Improving AI - what are the priorities?

[Yitzi]

Feel free to do with it as you like, Yitzi. But this is just the AI stuff, no bugfixes.

That's fine; I can handle bugfixes, but have more trouble with AI.

And I really cannot guarantee any results with so many changes your patch makes. The AIs usually looks very stupid with a lot of modding, they cannot cope with new rules, obviously.

My patch actually makes very few obligatory changes (just bugfixes and minor changes and closing off exploits the AI doesn't know to use anyway); if you explain what each change does, I should be able to figure out whether to include it in mine.

As for stockpile energy bug, thanks for notice, I did not know. I'll look into it later, I certainly need to have a clean bug fix on the AI patch as well.

That might get a bit tricky, as your fix is fundamentally inseparable from the bug it caused; what I'm doing is reverting your fix and introducing a new fix, but that's building on other stuff I did that builds on other stuff which took me a long time...

[kyrub]

Updated the patch (see file section).
From the original list (page 1 of this thread), I have tackled primarily points 3,5,6,7. Points 1,2 and point C were touched as well, more than solved.
I am curious how it will work.

     
1.  Production issues      Far too many mins tied up with supporting useless/unwanted units
2.    More combat units with mobility.
3.    AI never attacking with hurry penalty
4.    Production changing
5.    Improve AI-AI trading of techs
6.    Smarter air attacks
7.    Better air defense

UNDER REVIEW
A.    Smart probe attacks
B.   Worker placement
C.    Attack from base

[Kirov]

I played around with your patch yesterday. I'm surprised at the AI's trading techs so nicely, I still can't come up with the tech lead after 100+ turns (although to be fair, I do everything to help the AI, giving them techs and contacts as they ask). Their SE choices also seem interesting - Demo/FM/Knowledge for Zak, nice. I'm still surprised why some AI is so eager to pick up Fundie (Deirdre) and/or Power (Lal).

Oh, I forgot - I saw a Miriam's supply crawler actually crawling 2 energy! Granted, it was their only crawler and I fail to see another one anywhere else, but still it is a massively huge deal for me. I felt like I saw a unicorn. Thanks for that experience and keep up the work if you can.

[kyrub]

I played around with your patch yesterday. I'm surprised at the AI's trading techs so nicely, I still can't come up with the tech lead after 100+ turns (although to be fair, I do everything to help the AI, giving them techs and contacts as they ask). Their SE choices also seem interesting - Demo/FM/Knowledge for Zak, nice. I'm still surprised why some AI is so eager to pick up Fundie (Deirdre) and/or Power (Lal).

Hey Kirov, you have played with badly damaged version, the corrected version [n] from today has much better results, as I have seen. So, I suggest you retry with the new one.

SE choices are not affected, I had to call off my previous attempt. It is a huge task, huge priority now as well. Main plan is to learn AIs to pop-boom. I have noticed that once they get Cloning Vats, they immediately become mounstrously efficient. There is hope that with pop-booming they could keep the pace a bit longer.

Oh, I forgot - I saw a Miriam's supply crawler actually crawling 2 energy! Granted, it was their only crawler and I fail to see another one anywhere else, but still it is a massively huge deal for me. I felt like I saw a unicorn.

Hehe, unicorn. Crawlers are actually part of some previous versions. Sadly, they are badly targetted... 2 energy is a great succes even in my book.

[TarMinyatur]

I played around with your patch yesterday. I'm surprised at the AI's trading techs so nicely, I still can't come up with the tech lead after 100+ turns (although to be fair, I do everything to help the AI, giving them techs and contacts as they ask). Their SE choices also seem interesting - Demo/FM/Knowledge for Zak, nice. I'm still surprised why some AI is so eager to pick up Fundie (Deirdre) and/or Power (Lal).

Oh, I forgot - I saw a Miriam's supply crawler actually crawling 2 energy! Granted, it was their only crawler and I fail to see another one anywhere else, but still it is a massively huge deal for me. I felt like I saw a unicorn. Thanks for that experience and keep up the work if you can.

