SMACX Thinker Mod

[BFG]

My testing said the stock AI would build them but not use them.  Do you have different results from that?  Or are you testing Thinker Mod's AI.

I was talking about Thinker, but I'll admit my testing on this one wasn't as good as it should be.  I was going off the computer's ability to successfully automate the unit and find a good base square, and how they're handled when they're the only colony in Scenarios.  I'll try an actual game scenario and see.

If you provide this as a predefined unit, the stock AI will go after sea bases with these.  It's bad enough that I often refrain from building sea bases early on, since they're likely to get mind controlled before I can do anything to protect them.  However I can't see how this significantly boosts the AI in any way.  Rather, it annoys me.  I'm ok with that because I think it's fair, I've abused the AI with my own foil probe teams for years.

Oh yes, these annoy the crap out of me too.  I've lost at least one Thinker game as Zak due to getting hit by a fleet of these.  From Deirdre of all people.  But they do strengthen the AI, which is a big part of the goal here!

[Induktio]

> On a different topic, I have been testing various base units to see which can boost AI and which are useless.  I would be interested to hear thoughts or feedback.

> Probe Foil
> Enhanced probes and various other enhanced units

These have already been implemented by the design_units feature long time ago. It builds armored/foil probes and AAA garrisons when the proper techs are available. It's always possible to add new unit combos to boost the AI capabilities though.

> Needlejet Colony Pod (or any Aero Pod)
> Gravship Formers (AI can only use on land)
> XX Missile (Not sure it’s better than vanilla missiles)
> Aero Transports
> Supply Foil (AI doesn’t know how to use)
> Gravship Supply

These types of units are not currently implemented in Thinker, e.g. it does not provide any notable behaviour beyond the standard AI logic. I'm also evaluating which ones of these could be implemented in the next version. It does look like some of them could be helpful but we'll see later which ones will be implemented.

[bvanevery]

I may have found a bug in Thinker mod's unit design stuff.

In my SMACX AI Growth mod I do not allow ECM or Hypnotic Trance on probe team units.  I figure that armored probe teams are already enough of an exploit.  Consider how easy it is to make an Elite probe team unit, as they get PROBE bonuses, MORALE bonuses, and whatever increases in base probe team morale that various techs give.  Incidentally, I also eliminated increases in base probe team morale, because they're clearly overpowered as the game goes on.  It's like exploit city and in my mod I've reined all that in.

So I was quite surprised to see the AI make an armored hovertank probe team unit with Trance capability.  I checked my design changelog and indeed, I did eliminate this, it wasn't some mistake in the flags I set.  In my alphax.txt the flags are:

Code: [Select]


; Flags  =
;          000000000001 = Allowed for Land units
;          000000000010 = Allowed for Sea units
;          000000000100 = Allowed for Air units
;          000000001000 = Allowed for Combat units
;          000000010000 = Allowed for Terraformer units
;          000000100000 = Allowed for Noncombat units (non-terraformer)
;          000001000000 = Not allowed for probe teams
;          000010000000 = Not allowed for psi units
;
Hypnotic Trance,       1, CentEmp, Trance, 000011111111, +50% defense vs. Psi

Another possibility is Thinker is exposing a bug that was already in the game's code.  But the stock AI has never designed a hovertank probe team unit, so I've never observed it before.

Adding a Fusion reactor into the mix might be relevant as well.  I know that Scient's patch allows a reactor type to be specified for a predefined unit.  I don't do that in my mod, nor do I use Scient's patch.  My point is the Scient patch may have changed the flag behavior somehow.

The importance to my current game, is Morgan is dominating with all those nice thermal boreholes.  I'm dead last, and he's sending these Trance hovertank probe teams at me.  Fortunately I built the Hunter-Seeker Algorithm, just barely before he and others were going to.  But if it weren't for that, I'd have a real problem.  Mindworms are my armored probe team antidote, since they aren't affected by probes and regular armor doesn't matter to them.  I do allow probe teams to have 3-Res armor, for lack of any way to prevent it, and I think that's all the anti-mindworm buff a probe team should have.  I don't want invincible probe teams in my mod.

Details probe teams should not have Trance capability

probe teams should not have Trance capability

Note the unkillability.

