecessary amount of support ships (like I said in my original post, I was fighting fleets of 30 robotics cruisers at times from TEC opponents). As far as researching abilities, that might be improved, but they still spend very little on offensive weapons research. In a longer game, I had almost 60% more weapons research than any of my 4 computer opponents. So, in that game, even if I had an equal amount of forces to my opponent, I would still win if they would even bother to fight me.

In my opinion, this patch tried to put a band aid on a broken leg (with the AI issue), and then took an aluminum bat to your kneecap (with the very questionable black market "fix").

Reply #13 from Sins Forums Top
The 'researcher' AI will throw a vasari fleet with a good whack of Subverters and Overseers (and other race equivalents) together and use their abilities in combat. It suffers instead from failing to build a fleet with enough big hitters in it and being terrible at picking which of your units to kill first.
Reply #14 from Stardock Forums Top
The 'researcher' AI will throw a vasari fleet with a good whack of Subverters and Overseers (and other race equivalents) together and use their abilities in combat. It suffers instead from failing to build a fleet with enough big hitters in it and being terrible at picking which of your units to kill first.
End of quote


Again, this is irrelevant when AI fleets retreat if they even have the slightest chance of losing the battle. I've even seen a fleet with with a Kol, ~7 siege frigates and some cobalts retreat from basic planet defenses on a dead asteroid (4-5 turrets and a few hangers). "OMG I lost 2 siege frigates RUUUUN!"
Reply #15 from Stardock Forums Top
@ bonerfest > Thanks for your correction. I am not visually impaired : I had focused on the official 1.03 changelog, not paying any attention to the homepage.

My hypothesis, then, is that a marketing-oriented author, on the homepage, got carried away.

As far as I'm concerned, the official changelog is the most correct & precise, reliable technical info ...
and I do not see any « vastly improved » A.I. there.

Marketing ...
Reply #16 from Stardock Forums Top
@ bonerfest > Thanks for your correction. I am not visually impaired : I had focused on the official 1.03 changelog, not paying any attention to the homepage.My hypothesis, then, is that a marketing-oriented author, on the homepage, got carried away.As far as I'm concerned, the official changelog is the most correct & precise, reliable technical info ... and I do not see any « vastly improved » A.I. there. Marketing ...
End of quote


Actually, I have impaired vision and I saw it...I wasn't implying that you have impaired vision :p

But still...
-AI is now smarter about retreating and is willing to commit local suicide or accept local Phyrric victories if the global gain is higher and will also factor in local allied forces.

I haven't seen anything close to local suicide by the AI, not even to defend their most valuable planets (other than the home world, but that was a mechanic implemented in 1.02). In fact, they would rather retreat and lose a capitol ship and 20 frigates while trying to jump away rather than to try and fight. So I still don't see most improvements in the change log making a significant impact.

I also find it funny that posts like this, where I have gone into great detail to describe the problems that I am having so they can be improved upon, are ignored while posts like "STOP WHINING 1.03 IS GREAT" (where the OP wrote a 2 sentence post where he told everyone to "stop bitching") are actually praised by the devs. It kind of makes me feel like posting this is a futile effort.
Reply #17 from Sins Forums Top
I haven't seen anything close to local suicide by the AI, not even to defend their most valuable planets (other than the home world, but that was a mechanic implemented in 1.02). In fact, they would rather retreat and lose a capitol ship and 20 frigates while trying to jump away rather than to try and fight. So I still don't see most improvements in the change log making a significant impact.
End of quote


Sadly, that's all too true.

What's also still annoying is, that the AI retreats when defense buildings are built (ie. even before they're finished). Instead of killing all constructor ships the AI then just gets wasted by newly built defense installations.
Reply #18 from Sins Forums Top
I use to play a game called Submarine Titans and it had the ability to tell your ships to fight down to 75%, 50%, 25% health, or to the death. Maybe something like this could be implemented, but fleet wide. Then the AI will only turn tail and run at a certain loss level.
Reply #19 from Sins Forums Top
The real trouble with the AI is that it doesn't do what it SHOULD do in a losing strategic position: harry the enemy behind its lines.

Suppose you are reaching the point where you are building a larger military. At this point, when you drop in-system, you will likely have a tactical advantage from numbers. The AI deciding to retreat will make sense.

But, what it SHOULD do at that point is not to retreat to some planet closer to "home." Instead, it should find a way to get behind your lines, forcing you to commit ships to the defense of your own systems. This may well force a "pullback." It doesn't even have to win a battle on such efforts: Lee managed to prolong a losing position for almost 2 years with Gettysburg.

Sadly, properly programming an AI is quite beyond most developers. Short of giving the AI significant advantages in order to overcome its stupidity, not much else can be done. After all, it took them decades to come up with a chess AI that could do better than humans regularly, and that's a SIMPLE game by comparison.
Reply #20 from Sins Forums Top
Sadly, properly programming an AI is quite beyond most developers. Short of giving the AI significant advantages in order to overcome its stupidity, not much else can be done.
End of quote


Stardock's Galactic Civilization 2 A.I. is possibly the very best programming achievement in TBS games.

