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

Gandalf Parker · #1 November 13th, 2003, 03:05 PM

How about something simpler that would be "usually good enough". Such as, for every X number of provinces (5?)you own you should have a castle. Find a province with no castle, no neighboring castle, and no neighbor owned by another player, then build there. Maybe it should also have a minimum number of neighbors. 3? 4? 5?

And what about the cost? Should it be "do this before all recruiting"? I think that has worked for me (in fact Ive come to realize that I play rather like an AI). Doing it that way would tend to automatically provide some controlling limits since the AI would rarely (should rarely if other formulas are done right) be able to afford a castle at the beginning of a turn without saving up for one.

So the AI sees that it has 10 provinces and only 1 castle. If it finds a spot where a province has at least 4 neighbors with no castles, and none of them owned by another player, and IF it has enough money, then it builds a castle there. Oh yeah, and increase the defence (I always do that where I have a castle)

OH wait a minute. That would be better if it came first. In a province it feels is going to be a castle, it should raise defence to (21?) first and then build a castle. Hm that could get expensive.

OK stop me. I think too much in the mornings.

DominionsFan · #2 November 13th, 2003, 06:26 PM

Quote:

Originally posted by Gandalf Parker:
How about something simpler that would be "usually good enough". Such as, for every X number of provinces (5?)you own you should have a castle. Find a province with no castle, no neighboring castle, and no neighbor owned by another player, then build there. Maybe it should also have a minimum number of neighbors. 3? 4? 5?

And what about the cost? Should it be "do this before all recruiting"? I think that has worked for me (in fact Ive come to realize that I play rather like an AI). Doing it that way would tend to automatically provide some controlling limits since the AI would rarely (should rarely if other formulas are done right) be able to afford a castle at the beginning of a turn without saving up for one.

So the AI sees that it has 10 provinces and only 1 castle. If it finds a spot where a province has at least 4 neighbors with no castles, and none of them owned by another player, and IF it has enough money, then it builds a castle there. Oh yeah, and increase the defence (I always do that where I have a castle)

OH wait a minute. That would be better if it came first. In a province it feels is going to be a castle, it should raise defence to (21?) first and then build a castle. Hm that could get expensive.

OK stop me. I think too much in the mornings.

The AI should build 1 castle in every fourth province.

licker · #3 November 13th, 2003, 06:52 PM

1 per 4? Why? Should there be a distinction between land and water provinces? Should the number of forts be dependant on the number of players in the game? The size of the map? The cost of the fort?

Its far too easy to just say 'build 1 per X'. Of course doing that may be better than what there currently is, but I'd hope that the analysis of this question would try to incorperate more variables that are easily determined by the AI.

It shouldn't be that hard to find choke points anyway, in fact the simplest way to do it is for the map maker to designate certain provinces as choke points and let the AI work with that information. However, I doubt that is currently supported in the game

[ November 13, 2003, 18:57: Message edited by: licker ]

#4 November 13th, 2003, 08:43 PM

You could have the AI build a fort if there is a X amount of resources present or if those resources are not present within a 6 Province line then put it in one with most resources that is not adjacent to an enemy.

Also the behavior of an army to build a castle, are they going to take their main force and have them sit there while it's building or just place a small token force (with maybe some Provencials) and have the army more strategically placed.

MStavros · #5 November 13th, 2003, 11:34 PM

Quote:

Originally posted by licker:
1 per 4? Why? Should there be a distinction between land and water provinces? Should the number of forts be dependant on the number of players in the game? The size of the map? The cost of the fort?

Its far too easy to just say 'build 1 per X'. Of course doing that may be better than what there currently is, but I'd hope that the analysis of this question would try to incorperate more variables that are easily determined by the AI.

It shouldn't be that hard to find choke points anyway, in fact the simplest way to do it is for the map maker to designate certain provinces as choke points and let the AI work with that information. However, I doubt that is currently supported in the game

Yes it is lot more complicated, haha.
However, the size of the map really matters?
Right now the AI won't build a fort almost ever.
We should post a nice detailed list that when should the AI build forts.

#6 November 14th, 2003, 12:00 AM

Maybe we should tell them when we build Fortresses of what type and why.

Then they could code the behavior of the AI to that.

I'm sure the MP guru's who play pure efficency vs themic could come up with some pure formula/reasons to it.

Like, how often in MP do you build a fort turn 10 or below?

Never?

How often in MP do you build a castle between turn 11-15? If so, why and where?

Questions like that.

Kaljamaha · #7 November 14th, 2003, 01:26 AM

Another thing that I think should be looked at is AI dominion design. Now, granted, I'm no pro, but sometimes the AI makes design decisions that are utterly suicidal. It has effectively lost before the game even began. One example, from Dom I would be Marignon with Turmoil 2, Cold 3, Death 3. Can't remember if it had sloth also. Granted, those are worth a lot of points, but I'm pretty sure that whatever the AI bought, it wasn't worth it. In the Dom II demo, turmoil seems to be a favorite of many AI opponents, and I can't figure why, as I'm always strapped for cash. As someone mentioned earlier, human designed pretenders for AIs would probably make a lot of sense. As I play mostly SP, it always pains me (when playing Dom I) to see the AI shoot itself in the foot with an idiotic domain.

Next, some anecdotal evidence of AI brilliance. I was playing Abysia, and was besieging the Tien Ch'i capital. I had Anathemant Something, plus three Warlocks as combat mages (probably not a great choice). Anyway, while sieging, there was a break-out attempt, as there always is. However, this time was different. It consisted of one commander, a mage with 3 Astral and something else, plus a few traditional troops. What happened was that the normal troops drew my army's attention long enough for the mage to slay all three of my Warlocks

with Magic Duel. Naturally, the mage died in the end, but if this was a deliberate move instead of a random happening, it shows an amazing cost/benefit analysis by the AI. Anyway, just wanted to share the story,it gives me some faith in the AI.

K.