I completed a game under 444(m). I, too, noticed the AI trading effectively among themselves. I didn't establish a lead in technology until I built Tree Farms (which the AI could benefit greatly from...but, understandably, the Hive, University, and Believers can be at war with each another for 100 years, and can't afford to invest in ivory-tower facilities when hostile troops are attacking relentlessly.) The Spartans suffered late in the game due to horrid inefficiency. Some cities were losing 36 energy out of 36 energy produced. But, in fairness, the Spartans aren't meant to rule a vast empire containing dozens of cities. (This is not a criticism of kyrub's patch -- he didn't change SE-related stuff.)

As far as kyrub's specific AI work: Miriam did attack my combat troops with her air force many times. Aggressive indeed. She would have done serious damage if I hadn't scrambled SAM Choppers which had a slight edge in weaponry (8 vs 6, at first). Which is, in fact, a good way for Miriam to kill my SAM Air units if she doesn't have any Needlejet Interceptors. My SAM Choppers lose their +100% attack bonus when they scramble to defend ground troops.

I was stunned to see a Miriam using a Supply Crawler to harvest a distant 4-mineral mine early in the game. Yes, like spotting a Unicorn! She didn't build another crawler, as far as I know.

I'll try the 444(n) patch now.

[Geo]

More or less the same story here (just finished a game with patch 'n'. Lots and lots of air units whizzing around, especially from Yang. And I think 3 factions build a single crawler. The only surviving one (Yang's, the other two I killed during some Vendetta) must have crawled a mine somewhere, it returned 4 minerals. Tech trading must have happened alot too, since most factions were pretty close in weapon level.
I'm wondering how the AI would react to more 'hardcoded' early units in the game, like a probe foil or supply trawler, in the alpha.txt units list.

My gut feeling is this revised AI is best for large(r) maps because of its tendency to leave new bases undefended.
Something else I noted was that Alien Artifacts weren't around that much. I catched a good amount of seapods and a fair amount of land pods on small islands, but none came out.

[kyrub]

I completed a game under 444(m).

Thanks for the reports!
It is good that you enjoyed the game. [m] version is cake due to the bug, fingers crossed for the [n] version, that should be quite different, significantly more effective from the AIs.

Themes:
- the crawlers: it is easy to let AIs make them more. I just saw a lot of examples of how badly they are used by the AIs, mining 1 mineral typically... Also, they are wiped away by worms. Not sure about it. On the other hand, I thought I saw AIs using them for the SPs. What do you think?
- air attacks, nice to hear the changes are visible.
- is AI more aggressive? This specifically should mean, that AI sometimes attacks with odds slightly out of favor. However, it is important that it avoids "dumb"attacks.
- did you see AIs attack out of base?
- did you witness AIs attack with a hurry (e.g. 2/3 strength, AI moving and attacking in the same turn)?
- undefended bases are a problem, I noted. Will se what can be done about it.
- SE changes, I will do them next. It will be done quite simply, but hopefully in an effective way. Right now, I am basig it on Kirov's post in the other thread, plus few notes that were there.

Keep the reports coming, any information is valuable for the assessment.

[Geo]

I did notice several air attacks on the same location a few times. Can't confirm if it was on the same unit or another in a stack.

I definitely would let the AI make more crawlers. At the very least it returns something before becoming worm bait. And it probably saves a combat unit from being badly damaged or destroyed in doing so.

[Kirov]

So how should we play it to test your changes in the most effective way? At first, under [m], I tried to be a peaceful Morgan and traded away any tech they asked for/demanded in order to facilitate their trade and growth. I understand now we should check out their battle movements more?

You mention popboom as well - in order to do it effectively, the AI must do better terraforming. It's too long waiting for Tree Farms or Cloning Vats, which they are not gonna get anyway. CV is a game-winning SP even in multiplayer. Any changes here?

I sure hope the AI will use crawlers to cash SPs. One of the problems I have in single player is that once I get PTS, I build crawlers everywhere for many turns and grab as many Projects as I can, usually almost all of them. It's kinda boring for a while, but allows me to win any game in the long run. In my [m] game Zak managed to get VW, which is not that bad. If they start to beeline to AI and cash crawlers by standard that would be a huge deal. But yeah, any use of crawlers, even suboptimum, is a massive change in my book.

[BlaneckW]

You couldn't program crawlers to stay one hex away from Fungus?

[Kirov]

You couldn't program crawlers to stay one hex away from Fungus?

I imagine that could potentially drive them away from juicy tiles, besides a bulk of native life comes as worms disembarking from IoDs, you can't do much to stay away from them.

