Improving AI - what are the priorities?

[Nevill]

There seems to be something affecting this type of tests with scenario editor. If the normally built AI units don't act like this, it's not probably a bug, no? Maybe the problem is that these scenario created units have not got the right unit plan, as you suggested. This messes AI up, obviously. I cannot blame it.

The first logical step would be checking up which plan these units were assigned with. There are only a dozen of them, and I think we can exclude non-military ones. Whatever the plan seems to be, the units behave erratically. If it is Offencive/Combat, then why do they not attack bases? If it is Defensive, then what are they doing in my territory? If it is set to Reconnaisance, then why do Scout Patrols (that have the same plan assigned) function correctly?

Another thing that I want to point out is that the units that are built through conventional means do behave the same when it comes to 'hasted' attacks. If, however, you will only accept a 'clean' sample, I can arrange that, though obviously I would be limited in what I can test.

And 'the AI gets things right eventually' does not mean that 'normally built AI units don't act like this'. It is only natural to see units with a correct plan attacking my bases, but that says nothing about the number of units that do not have the plan set correctly (and they exist, as indicated by 6-1-1 normally built  'defenders' from a previous SMAX save, which I haven't reproduced in SMAC yet).

Why do you think its plans make little sense? It is the mechanics AI uses, indeed.

Because I have no idea what the AI was trying to do. It is hard to explain things when I myself do not understand them.

[kyrub]

If it is set to Reconnaisance, then why do Scout Patrols (that have the same plan assigned) function correctly?

What do you mean, correctly? Do Scout patrols ever attack? I cannot remind me of one such occurence.

It is only natural to see units with a correct plan attacking my bases, but that says nothing about the number of units that do not have the plan set correctly (and they exist, as indicated by 6-1-1 normally built  'defenders' from a previous SMAX save, which I haven't reproduced in SMAC yet).

These are "emergency defenders", I think. Ai has such a routine inside and it seems very normal to me. It's quite obvious, when there are too few defenders, let others defend. These units, however, are not able to counter-attack. This, I hope, can be changed.

I must say I am having very hard time to find the location of unit data in the saves. I need to understand which unit_plan do they use and change it. With this, your saves are good material and can be used for my testing.

[Yitzi]

I don't think this is particularly a priority, but I've found that while the moddable psi offense/defense ratio for land finds 18 different mentions in the code, the ones for sea and air use only two of them and no others.  Presumably those two are the ones actually in combat, and the others are for the AI, implying that when the AI decides how to deal with psi, it only looks at the land ratio.  Of course that's SMAX, so SMAC might be different, but it's still something worth noting.

[illumined]

Great project, I've been off and on following some of Kyub's work with Alpha Centauri and Master of Orion and I'm always impressed by it. Recently I decided to fire up SMAX once again use the new Yitzi 1.3 patch, playing the standard Huge Map of Planet on librarian. I played Cybernetic in place of the University, with all the other factions being from the original. I ended up starting near the planetneck region, and after expanding to most of the southern portion of both continents (we really should give these continents names) I met the Hive, and as it turns out above him in the Upland Wastes and Eurytion Bay region was the Peacekeepers. The Hive went to war fairly early and wiped them out completely, and of course went to war with me soon after for believing in democracy. Here's what I noticed about his behavior:

-His offensives just didn't seem to pack much of a punch. Periodically he would send a few infantry units and maybe a rover or two over the border, but these were just lambs to the slaughter. With taking little to no losses of my own I was able to wipe them out with my speeders. Perhaps what would make it better is if he focused his attacks on one or two bases instead of trying to fight across the whole front (which is quite lengthy) all at once. Another strategy is to have a small "micro-offensive" in one part of the front as a diversion and then hit hard somewhere else. 

-His production didn't seem to be very high. One thing that would definately solve this is if the AI used more thermal boreholes, so far I haven't seen a single one besides the ones in the Borehole Cluster that was already on the map. It would hugely enhance his productivity if he had these, like me. I have the Empath Guild so I can always see how many units he has, his ground units were not much more than mine, at least not during round one of the war.

-His Air Force was practically nonexistent. At any given time he only had 4 or 5 aircraft, all of which were bombers. This is really surprising, before when I played previously using the vanilla build they usually have fairly large amounts of aircraft, both fighters and bombers, I don't think I've seen this before. Looking at other factions, the one that has been the most successful has been the Spartans who now control the continent that has the Freshwater Sea in its entirety, yet at this point in the game they only have 7 aircraft. Anyway, towards the end of round 1 of my war with the Hive I was pounding away at his positions and causing major casualties with my bombers, and despite that he did not produce a single interceptor. It used to be that the AI would build a large amount of fighters under this circumstance, I used to have major Battle of Britain types of air duels with them.

-He isn't really investing in his navy. In this situation that's a good thing because he's fighting a large scale war across two continents so he should be focusing on ground and air. What doesn't make sense is that even though the Spartans are now totally alone in their continent they have almost no navy. This is not good not only because it impairs her ability to defend her continent, it also makes it very difficult to wage an effective campaign to finish off Gaians and Believers by taking over the small smattering of sea bases they have left. She also recently declared war on me which is impossible for her to prosecute effectively since my navy is so much better than hers and the AI is evidently not capable of launching a major transcontinental invasion. While I do believe that development priority should be on the AI's terrain improvements (which do appear to be better than before), ground game and air force, it would be nice to eventually have it play an effective naval game. Civ 4 actually did a pretty good job with this.