Details armor determines probe team combat

armor determines probe team combat

I can't take it out with a probe team either.  Note that I have an Elite probe team vs. his Disciplined team.  Note that I'm using an infantry chassis probe team, to avoid the enemy getting any bonus from his 3-Pulse armor.  Finally note that I'm the Data Angels.  In my mod, they get the usual +2 PROBE bonus, and no other factions get any PROBE bonus at all.  Furthermore in my mod, the only ways to get a PROBE bonus from social engineering are +1 for Power and +2 for Thought Control.  I've chosen Power so I have +3 PROBE.  Now, maybe that just raises my probe team morale, I forget.  The point is, I'm the baddest ass probe team faction that anyone can possibly be in this game, no one else has my advantages, and I still can't even crack these things.  They are invulnerable in probe-to-probe combat, because you can't add a weapon to a probe team to counteract the armor!

Sorta forgot that he's got a Fusion engine and I'm attacking with Fission, but I still think I should be doing better.  I'll try throwing a few units at it to check...

It's quite an exploit.  I did choose to teach the AI to use this exploit, by providing a 3-Pulse Speeder Probe Team predefined unit.  I figure if the player can do it, why not let the AI do it too.  I wouldn't mind changing the game so that probe team armor is not allowed.  However there's no way to do that by .txt modding the stock game.

[bvanevery]

I'm not seeing a way to delete my posts, so I'm reusing this slot for another possible bug.

In my mod, as in the stock game, the compulsion of the Cybernetic Consciousness is to choose Cybernetic Future Society.  Nevertheless, I check on my ally in my current Thinker test game, and I see they've chosen Thought Control!  I've never seen that happen before.

Details Cyborgs that do not believe in Cybernetic

Cyborgs that do not believe in Cybernetic

My compulsion line in cyborg.txt is a little different from the stock game.  Instead of:

Code: [Select]

  Future Society, Cybernetic, GROWTHI have:

Code: [Select]

  Future Society, Cybernetic, RESEARCHI have no idea if that would make any difference, but that there is a difference, is worth mentioning.

[bvanevery]

I'm reusing another one of my undeleteable thread slots for the same bug.  This time it's Deirdre!  Why the heck is she Socialist?  (My version of Planned.)

Details Gaians who are not Green

Gaians who are not Green

Both the stock game and my mod have the same compulsion line:

Code: [Select]

  Economics, Green, PLANET

[bvanevery]

Oops, sorry, had multiple tabs open in my browser and posted some AAR stuff in the wrong place.  I've replaced those errant posts with a bug report.

[Induktio]

So I was quite surprised to see the AI make an armored hovertank probe team unit with Trance capability.  I checked my design changelog and indeed, I did eliminate this, it wasn't some mistake in the flags I set.  In my alphax.txt the flags are:

Another possibility is Thinker is exposing a bug that was already in the game's code.  But the stock AI has never designed a hovertank probe team unit, so I've never observed it before.

Adding a Fusion reactor into the mix might be relevant as well.  I know that Scient's patch allows a reactor type to be specified for a predefined unit.  I don't do that in my mod, nor do I use Scient's patch.  My point is the Scient patch may have changed the flag behavior somehow.

The importance to my current game, is Morgan is dominating with all those nice thermal boreholes.  I'm dead last, and he's sending these Trance hovertank probe teams at me.  Fortunately I built the Hunter-Seeker Algorithm, just barely before he and others were going to.  But if it weren't for that, I'd have a real problem.  Mindworms are my armored probe team antidote, since they aren't affected by probes and regular armor doesn't matter to them.  I do allow probe teams to have 3-Res armor, for lack of any way to prevent it, and I think that's all the anti-mindworm buff a probe team should have.  I don't want invincible probe teams in my mod.

So yeah, that marvelous piece of engineering was indeed created by Thinker's design_units feature. And that's not even the end of story. When it gains more techs, it will upgrade both the armor and the reactor. And also add Algo Enhancement when it's available. Thinker calls Smacx's native propose_proto function to create those prototypes, but the problem with those abilities looks like this function does not do any checks if the rules allow the ability for the unit. So that looks like an oversight. Probably should rewrite some parts of that code.

What those lopsided combat odds really reflect is the overpowered value of fusion vs. fission reactors ("Power" number). Anyone fighting with obsolete reactors is at a huge disadvantage. It would be nice to mod those reactor values but the last time I checked, they were hardcoded in the binary in so many places it's difficult to change them.

In my mod, as in the stock game, the compulsion of the Cybernetic Consciousness is to choose Cybernetic Future Society.  Nevertheless, I check on my ally in my current Thinker test game, and I see they've chosen Thought Control!  I've never seen that happen before.

