Dominions 2. AI. Suggestions, that how to fix it.

[Gandalf Parker]

originally posted by cbeller

Anyways, there is my report for this game....what i have seen so far....I have come across several times 20+ armies on my borders, but they pull their troops elsewhere...possibly to fight on another border with another nation? IT would seem....cause the AI's are losing/gaining provinces quite a bit...

Ive seen that. Im just guessing but it might be a simple way of providing some "smarts" to the AI. He moves the big army around inside his borders. If he kept it still Id be able to plan better. And many times Ive moved on a province of his that I thought was clear only to meet his big army head on.

It would probably work better if he had more armies, a variety. Maybe kept up with the number of provinces he had such as armies=provinces/5

[licker]

Originally posted by Gandalf Parker:

quote:

---

originally posted by licker
I don't see why this wouldn't work for Dom as well. You'd have a list of commander types (maybe also broken down into magic types for different mages) and a list of unit types (classes rather than specific units so indies fit in). Then a weight is assigned to each type, a crude example for just infantry could be:
LI: 25%
MI: 50%
HI: 25%
So the AI would look at its current army and see that it had only 10% LI, then it would build more LI to get the ratios set. Something like that should be made external so that modders can tweek away to their hearts content. In fact you can use lists like that to control many aspects of AI choices, you can do it for magic, artificing, recruiting, summoning, ... and have master matricies to control the embedded ones...

Seems simple and workable. On the other hand, lets pick one race and try to write it all out. Ulm should be simplest. It has a cheap and expensive Version of each of its units which is nice. But I suspect that anything we come up with to "fix" this will amount to fixing it for the indep 2 games people are playing (fast player conflict) and will need fixing again for 4, 6, 8. Then for levels of magic research, and levels of resources, etc. Im thinking that this level of "thinking" might end up being equal in size to the game now (Ive run into this problem before).

---

I don't know if its as large of a task as you make it out to be

There would be a series of matricies, and it would be a bear to optimize them, but if they are moddable... then we can count on the relentless fans tweeking their fav nations to get them to be kick *** wicked

As to the differences in game settings... I agree completely that these need to be factored in, but I don't think its that hard to do. You have a matrix for the game settings with 'correction' values to the national matricies. Or you may have a game setting matrix for each nation (though some nations are similar enough...).

What I think we would find if we did this is that there are indeed preferable units for nearly all situations, especailly when we are dealing with shades of grey in Ulm Infantry, in fact this would give the AI a boost as it wouldn't be tempted to use inferior units (slight though the difference is) as a human would be.

I'd still like to hear from the devs is such a system is at all possible before I devote any more time to specifics. Though if others want to take a nations unit roster and break it down go for it. Remember that the %s should differ depending on what the goal of the nation is (magic, might, dominion, ...) also modified by the available richness of provinces, income, ... its alot of tables, or inputs, but its not that hard to do (I think )

[Gandalf Parker]

I would love to see a scripting language for AIs. Ive said since before Dom 1 that games like this should have one but its probably just my MUD background talking.

It would be great to have player-made scripts. For some that would mean that the great players could create smart scripts for their favorite race so that we solo players could play "them" without playing MP. Or at least get a feel for how we would do against their tactics (a script cant be as good as a player). Of course some scripts would suck but for me (Fanatic of Random) that would be fine also. We could have scripts for playing styles of barbaric horde, home defender, slow careful expander, diplomatic, backstabber, aggressive expander, comes in waves, elite units only, etc etc etc. Then I could write a script to randomly select from the pile for each game (I love it). Player-made scripts will always allow far more variety than one set of programmers straining to think in styles different than they wouild play themselves. OR at least (as it has in some other games) it lets them say "we gave the players the means and they didnt do anything with it".

The problem is that the variables and hooks needed for such scripting to work usually has to be written into the game from the beginning.

[ October 30, 2003, 12:14: Message edited by: Gandalf Parker ]

[Gandalf Parker]

The area of unit selection is really hard to move from complaining to real suggestions. To keep a database of each race and each gaming situation would be extreme. They might as well move the AI to a seperate program so it doesnt slow down the MP play. Many of the people complaining about the AI are playing with independents set at 2 or 3. The strategies and units for such a game would be very different than if it were set in the middle, and that would be different than the high setting. And thats just ONE of the pre-game settings.

