Note from MM - AI MODDERS MUST READ THIS - Page 6

dmm · #51 January 16th, 2001, 07:39 PM

Besides memory, the AI needs to have at least one turn of forethought. (Even the stupidest chess program has this.) The AI needs to look at known enemy forces (at least forces in the same quadrant) and ask itself, "What would I do next turn, if I were him?" And then it ought to base its moves on that. Also, at present it perhaps is seeking to maximize its score. (At least I hope it is!) In addition, however, it needs to seek to lower enemy scores (e.g., by destroying undefended enemy colonies). At present, it has been observed to leave enemy colonies unmolested if it hasn't claimed the system, even when it is at war with that player. (Maybe the AI ships didn't have supplies or were desperately needed elsewhere. But I doubt it.)

Nyx · #52 January 16th, 2001, 07:39 PM

quote:
inconsequence and lack of force concentration ARE the main things everybody is complaining about. I mean, all the AI's logistic errors hurt them, but these are what makes them victims instead of adversaries

My biggest complaint is that war with the AI is always to the death. There is no "limited conflict." It is just like in Civilization where you get into a war in 2000 B.C. and when it rolls around to A.D. 1790 you're still in that same Da$#$% war with that same player. Who the hell ever heard of a 4,000 year war?!?! Why is this the exact model used by every single strategy game ever made? When the AI declares war it should have clearly defined goals. Drop them into a temporary text file or something. Things like, "Capture planets x, y, and z; and kill all Drushocks ships in system A." When the goals are met, it offers a non-aggression treaty. Similarly, it should have clearly defined, "Well, I lost this one" settings. Something like, "If I've lost 2 entire systems and not defeated a single of my enemy's fleets... Then offer protectorate status or tribute plus non-interaction treaty."

Secondarily, I'd like to see a setting where the AI doesn't just get angry, but gets submissive. After a point, the AI should get afraid of you, not just angry at you. It should actively seek out allies before attacking you suicidally.

Easy and very important: The AI should get cumulatively less likely to declare war the more wars it is already involved in. So it should be way, way less likely to declare war on a third empire if it is already at war with two. When it begins losing battles, it should start getting out of the multi-front wars and try to concentrate on killing the weakest of its enemies.

There needs to be some way to get the AI to give up a claim on a system. Right now once they claim something, they never give up trying to get it.

------------------
Compete in the Space Empires IV World Championship at www.twingalaxies.com.

dmm · #53 January 16th, 2001, 07:49 PM

More on forethought: Besides asking itself about the other player, it might start by asking itself about itself.
Examples:
What would happen to these ships if I moved them through that unstable warp point with a black hole on the other side?
What would happen if I attacked that big fleet with my one ship?

Also on forethought: It needs this in tactical (battles) AND strategic.

Lastly,
this is probably a lot harder to talk about than to do. But scripting has its limits. In an unscripted situation, the AI needs to have some way of optimizing the outcome. I think that this is the basic weakness of the AI as it is now.

DirkHowitzer · #54 January 16th, 2001, 07:58 PM

I don't know if anyone can answer this, perhaps Mephisto or one of the other big modders...How difficult, realistically, will it be for MM to enact the changes that've been suggested here? I mean it seems that were going beyond simple editing of txt files into unexplorered territory. I mean all of these suggestions are terrific but were talking turning the AI into something near perfect and almost humanlike. Is it possible at this junture?

------------------

"He's dead, Jim."-- Lt. Commander Leonard "Bones" McCoy |Chief Medical Officer / USS Enterprise (NCC-1701)

[This message has been edited by DirkHowitzer (edited 16 January 2001).]

KiloOhm · #55 January 16th, 2001, 08:00 PM

One thing I think the AI should do is multiple attacks with the same fleet per turn (as a human does). If he attacks my planet and can't quite defeat it in one attack AND he has additional movement points after the attack, he should attack again if he can do more damage or win.

------------------
Regards,
KiloOhm

Emperor Zodd · #56 January 16th, 2001, 08:05 PM

Absolutely! Tell MM!

Nyx · #57 January 16th, 2001, 11:59 PM

quote:
How difficult, realistically, will it be for MM to enact the changes that've been suggested here? I mean it seems that were going beyond simple editing of txt files into unexplorered territory. I mean all of these suggestions are terrific but were talking turning the AI into something near perfect and almost humanlike. Is it possible at this junture?

The vast majority of the suggestions in this list can be scripted as if-then statements. The problem is figuring out how to phrase them, how much this will slow the AI's turn, and making sure they don't conflict. Meaning most of what's here should be doable. Some of it will take more than a month, and/or one or two more patches worth of tweaking/play balance testing. But Aaron should be able to handle the majority of what people have suggested. Judging from how well the AI works already if you just fix the text files, and how well it worked in SE3, he does know how to script a dangerous AI. I have faith in him.

It would be nice if the people modding the AI scripts would continue to share their discoveries with him so he can keep abreast of what we all discover.

------------------
Compete in the Space Empires IV World Championship at www.twingalaxies.com.