Does the AI Boost?

[jutetrea]

I like the idea.

Any idea how the AI prioritizes rituals or if it does at all? Prioritizes commander purchases?

Possibly make it a research lvl - 0 national spell, muli-path'd to their rarer capital only magic combination and a moderate gem cost (maybe equiv to a lvl 3-4 summons)?

Possibly a high resource/high gold cost commander as well, no leadership, thug stats/abilities.

I'm not sure if the AI:
site searches well
accumulates gems towards goals
alchemizes
boosts
summons "bridge" magic users
summons "diversity" magic users
casts gem producing/beneficial globals
Summons troops effectively

On a random note, can rituals ever cost 2 gem types? Even via mods?

[Kristoffer O]

> On a random note, can rituals ever cost 2 gem types? Even via mods?

Nope

[Edratman]

jutetrea said:


I'm not sure if the AI:
site searches well
accumulates gems towards goals
alchemizes
boosts
summons "bridge" magic users
summons "diversity" magic users
casts gem producing/beneficial globals
Summons troops effectively

?

I think you have summed up the major deficiencies of the AI and why a person can beat the AI once he learns how to play the game.


Despite these shortcoming, the AI kicked my butt for a month or so until I finally quit playing the game like CIV. So I think the AI is pretty impresssive. Because it doesn't perform the items on your list well and I guarantee that your list will be expanded (no Nostradamus points available for the prediction) it is not perfect. I don't know if a perfect AI is achievable.

Look at the time and resources IBM put into making a computer that can beat chess champions. That is the type of effort required to make a near perfect AI. Is it reasonable to expect that from a game company?

Everyone's suggestions are probably helpful to the Dev's and I imagine they evaluate all, or at least most, of them. And the devs probably put a good number into the category of "WOW, good idea, wish I knew how to do that".

[Jazzepi]

Chess is also extremely simple when compared to games like Dominions 3. There is no hidden information, and there are only a certain set of given moves.

Jazzepi

[Folket]

Sure chess is more simple then Dominions but on the other hand good chess players are probably much better at chess then good dominion players are at dominions.

[Edratman]

To further buttress this line of thought, two weeks ago there was significent ballyhoo that someone developed a computer that was absolutely unbeatable (could not lose, worse case is a draw) in checkers. It only took them two decades. I may be wrong, but checkers seems to be a lot simplier than Dominons.

As an aside, in the same articles, there was a sidebar story about a chap who was the long time target of the checker playing computers. This chap held the world checker title for 40 years, never lost a match and only lost 9 games in 40 years, counting computer opponents.

No matter what your opinion of checkers is, that is a gaming accomplishment of the ultimate magnitude. Maybe the dev's should name a god after him?

[Gandalf Parker]

jutetrea said:
Any idea how the AI prioritizes rituals or if it does at all? Prioritizes commander purchases?
casts gem producing/beneficial globals
Summons troops effectively

The AI seems to use more of the higher-level national units if there is enough resources.

I have seen it cast gem producing globals quite abit. Unless its the "effectively" that you are making a point of.

I have also seen lots of summons troops. Again, unless you are pointing at the "effectively".

The problem with "effectively" is that it tends to require a lot of "IF nation=xxxxxx" branching in the code. And even then there are variations. The maenad version of Pangaea has different needs than the carrion version.

[Ironhawk]

Yeah those are the really big questions when designing SC/thug summon spells for the AI. Without knowing how it chooses one ritual from another it will be very difficult for me to figure out how to incent the AI to cast this particular spell. My worst fear is that the AI will assess a summon spell based on the number of effects which would pretty much shoot the whole idea in the foot. But then... iirc, I remember the AI using cave and fire drakes occassionally so it must have some understanding of summoning.

My current thought on the topic is to make it a 0-level national spell with 1 gem cost to summon a thug. And then just see how many of them the AI ends up casting. If I am overrun immediately by thugs then I will have my answer. But honestly I'd rather have some more solid data from people who have played a lot of SP games...

[Foodstamp]

The AI loves to summon. If you want to make sure that it uses the custom summons, be sure to set them to research level 0.

I use quite a bit of custom summons in my mods and it has been my experience that if you set it to research 0, the AI will summon like crazy, but if you set it to a different research level, it may take the AI a while to get there.

[jutetrea]

I haven't tried playing against the kobold mod, but it has a nice summon mechanic. It has a rare expensive (pretty nifty) commander that gets (I think) 3F2H +50%FE. Then it has a summon spell at research 0 that requires 4F2H. So a good portion of the commanders can cast it, but the commanders are about 350-400g apiece so going for that random doesn't happen to often.

His summon spell is 3-5? nice units, similar to upgraded celestial soldiers, for 15F gems. It could easily be 1 of those as a commander and then you have a thug.

High priced casting commander (wonder if the AI would even build?) but good abilities beyond paths, sacred, almost a thug in its own right. Moderate priced summon at 15 of a primary gem, and beefy when summoned.

If the AI won't recruit the high priced commander it could even be made more simple by keeping it at a common path lvl, but you also want to avoid the summons that is so worth its cost it becomes the nations primary goal.