-He isn't taking an easy island expansion. Ok, the war hasn't gone well for the Chairman and he is slowly but surely being driven back and has  lost several bases. But nearby to the northwest of his position is the Isle of Dierneda (or whatever it's called). There's a couple of small Believer seabases on its coast, but aside from that it's totally uninhabited and ripe for the picking. While making a minimum invest of a transport and a handful of colony pods he could quite easily establish several bases there, build them up, and then have them start cranking out combat units to reinforce his position on the mainland. This should be a viable strategy, even with having almost no naval presence the transport distance is very short, so protection shouldn't be a big problem.

[Buster's Uncle]

You know, Yang's what wrestlers call a jobber.   He builds a bunch of puny bases close together and gives you someone to fight, but rarely makes for much of an actual challenge.  Morgan's usually like that, too.

Welcome to AC2, BTW, illumined.  How did you find us?

[illumined]

Heh, well in fairness Yang already managed to take out the Peacekeepers, and based on the number of bases they had that was no easy feat. His production bonus should give him more of a leg up in units than he actually has. After he took so many losses he called for a peace treaty (it was good that he recognized he was beaten) and, feeling my army was a bit overextended, accepted. I then was going to transition back to a free market economy (which didn't happen) and I expected he would build up a big offensively oriented army with lots of speeders, fighters and bombers. That didn't happen and not too long after the peace treaty he declared war again, and the initial offensive this time was seriously underwhelming, so much less capable than the last time. So, I've started rebuilding my bomber corp to go on the attack again, I've already succeeded in taking one or two of his bases. But still, shouldn't he have waited until he had a bigger edge to launch his attack?


So here's how I found you guys. I have a library of old games that I periodically cycle through, whatever I'm in the mood for. I was playing through Master of Orion 1 and discovered Kyrub had made an official patch for it on Realms Beyond. Eventually I found out he also made a patch for SMAX on Aptylon. Recently I started playing SMAX again, and found a link on that thread saying to come here since he doesn't go to Aptylon very much anymore. So, here I am.

[Buster's Uncle]

Well have a good look around and don't be afraid to speak up; we'll take good care of you.

[illumined]

Well have a good look around and don't be afraid to speak up; we'll take good care of you.

That sounds ominous.  I also forget to mention, the aircraft refueling bug is still present.

[Buster's Uncle]

Yeah; I get my kicks freaking people out with kind treatment...

[illumined]

Yeah; I get my kicks freaking people out with kind treatment...

I was just pulling your leg.


So now I'm a little farther along in round two, Yang finally started building his first interceptor, but its at a sea base with no minerals. He's finally built his first bunker, right on the front which is good, but so far has left it unmanned and open to be occupied. I'm waiting to see what he'll do next on this front while I attack in Eurytion sector. He did make another small attempt at an invasion around Mesa and Hive sectors, but once again it was easily repelled by my speeders. One thing I like is his increased use of artillery to destroy enhancements on my side of the border, it's a reasonable move, especially destroying roads leading to his bases to slow down any future attempts.

[Buster's Uncle]

Hey kyrub - is there anything we can do to help?

[kyrub]

Hey, BUncle, it's nice you've noticed me.

What can you do to help? Please wait a bit more, I am almost done.
At the moment I am consolidating what I've analyzed and putting it all in the patch. It looks good on paper, but let's wait and see how it works in game. 7 points from the original 10-point shortlist (see 1st page of the thread) have been adressed with two more issues pending to be finished. The only exception is, sadly, the smart probe movement, as I don't seem to find the way how to change it. Pity.

Looked from the other side, there seems to be massive room for improvement in all other areas. AI air attacks have received major boost, no more former targetting, AI should attack with hurry and it could/should even counter-attack from its bases. I have even found (to my surprise) a piece of code suggesting AI was taught to use "bomb and drop" technique - big thanks to the guy who mentioned it in the thread! AIs will trade technologies freely and build lot more of buildings. Again, let's see how it works.

I'll need help later, with testing and feedback. And more patience before the result is put into SMAX version.

[Dio]

Hey, BUncle, it's nice you've noticed me.

What can you do to help? Please wait a bit more, I am almost done.
At the moment I am consolidating what I've analyzed and putting it all in the patch. It looks good on paper, but let's wait and see how it works in game. 7 points from the original 10-point shortlist (see 1st page of the thread) have been adressed with two more issues pending to be finished. The only exception is, sadly, the smart probe movement, as I don't seem to find the way how to change it. Pity.

Looked from the other side, there seems to be massive room for improvement in all other areas. AI air attacks have received major boost, no more former targetting, AI should attack with hurry and it could/should even counter-attack from its bases. I have even found (to my surprise) a piece of code suggesting AI was taught to use "bomb and drop" technique - big thanks to the guy who mentioned it in the thread! AIs will trade technologies freely and build lot more of buildings. Again, let's see how it works.

I'll need help later, with testing and feedback. And more patience before the result is put into SMAX version.

I to can be patient. Also, it is nice to see you active .

[Yitzi]

When you have stuff, please let me know so that I can incorporate it into my patch as well.

(By the way, your stockpile energy fix seems to have a few associated problems, most notably that when you change production, even to the same thing, it deletes your production gained that turn, so I'm redoing the whole thing.)

[kyrub]

When you have stuff, please let me know so that I can incorporate it into my patch as well.

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

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.