My compulsion line in cyborg.txt is a little different from the stock game.  Instead of:

Code: [Select]

  Future Society, Cybernetic, GROWTHI have:

Code: [Select]

  Future Society, Cybernetic, RESEARCHI have no idea if that would make any difference, but that there is a difference, is worth mentioning.

That one is intended behaviour. New social_ai decides what society models they will run and the old code has been replaced. They are not allowed to select social opposition models, but currently the social preference category does not get any specific priority either. I could maybe increase that priority but I'd rather avoid situations where the AI's always choose one model regardless of other considerations. Now it has some situational awereness, where if the AI has enemy factions closeby, it will prioritize POLICE, MORALE, and PROBE more, among other effects.

[bvanevery]

Thinker calls Smacx's native propose_proto function to create those prototypes, but the problem with those abilities looks like this function does not do any checks if the rules allow the ability for the unit.

Well that solves a mystery at least.

New social_ai decides what society models they will run and the old code has been replaced. They are not allowed to select social opposition models, but currently the social preference category does not get any specific priority either.

This isn't good from a character narrative perspective.  Deirdre could chew you out about not being Green, when she herself could be the very Planned or Free Market economy she's railing against!

It's also not necessarily good from a game political perspective.  The original 7 factions were carefully positioned to be at each others' throats in various balanced and predictable ways.  Possibly similar with the 7 factions in the expansion, by themselves at least.  Now, randomly combine any / all of them, and social choices definitely slant one way or another.  Player makes choices like, "No one will get pissed about my Politics this game.  Free ride."

So, I've modded to get rid of "weak" factions.  There's nothing like a stock Morgan in my mod, he's much tougher because he's got piles of instant money.  No SUPPORT penalty either.  Most of my factions are supposed to be retaining their original character, i.e Morgan builds.

Some of the social engineering choices weren't championed by anyone, so I changed that.  The Pirates champion Wealth.  The Data Angels champion Thought Control.  The Believers champion nothing!  Only faction that has complete free will like that.  The Cult of Planet takes over the championing of Fundamentalist.  Marr goes for Power because he is a Conqueror, and Svensgaard isn't doing it anymore.

Lots and lots of playtesting to balance these.   Don't really want the factions or the techs to be genericized, with some kind of "one true borehole" path through the game.  It makes TOTAL SENSE for Morgan and is TOTAL ANATHEMA for Deirdre.

Some of the factions, they retain a secondary compulsion, i.e. ECONOMY for Morgan, PLANET for Deirdre.  Other factions, in my mod they don't have a secondary compulsion.  It's a limiter, it straitjackets their choices in the social engineering table.  I made a lucky find, that setting this to Nil, gives the stock AI more choices.  And that the stock AI will actually make those choices!

I only want them limited if it's totally to the faction's advantage, and also true to their narrative personality.  So yes, it's ok if Zhakarov fixates on RESEARCH.  Desired, even.  Cha Dawn, however, I do not require to fixate on PLANET.  He's already got +2 PLANET, maybe the AI will decide that's enough?  Maybe he's like a televangelist, completely cynical.  Will damage the planet with Capitalism to keep his mindworms well funded.  Dishonesty in cults and religions is a thing!

I could maybe increase that priority but I'd rather avoid situations where the AI's always choose one model regardless of other considerations. Now it has some situational awereness, where if the AI has enemy factions closeby, it will prioritize POLICE, MORALE, and PROBE more, among other effects.

My view is, the .txt modder should be deciding what's constrained about a faction and what isn't.  That's what makes a faction a faction.  But to the extent that constraints are relaxed, the AI should be doing whatever it thinks best.  I've changed things around so that Chairman Yang can end up with a Capitalist Police State, even with the stock binary.  Yes the AI actually chooses that, no new code!  Or he may choose something else.  Point is, Capitalist Police State is modern China.  It's a thing.

One thing I would like all factions to do, regardless of orientation, is spread better at the beginning of the game.  Not do stupid things with colony pods.  With the stock binary, I and others noticed that an Explore focus seems to trigger better early empire growth.  Don't know why, only that it does.  Hence how I got the name, "AI Growth mod".  The Explore trigger wasn't an original idea, it was something I found in the forum archives.

[Induktio]

Lots and lots of playtesting to balance these.   Don't really want the factions or the techs to be genericized, with some kind of "one true borehole" path through the game.  It makes TOTAL SENSE for Morgan and is TOTAL ANATHEMA for Deirdre.