OK Ive heard that the AI should not buy cheap units but instead buy a "real army". So insteadof figuring out how to get the most units with the least upkeep (patrolling strtegy) they should buy the most expensive? That would lead to armys of hvy cavalry, which I guess isnt too bad.

Personally I think an AI best bet is to buy everything. Start at the cheapest and buy 1, move up thru the choices then start over. They would reach a point of not being able to buy another hvy cavalry or elephant or hydra and would use up the left over amount on cheap troops so they would end up with slightly more lesser troops than expensive ones. But that wouldnt be too bad.

What about leaders? How do you make sure the AI buys a mix of leaders? It has no memory so it cant say "I bought this Last time and now need to buy this". Hmmmm well actually it might be able to fake that. If turn number divides by 5 evenly then buy the most expensive leader THEN buy troops. *sigh* still not great but maybe it would be better?

[Aristoteles]

I like the idea of this kind of scripting.

Another problem: The AI is sending in his pretenders to fight in the Arena. I think it should send in commanders instead of pretenders.

Forts and castles: I sorta agree with this, in the demo the AI didnt built a single one. Maybe 1 or 2 in like 6 games.

Strategic AI: I am sure that it can be improved but it will need precise scripting, also propably lot of time.

[DominionsFan]

Yeap. It seems, that the Dominions 2 AI love to use these worthless units.
Ok they are not worthless, but without good units and monsters, those armies are very very very weak, compared to the elite human armies.

[licker]

I think its hard to tell from the demo how effective the AI is with units. The game settings do make alot of difference. I still think the discussion and suggestions in this thread are valuable, and I wish a dev would deign to let us know how things are currently set up.

I had some time to fool around Last night and I was playing with Indies at 6 or 7 (can't remeber exactly) with all nations on normal or difficult on the bigger map. Anyway, by turn 10 I had expanded or scouted to other nations borders (not surprising as the map was crowded), by turn 15 I had 4 nations declare war on me, though only one was actually adjacent to me. I routinely saw armies of 100+ units moving around though, and I managed to catch Marginon's big force with mine (playing Pythem) I slaughtered them, mostly because I had sent in an assassin who picked off a couple of mages (well inquisitors... you know...) and my Arch Therug with 3 slaves doled out some nastyness...

Anyway, a few turns later I regrouped and reenforced and set seige to their capital. The next turn I had to repell a relief army with 8 summer lions and assorted pikeneers and x-bows. I beat them back, but it wasn't clean... So the AI does seem capeable of using summons, and it will make bigger armies.

Part of the problem some of you are encountering is probably due to not having enough opponents on a too big map. With a cramped map the AI is forced to consolodate more (though it could still probably use some tweeks to the units it recruits). In the next 20 turns of this game it will be interesting to see what the other nations who declared war on me do, 2 more of them now have provinces boarding mine, and if I can't finish of Marg quickly I may have to recall that beseiging army to deal with other threats. All in all this game has been entertaining, at least for the short duration it's Lasted.

Again though, there is alway room for improvement of the AI, I hope that the discussion in this thread is not being ignored

[-Storm-]

I am sure that we are NOT ignored, but IW tends to don't reply usually. It is all cool, it is good enough, if they are READING our suggestions at least.


I have a very good idea. Well hopefully it is a very good idea.

So we all know about the debug mode, and that we can pass the control to the AI.
What about adding a special debug mode.

Special debug mode:
((
1. make a god
2. set up a game
3. let the AI control our nation
))

-->>
We will be able to see the actions of the AI. That would be awesome. Than, it would be lot easier to post suggestions about the AI. {{Now, after you've gave the control to the AI, the game will quit.}}

What do you think? [I hope that this is possible.]

[ October 31, 2003, 15:26: Message edited by: -Storm- ]

[Mortifer]

Man, that is an excellent idea! If we could see the "moves" made by the AI, after each and every turn....that would help hella lot!
Seriously I love this idea.

[licker]

It is a good suggestion, as would be some form of autoplay, where the Comp would run for a set number of turns and then allow the user to view the contents of every province.

That way you could set different nations to different levels and get a better feel for how well each nation is tuned rather than just watching one nation at a time.