[BlaneckW]

I imagine that could potentially drive them away from juicy tiles, besides a bulk of native life comes as worms disembarking from IoDs, you can't do much to stay away from them.

That depends on map-size and land-to-sea ratio.  The AI's crawlers are also more valuable, as the AI is less efficient in build-planning.  Anything that might conserve them to begin with would be valuable.  One can have more than one AI behavior tendency anyway, and switch between them.  I.E. start off more conservatively.

[kyrub]

I did notice several air attacks on the same location a few times. Can't confirm if it was on the same unit or another in a stack.

Interesting, but next time the details would make the info useful. AIs typically gang an attack on a city, like, an air raid.

I understand now we should check out their battle movements more?

Yes. I strictly tried to make AI less "cautious" when assessing its attacking options. It may be subtle, but should be visible. Like: an infantry unit moves along the road and actually attacks your 10-1-2 rover. Hopefully also: an infantry unit in a city attacks your stack of attackers and deals good damage to it, instead of sitting back doing nothing waiting to be slaughtered as a defender. Note: sometimes, the moves can be too risky, sub-optimal for the AI (because it is less cautious). I'd like to know if it feels like AI gained an edge, in general, or not. Or if it feels like nothing has changed.

Air attacks are a bit different story, AIs should avoid your formers (although there may be a sudden swipe as well). Air attacks are lot less cautious, maybe gung-ho a bit. AIs should try to air attack a base, and then drop-pod an invader - at least sometimes! AIs should counter info war units more as well.

In my [m] game Zak managed to get VW, which is not that bad. If they start to beeline to AI and cash crawlers by standard that would be a huge deal. But yeah, any use of crawlers, even suboptimum, is a massive change in my book.

All AIs, regardless of orientation, now understand better the value of a few key SPs. Virtual World, Hunter-seeker algo, Cloning Vats and a few others are top targets, they'll try to beat you to them and build them ASAP if they can. I'll try to rapidly increase amount of crawlers, we'll se the problems with their usage better.

(By the way, if you think we can achieve anything near the level of MP games, you should probably lower your expectations. Personnally, I restrict myself from using crawlers in single player games, to help AIs, as I am sure, more players do.)

You mention popboom as well - in order to do it effectively, the AI must do better terraforming.

The terraforming has a lot of room for improvements. But if CV makes the AI peak already with this level of terraforming, I don't see why pop-booming could not be a big improvement now.

Terraforming (higher level) can come back as a theme in future... Please understand, that this is a step by step work. Everything is inter-connected, only after you fix the problem on three fronts, it starts to work. Example: I can push the bomb-and-drop tactics more. But then I realize that without the AIs actually building lot more drop units, the tactics will almost never occur. So I do it. But then I realize that without the needlejets targetting something else than formers, I'll still never see a dropping unit take a base.

For me, this is about priorities and a my lack of RL time. Right now, I have SE on the platter. And ensure that FM does not ruin the AI (which is a huge task in itself). This, in my opinion, is the big fish at the moment, AIs are self-destroying with social engineering.  Occasional pop-booming is half done (with Creches). So I'll try to tie the ends, no?

[Nexii]

I haven't played with this recent AI (will this be put into Yitzi's patch?) but I have to agree terraforming is probably where the AI could stand to improve the most.  Improve that alone and a lot of their more extreme cheating bonuses could be toned down.  I've noted some factions like Spartans only build like 2-3 formers tops at any time.  AI never builds advanced terraforming from what I've seen (borehole,condensor,echelon).  I suspect this is because the AI doesn't pollute.  Maybe a middle ground (to prevent AI from flooding the planet) would be to just make AI suffer from the fungal pop portion.

SE choices, yea the AI doesn't always take ideal SEs.  Pop booming is important but I believe with AI cheating bonuses (++GROWTH) that even when they take some mild +GROWTH SEs and Creches they end up booming.  It's usually more due to lack of nutrients that they stop that explosive growth.  I'm not sure if SEs should be the biggest priority though - I'm sure there's some hard coding for them to prioritize their affinities? 

Use of military units would be second and probably a lot more difficult to code.  Not just suicide attacks but more importantly suicide zones should be considered.  For example AI will stop low defense units next to yours, and they die easily.

[kyrub]

Hi, Nexii,
I have a different vision of the game and the current state of the patch. Please, try the patch first.

[Edited]