Some of the factions, they retain a secondary compulsion, i.e. ECONOMY for Morgan, PLANET for Deirdre.  Other factions, in my mod they don't have a secondary compulsion.  It's a limiter, it straitjackets their choices in the social engineering table.  I made a lucky find, that setting this to Nil, gives the stock AI more choices.  And that the stock AI will actually make those choices!

The thing here is that playing according to some narrative doesn't appear from nowhere, instead somebody has to code very specific rules how that should be handled. And often in the stock game there are situations where playing strictly according to a factions narrative, e.g. using only their prioritized social settings, results in inferior play. The dilemma didn't rise earlier with borehole building because the AI basically never built them anyway. Sea level rise wasn't really relevant when the AI had such poor production etc. Like if one AI is planet friendly, should it stop borehole building altogether? How is it going to deal with the disadvantage in production vs the other factions then? Possibly they could have a reduced amount of borehole construction for the sake of the narrative but not sure how well they would do.

The social_ai formula is still a work in progress, it will probably receive tweaks to the scoring system. But people shouldn't expect the choices to be exactly similar to the stock AI's choices because there will probably be significant differences.

[Induktio]

I think I now remember an issue that came up during testing the social_ai feature. If we look at Gaian faction file for example, it has these lines:

  Economics, Green, PLANET
  Economics, Free Market, nil

The last values on each line are supposed to be the social priority and opposition effects, in that order. In the actual data structure, these values are always -1 which is an undefined value. It probably means the rules parser ignores these two values and they have no actual effect on the AIs. These values are also mentioned here: http://alphacentauri2.info/wiki/Faction.txt_editing_guide

Scient might know more about this since he contributed to the guide but don't know when he's going to release the decompiled code. I notice the guide also states that "The Social Priority line specifies the social choice a faction will oppose and be unable to select in social. This has no effect for a human using the faction." This seems incorrect, it should refer to "Social Opposition line" and the opposition category value should also affect human factions equally.

[bvanevery]

And often in the stock game there are situations where playing strictly according to a factions narrative, e.g. using only their prioritized social settings, results in inferior play.

That's why I sat around for 10 months rebalancing factions to make them superior again.  My Gaians aren't weak, they kick ass.  They have free Biology labs.  I had to give Zhakarov free Biology Labs and Network Nodes, because he really wasn't tough enough in the face of escalating competition between all my rebalanced factions.  And he doesn't have any PROBE penalty anymore, nor any unhappiness every 4 citizens.  All of that was necessary, partly because the Discover part of the tech tree doesn't give any weapons anymore.  You can't count on the University being the guy with the Chaos gun now.  Nor the Hunter-Seeker Algorithm.  Nor are the Gaians the ones with the Missiles now.  Nor do the Morganites have the Silksteel Armor.

I don't want your AI to try to adjust every flaw of the game.  Other modders deal with that stuff. 

You can't anticipate every balance issue that arises with mods.  Yes, you can focus on the stock factions and alphax.txt rules.  Please don't.  Not in ways that mess up what other modders are doing in their mods.  There's a difference between being compatible and having an interface boundary with other mods, and thinking you personally are supposed to solve every problem yourself, with Thinker mod as some kind of standalone ultra-competent game experience.

Many of us like SMAC, and still talk about it and recommend it to others 20 years later, because of the narrative and world building.  I mean, that's why /r/4Xgaming actually gives me the most downloads, because people over there still have collective memory and still care.  Nobody, literally nobody, has done anything remotely close to SMAC in the narrative dept.  It's still the exemplar.

Stepping on that in the interest of gameplay, is not a feature for some of us.  Yes we want to be more challenged, because we are bored.  But for many of us, we don't want the Gaians to be turned into the Morganites, or really every faction to be the Morganites.  If we wanted that, we could play with Randomize Faction Personalities, or just start a game with 7 Morganites!

I think it's notable that Pandora happened, and didn't get traction, very exactly because they didn't really bother with all the narrative and world building stuff.  I'm told they eventually achieved a somewhat modern AI, due to some uber-modder's work.  Sort of like you, Induktio, you could have easily been that guy.  Going that road, Pandora's your endgame.

Sea level rise wasn't really relevant when the AI had such poor production etc.

Actually it is still not relevant.  I have proven in the other thread, that the game is cheating on AI eco-damage.  3X difference compared to a human playing the same faction!  There's a reason I've never adopted the play style you've implemented for the AIs.  I'd get slaughtered by mindworms doing that.  The whole world would end in a dripping wet mindworm apocalypse.