Stardock is the very involved producer of SINS, it is not the developer : Ironclad is ... so, the latter remains solely responsible for the coding of the SINS A.I.
Reply #21 from Sins Forums Top
It's MUCH MUCH easier to program an AI for 4x games than an RTS game (yes, sins takes place in real time so it's considered RTS in major aspects)

I've created a small one for one of my wc3 games where you try to trap a black-dot (AI) that moves around on a checkerboard. The players objective is to capture the black dot and make it impossible to move.
The AI was pretty basic (only looked outward 5 spaces) but was difficult to trap it.

In realtime though, it's much much harder to program an AI; short of telling it "You need X number of this building at 5 minutes" and other such orders.
Reply #22 from Stardock Forums Top
But, what it SHOULD do at that point is not to retreat to some planet closer to "home." Instead, it should find a way to get behind your lines, forcing you to commit ships to the defense of your own systems. This may well force a "pullback." It doesn't even have to win a battle on such efforts: Lee managed to prolong a losing position for almost 2 years with Gettysburg.
End of quote


That is a good idea, and I would approve of that. If the AI did that, even in a scripted situation, it would at least provide the illusion that the computer is using strategy.

I think they should also think they should refine the AI's ability to make capable fleets (there is no need to have more that 10 of each support ship class per fleet, otherwise you screw yourself by not having enough firepower. As it is, and has been since day one, the AI builds a ton of support cruisers, which may be helpful in the earlier stages, but late in games is only hurtful.

Many times, I see TEC AI's building lots of (What's the name of the TEC's colony cap ship?) instead of Kols, Dunovs, and carriers, all of which bring more firepower to the table. Instead of, say, having a pair of Dunovs to "cross heal" with the shield regen ability, or 2 Kols and a colony cap ship to supplement with the targeting ability. Plus, there are lots of other effective cap ship combinations that are not used by the computer. They also build a lot of Cobalts, but very few LRMs in comparison. The AI needs a better "equation," if you will, of what it takes to create a capable fleet. It focuses too much on support, and not enough on firepower. They also focus much more on planet development over fleet construction, which doesn't really make sense since your fleet is what makes or brakes you in the end. That's another thing that might help.

Another idea might be to help their build template. Right now, the computer seems to randomly place buildings, using the "auto place" option, which is the stupidest part of the UI and is a hindrance more then a helper. The AI needs to know where to make its buildings. Turrets are the most useless building in the game unless they are bunched together (at least to the point where more than 1 turret at a time will be in range of a given enemy ship) and upgraded. Yet, the computer does neither of these things, and it makes conquering their planets too easy. Giving the computer a build template is necessary, and is not impossible.

These are just some ideas, but I don't know what is or is not in the realm of possibility since no devs want to comment on the AI problems in any thread in which they have been discussed, at least not in a manner which would insinuate what they can or cannot do/what they plan to do about it.

Reply #23 from Sins Forums Top
From what I have been seeing (just finished another game), is that the AI is unwilling to fight a battle of attrition. A normal strat when outnumbered near or at a production center is to keep em busy and build/rebuild ships as they are destroyed and send em right back in. It means adjusting whatever economic plan you may have had to meet the situation, perhaps upping the fleet supply earlier than you would like. The closest I have seen is the AI still faithfully spamming light frig as they retreat.

The building placement the AI sets up for itself is one improvement I have noticed, can't say for autobuild as I never use it. Everything seems more thoughtfully placed, it tends to have two clumps of structures on totally opposite ends of the well, an interesting thought. The AI now uses all of the buildings available to it, very nice to see. It really just needs to know how to fight and produce backup for the battle on demand, then we'll be talkin.

And on that note the AI should have a higher priority for attacking ship producing structures. Now thay are all but ignored making it too easy to hold out while cranking out ships to keep them occupied. If it is going to harass me it may as well do it properly.

KP
Reply #24 from Sins Forums Top
@ bonerfest > Thanks for your correction. I am not visually impaired : I had focused on the official 1.03 changelog, not paying any attention to the homepage.My hypothesis, then, is that a marketing-oriented author, on the homepage, got carried away.As far as I'm concerned, the official changelog is the most correct & precise, reliable technical info ... and I do not see any « vastly improved » A.I. there. Marketing ...
End of quote


I'd just like to say, that the changelog, didn't even add most of the changes. Calm down with the zealotry to defend the game's flaws. It's a great game, but you seem to think it's almost perfect and the dev's can do no wrong. But they're human too.
Reply #25 from Sins Forums Top
@ bonerfest > Thanks for your correction. I am not visually impaired : I had focused on the official 1.03 changelog, not paying any attention to the homepage.My hypothesis, then, is that a marketing-oriented author, on the homepage, got carried away.As far as I'm concerned, the official changelog is the most correct & precise, reliable technical info ... and I do not see any « vastly improved » A.I. there. Marketing ...
End of quote


I'd just like to say, that the changelog, didn't even add most of the changes. Calm down with the zealotry to defend the game's flaws. It's a great game, but you seem to think it's almost perfect and the dev's can do no wrong. But they're human too.