Improvements for AI Opponents

NTJedi · #1 July 8th, 2008, 03:57 PM

From what I've seen the computer AI opponents haven't been improved in any patches despite having multiple problems. Some of these are probably difficult to fix yet I'm sure others are not so complex.

1) EASY FIX DIFFICULTY: Computer opponents sending pretenders into the arena death match. Pretender tag should make it easy to stop this unit from participating. It's so sad to see 10+ AI pretenders getting killed because the AI doesn't know better.

2) HARD FIX DIFFICULTY: Computer opponents starving its own troops. This causes the AI opponents to lose battles, even important battles. This would probably be more difficult to fix, yet at least the human player can set supplies to 300 as a semi-workaround.

3) HARD FIX DIFFICULTY: Computer opponents blindly drop items on mages/commanders. This causes AI opponents to severly hurt its own mages as it places full steel plate on a low hitpoint caster. It's rarely able to use important magic item boosters as well. This is another task which would probably be difficult to fix.

4) EASY FIX DIFFICULTY: Computer opponents lack knowledge of which globals it should NOT be casting. Globals such as Utterdark and Perpetual Storm will do more damage to the AI opponents 99% of the time. If given a high astral and blood pretender I worry whether it may even cast Astral Corruption which would totally devaste any AI opponent. It should be very easy to prevent AI opponents from casting specific spells since it's already done for spells such as Wish.

5) AVERAGE FIX DIFFICULTY: AI opponents choose wrong scales causing even an Impossible AI setting to choke and wither away. The higher difficult settings should include wiser scale/pretender builds. Poor scales and dominion has been known to cause AI opponents to die very very early... such as Mictlan. Some nations already have programmed scales such as Abysia for its heat so the difficulty is probably average.

Hopefully one of the DOM_3 patches will provide improvements for the AI opponents.

MaxWilson · #2 July 8th, 2008, 04:07 PM

Hopefully the current (secret) Illwinter project will turn out to be as much fun as Dom3.

KO is always hinting at the current project and I want to know what it is. Hopefully within a year or so.

-Max

HoneyBadger · #3 July 8th, 2008, 04:10 PM

Illwinter doesn't even need to improve the AI themselves, we just need the tools to program it. I'm completely confident that there are people in the forum who will positively salivate over the idea of improving the AI.

NTJedi · #4 July 8th, 2008, 04:16 PM

Quote:

HoneyBadger said:
Illwinter doesn't even need to improve the AI themselves, we just need the tools to program it. I'm completely confident that there are people in the forum who will positively salivate over the idea of improving the AI.

Yes a scriptable AI would be a huge improvement and it's worked wonderfully for many other games. Unfortunately we probably won't see this feature until DOM_4. With JK and KO working another project it may take 5 years or more before we see DOM_4, but I believe it will arrive. Stardock is releasing a fantasy TBS game during 2009, the developer responding to my email shows this game has a bright popular future.

Amhazair · #5 July 8th, 2008, 05:55 PM

Quote:

NTJedi said:
1) EASY FIX DIFFICULTY: Computer opponents sending pretenders into the arena death match. Pretender tag should make it easy to stop this unit from participating. It's so sad to see 10+ AI pretenders getting killed because the AI doesn't know better.

4) EASY FIX DIFFICULTY: Computer opponents lack knowledge of which globals it should NOT be casting. Globals such as Utterdark and Perpetual Storm will do more damage to the AI opponents 99% of the time. If given a high astral and blood pretender I worry whether it may even cast Astral Corruption which would totally devaste any AI opponent. It should be very easy to prevent AI opponents from casting specific spells since it's already done for spells such as Wish.

5) AVERAGE FIX DIFFICULTY: AI opponents choose wrong scales causing even an Impossible AI setting to choke and wither away. The higher difficult settings should include wiser scale/pretender builds. Poor scales and dominion has been known to cause AI opponents to die very very early... such as Mictlan. Some nations already have programmed scales such as Abysia for its heat so the difficulty is probably average.

While these 3 points do indeed (to my uninitiated mind at least) seem doable to change the programming for, I think it's a conscious design decision by devs not to do so. If I read their (well, KO's especially) various posts it seems to me they want the AI to do various random, not strategically optimal, even potentially harmful things, simply because the various pretender gods might feel it was fun to have horrors randomly popping up everywhere, or believed a productive society was the one and only road forward in this world, even if they happened to be ruling Mictlan.

I'm not saying you have to agree with this reasoning. I do even agree that these (comparatively) easy to implement changes would bring a welcome improvement to the challenge level of SP games. In the end though - at least if I interpret the devs correctly - these issues are about the very philosophy of the game, and as such transcend prosaic considerations such as AI competence. And I, for one, am very excited about the philosophy of the game. And while I would be very happy to see AI improvements, I am also more than willing to follow the devs overall philosophy of possibilities, options, and diversity, as this is one of the main reasons this game is so exceptional.

********* DISCLAIMER *********

I have no special connection in any way to the devs, and am most definitely not a mind reader, so the above just is my own guess of the devs point of view, cobbled together from their various posts over the last ~2 years.

If, in fact, my interpretation is drastically wrong, I appologise to the devs for being a socially challenged individual who isn't capable of understanding my fellow men.

******* END OF DISCLAIMER ********

Gandalf Parker · #6 July 8th, 2008, 06:15 PM

Quote:

HoneyBadger said:
Illwinter doesn't even need to improve the AI themselves, we just need the tools to program it. I'm completely confident that there are people in the forum who will positively salivate over the idea of improving the AI.

We would like to think so but many offers have been made for AI improvement projects by "the AI sucks" experts. Few have ever bothered. They mostly just want Johan to fix it.

On lists of AI fixes about half tend to be "we can do that now", then theres a bunch of "can you lay that out more exactly and maybe Johan can look at it", some that are IMHO not even good things to put into the game, and then a few possibly workable good ideas.

Maybe the problem is that Im not sure what list such things should go on to get Johans attention. Maybe the AI needs its on wishing shortlist.

MaxWilson · #7 July 8th, 2008, 11:35 PM

Quote:

Gandalf Parker said:
We would like to think so but many offers have been made for AI improvement projects by "the AI sucks" experts. Few have ever bothered. They mostly just want Johan to fix it.

Gandalf,

What does this mean? I've never seen Illwinter offer to let anybody else touch the AI. Are you talking about the map-editing projects like SemiRandom, for giving AI pretenders better scales?

-Max

Gandalf Parker · #8 July 9th, 2008, 12:19 AM

Well that would be one.

Also calls to lay out specific AI instructions the way its in the game. We gave info on looking at the debug code, then giving specific suggestions. The AI is linear and general so most peoples suggestions tended to be too specific. Altho I do remember one that gave very specific suggestions for better castle building which seems to have been put in. Also some suggested changes to spell sequence, and unit building.

MaxWilson · #9 July 9th, 2008, 12:23 AM

Hmmm. Did these discussions take place on the mod forum or here in the main forum? If the mod forum I must have missed them.

I've been thinking lately that the AI seems to be better than it used to be, and I wonder if this is why.

-Max

Loren · #10 July 9th, 2008, 12:41 AM

*PART* of #3 could be implemented easily:

Check to see if the commander will benefit from the item. When I take an air-breathing item off an air-breathing commander, something's wrong.

I would also divide all commanders into caster or non-caster. Items would be flagged as to whether they are for casters/non-casters or both.