I hope you decide to make it relevant, i.e. make the AI play by the same eco-damage rules as a human player.

Like if one AI is planet friendly, should it stop borehole building altogether?

I'd go so far as to say, the Gaians should never be allowed to build a borehole at all.  If that was a modding feature, prohibiting some particular action, I'd be all for it.

In contrast, the Cult of Planet would build the most boreholes.  They're complete jerks.  They'll tell you anything, long as they get you indoctrinated into the cult.  Anyone remember In God We Trust with Marty Feldman, and the Church of Divine Profit?  That's Cha Dawn!  Or at least, that's how I imagine Cha Dawn in my mod, certainly a direction they can be going in.  Deirdre is absolutely earnest.  Cha Dawn could easily be a big fat liar.

How is it going to deal with the disadvantage in production vs the other factions then?

In an ideal world, the Spartans would deal with it by invading.  The Peacekeepers would deal with it by turtling up, growing population, and voting.  The Gaians would make mindworms and attack -3 PLANET factions.  The University would turtle up and research.

Yeah, that's all much more AI than just cranking up the condenser, crawler, and borehole production.  I have no idea how much more, or what the stock AI already did.  I've adjusted as much as I could at the .txt mod level, giving the few AI inputs I was allowed to give.  Some of those did help, so I know that the stock AI isn't a complete dunce at differentiating play styles, as a matter of faction input.  The original designers did try to do this, it's only a question of how successful they were.

[bvanevery]

It probably means the rules parser ignores these two values and they have no actual effect on the AIs.

I know for fact that setting the secondary compulsion to Nil, caused a greater range of stock AI choice than I otherwise saw them do.  The Hive never went Police State Free Market until I did that, for instance.  Obviously I've never looked at the plumbing, the machine code.  But I would submit, it's not so easy to observe how a piece of machine code behaves.

I also observed factions being much more willing to move back and forth between Fundamentalist and Democratic, after I gave them this freedom to do so.  If for instance you compel a faction to seek MORALE, well then they're always going to go Fundamentalist.  So I refrain from compelling them.  Or I'm careful about what compulsions actually result in meaningful choice in the SE table, and what compulsions don't.

I believe that choosing a secondary compusion, such as PLANET, makes the AI seek those entries in the social engineering table, and prefer those over others.  However, it is not ironclad.  I have often seen a faction choose the None column choice, if the AI thinks it's a better deal.  For instance, my mod gives +1 GROWTH for Survival.  Factions often choose it, rather than Capitalist or Socialist or Green.  It gives 1 benefit and 0 penalties, which might be better than the 3 benefits and 2 penalties of other choices, depending on what the AI wants to do.

I don't have rigorous proof of what an "affirmative" secondary compulsion does.  I think of it as a "strong suggestion" to the AI.  I'm guessing the AI will not choose something that goes against the suggestion, i.e. a PLANET penalty instead of a bonus, but I'm uncertain of this.

[Induktio]

You can't anticipate every balance issue that arises with mods.  Yes, you can focus on the stock factions and alphax.txt rules.  Please don't.  Not in ways that mess up what other modders are doing in their mods.  There's a difference between being compatible and having an interface boundary with other mods, and thinking you personally are supposed to solve every problem yourself, with Thinker mod as some kind of standalone ultra-competent game experience.

Many design problems are balancing acts, including finding a balance between difficulty and interesting narrative. It certainly doesn't have to be exclusive, although currently this mod has been slanted in favor of increasing challenge. I will say straight up though, Thinker is supposed to be a playable experience without extra mods, but modding should always remain a possibility. There's simply so many players that prefer the vanilla game mechanics.

Sea level rise wasn't really relevant when the AI had such poor production etc.

Actually it is still not relevant.  I have proven in the other thread, that the game is cheating on AI eco-damage.  3X difference compared to a human playing the same faction!  There's a reason I've never adopted the play style you've implemented for the AIs.  I'd get slaughtered by mindworms doing that.  The whole world would end in a dripping wet mindworm apocalypse.

I hope you decide to make it relevant, i.e. make the AI play by the same eco-damage rules as a human player.

It seems like the post is jumping to conclusions pretty fast. Although it's not fully documented, in the stock game the AI gains an advantage in eco damage at thinker and transcend difficulty levels, as can be seen from here: http://alphacentauri2.info/wiki/Difficulty At librarian and below, the eco damage should be equal for human and computer bases, unless I missed something.

