AI concept of the '80ies ... why?

[thejeff]

Note that almost all the current turn processing time is battle calculation. Very little is spent on generating the AI orders. Maybe 10%?

I suspect most of the AI improvements fall more into the category of lots of programming time, rather than calculation time.

A lot of simple things could improve the AI difficulty drastically. Better pretender design being high on the list. Better use of sacred troops. (Buys expensive sacred troops, even when the pretender has no bless. Takes a bless not suited to the units. Gets a good bless and then doesn't send a priest into battle to bless the troops.) Some ability to make basic SCs.
I doubt fixing any of that would require much calculation time, just better design.

[Hullu]

The problem with AI is definitely design, not calculating time.

If time was the problem they would use multithreaded approach, I'm pretty certain of it.

You have a pretty good idea how AI works, in theory. Putting what is said in this thread into reality is EXTREMELY DIFFICULT. Creating even a semi-intelligent AI for a strategy game with any complexity is _EXTREMELY_ hard. Or, why do you think no one has yet succeeded? I can't think of a single game, no matter the budget(Civ4 had a huge one) that would be the least bit smart.

It's VERY hard to code intuition and judgement for computers, and that's what you need for strategy games. Also learning is very hard to do. It's easy to tell the computer what YOU have already learnt, but it's hard to make the computer learn from it's own experiences.

It's just, hard.

[B0rsuk]

One problem with creating good AI is that humans operate on very high-level concepts. Computers don't understand very high-level concepts. Computers merely do exactly what they're told. You have to struggle hard to make it seem computer produces random numbers, for example.
--------
I'd prefer something more realistic, and within the reach. I don't have the decades of spare time to wait for someone develop good AI, and someone else design powerful enough and affordable processors.

For a start, let's forbid AI doing obviously stupid moves. The first thing would be forcing AI scales into (-1, 3) range except for races with preferred cold/heat values and bad scale tolerance (Ulm, Pangaea, mid-late Ermor, Abysia, Atlantis(cold) etc)
Very bad scales on ordinary nations can be worth it, but NOT if the nation is controlled by AI. Let's not pretend AI understands consequences of very bad scales. Simply lock it down to (-1, 3).

[Esben Mose Hansen]

Personally, I would love a way to create a script-client so we could write our owns AIs. I think I'd take a shot at doing it as a LP or maybe a MIP problem

[PhilD]

Esben Mose Hansen said:
Personally, I would love a way to create a script-client so we could write our owns AIs. I think I'd take a shot at doing it as a LP or maybe a MIP problem

Yeah, but scriptable in-game is way too dangerous - either the scripting language would be extremely limited, or 99.99% of scripts would just crash the game 99.99% of the time (though, Neverwinter Nights' toolset scripting language seemed to be a pretty decent sublanguage of C, and that didn't seem to cause too many stability problems; but then, the development team of NWN was orders of magnitude larger than that of Dom...)

What could work would be if the turn file format were more open, so that players could have an external program read an AI's turn, and generate orders. There would be some minor cheating issues (need for a way to have AI nations' files either in encrypted format, to prevent cheating, or unencrypted, to allow AI scripting), and some safeguards to prevent buggy programs from crashing the game through buggy orders files (simply parse the file, and if anything goes wrong, ignore it and have the in-game AI take over), but it would make it possible for all those "how hard can it be to write decent AI, this game's AI sucks" sayers to prove how much better they really are...

But then, this was a suggestion we made back in the days of Dom2, and the devs weren't too receptive.

[GwyrgynBlood]

Scripting languages for AI aren't necessarily hard to write... just depends on how the AI was setup. Kohan series has always had scriptable AI, TA Spring also has it.

Most AIs in strategy games have their AI scripted in some way, even if it's internal. For example, in your average RTS game, the AI generally gets a basic list of what buildings they need to build what units, and then a basic understanding of how many buildings of each type they should have, and then an idea of what relative number of units to produce. Also things like 'how many units should I stock up before sending out an attack force'.

Of course, those AIs are generally very dumb unless a LOT of effort is put into them.


Best example of 'good' AI I've seen is Virtual Fighter 4. Very good AI that is modeled after real players.

[HoneyBadger]

I think atleast we ought to be able to have the computer coming up with good Pretender designs. It should be fairly easy to do this without the use of any true AI at all. Just come up with several designs for each and every pretender available per nation. Preferrably, these designs would be influenced by the forum community, because inevitably, it takes a village to come up with truly dangerous strategies. Based on the design the computer randomly chose, the AI could then be influenced to follow one of several basic strategies, such as from building a lot of sacreds to not building any sacreds at all, depending on the bless effect and the specific nation the computer was playing. It would require a significant effort on the part of the programmers, but not much computer time at all, and if done right, should certainly improve the AI. By the way, has anyone noticed whether or not the AI in the game makes better decisions (or seems to) on higher difficulty levels? I'm still learning the finer points of the game, so I mostly stay on Normal setting.

[Hullu]

You have to struggle hard to make it seem computer produces random numbers, for example.

No matter how much you struggly, computer does not generate random numbers (except perhaps with an outside source of randomness, which also most likely is just pseudorandomness

Yea, offtopic, I know, just had to say it!

[Gandalf Parker]

I would love scriptable computer opponents but the handles needed for that have to be written in from the beginning.

And yes, as its been stated, "human" computer players are hard. Artificial Intelligence (AI) is actually easy, but artificially human (AH) is much harder. Its a common subject in the AI newsgroups.

An AI would figure a path from A to B as the strightest and easiest route. AI is simple and gets done often in programming. But always taking the straight line in a game would quickly be abused by human players. To be more human would have to include concepts like flanking, sneaking around behind him, multiple fronts, etc etc.

The closer to AH that programming gets, the more obvious it is that the easy route is just to add more randoms. Now it becomes deciding whether you can get closer to the goal by starting with AI and adding some randoms, or starting with totally random actions and weeding out the most extreme ones.

I do agree that this game could make good use of "thinking" time. As a hosted game it wouldnt bother me if it took hours to think out its moves. As a solo play it might be able to use spare CPU cycles while Im doing my turn but games that do that tend to become games that take over your machine. I dont think Id like that. I like Dom because it lets me keep all my work windows open and skip back-and-forth

[Endoperez]

I was surprised, yesterday, to see AI using Terminus the mercenarey assassin to conquer my weakky-defended provinces. The first one had PD of just 1 or 2, and he killed the Markata and the Bandar Commander pretty easily. In the next battle, one of the markata happened to score 7 damage to him in the very beginning, and he died. Still, I didn't think AI would consider attacking with a lone commander.

It's still far from being able to build thugs or combatants, but not quite as far as I had thought.