The post also looks like you've entirely ignored how the clean mineral formula works or are not aware of it. As a player, it's perfectly possible to run 50 mineral production from boreholes with zero eco damage. This just requires some timing with the first fungal pops and then rushing lots of tree farms or similar clean mineral boosting facilities. The AI might be able to do this too if the behaviour was coded in, but currently it's not implemented obviously. http://alphacentauri2.info/wiki/Ecology_(Revised)

If we were to go on increasing AI eco damage, then the increased fungal pops would also act to lower the eco damage because the clean mineral cap increases faster. The increased number of fungal pops would also cause sea level rise even sooner. Fungal pops can also be highly exploitative mechanic for the player since the worms generate planet pearls. That's just the way eco damage works in the stock game. Now, Yitzi's patch did attempt to do lots of changes here, but not sure if it makes sense to duplicate them.

It's always possible to change librarian level cost factor to 7 and see how the equal eco damage would change things when the cost factor is similar to transcend level. I just doubt it makes that much of a difference here.

[bvanevery]

As a player, it's perfectly possible to run 50 mineral production from boreholes with zero eco damage. This just requires some timing with the first fungal pops and then rushing lots of tree farms or similar clean mineral boosting facilities.

It "just" requires you to do a whole lot of stuff.  As opposed to say, spamming whatever you like with all kinds of nifty boreholes, free pass no problemo.  All those requirements constrain your growth and conquest.  The AI suffers none of that by comparison.  I mean the game "just" requires you to conquer everything on the map in order to win it, just about everything could be spoken of this way.

If we were to go on increasing AI eco damage, then the increased fungal pops would also act to lower the eco damage because the clean mineral cap increases faster.

You still have to stay on top of it, and some factions with their Economy choices will not have the same options for mitigation.  What's AI Morgan gonna do?  Double down on Free Market like a good little climate denier, that's what.

The increased number of fungal pops would also cause sea level rise even sooner.

I'm not worried about that.  I want the AI to suffer the consequences of the eco-damage it does.  So that it isn't so easy for it to make 40 minerals cities and invade me like I'm standing still.  I may get some flooding, they may get giant mindworm stacks that completely wipe out their cities, no matter how big they are.  Have you played through that sort of thing?  Did it because of illegal chemical weapons a couple of times, it wasn't pretty.

It's always possible to change librarian level cost factor to 7 and see how the equal eco damage would change things when the cost factor is similar to transcend level. I just doubt it makes that much of a difference here.

And then other things change like how many drones are unhappy.  Shifting these things up and down doesn't work unless all of the factors can be adjusted.

[Induktio]

I'm not worried about that.  I want the AI to suffer the consequences of the eco-damage it does.  So that it isn't so easy for it to make 40 minerals cities and invade me like I'm standing still.  I may get some flooding, they may get giant mindworm stacks that completely wipe out their cities, no matter how big they are.  Have you played through that sort of thing?  Did it because of illegal chemical weapons a couple of times, it wasn't pretty.

Did you read how that eco damage formula works? Of course you're going to get lots of problems when doing atrocities while UN charter if effective. I suppose that part of the formula works as intended, e.g. limits certain exploitative strategies. Otherwise the original eco damage is pretty broken in the sense it can be gamed by very simple strategies. It doesn't make sense to say eco damage gives some huge advantage to the AIs since human player can easily exploit it. With higher eco damage, it's even easier to trigger the first fungal pops and increase the clean mineral cap...

Sure it's possible to play a variant where the player is not allowed to cause any eco damage, but that is a self-imposed limitation and the game will get much tougher as a result. But then again, giving AIs cost discounts is a self-imposed choice for the player too. What maybe could be done here, is have some kind of a config option to have the AI adjust the extent it uses heavy terraforming improvements, like boreholes and condensers. It could even adjust it according to the faction definition, but I'm not sure yet how that might be implemented.

Edit: So looks like the wiki has two guides for eco damage:
http://alphacentauri2.info/wiki/Ecology_(Advanced)
http://alphacentauri2.info/wiki/Ecology_(Revised)

There are some discrepancies in the details how the eco damage is calculated, in this case the advanced version is probably correct. Improvements number should be divided by 8 before added to the minerals, and one of the terms in the final eco damage number is "3 - PLANET", and not "PLANET + 3". But the textual description of eco damage should be correct there.