The Big Question - How Does The AI Choose Which Units To Build?

Originally posted by ozymandias


:eek: My Dear Rocoteh!

I would NEVER suggest you or anyone else do such a thing! -- I know I won't ;) ...

All The Best,

Oz
Click to expand...

ozymandias,

I know that, but I did edit out my first comment
so I replaced that with a general comment:) :)

Rocoteh
 
ozymandias,

Back to the subject.

Have you read ACW-thread page 40, post 795? If so,
what is your comment?

(OK, my first comment was something like:since Dunnigan
once said wargames and strategy games are illusions (like
movies) we should sit back and "imagine" AI is okay)
It would work as irony in my own language but maybe
not in english.


Rocoteh
 
Originally posted by Rocoteh
Have you read ACW-thread page 40, post 795? If so,
what is your comment?
Click to expand...

Ah! Found it. :)
Here is the link:-
http://forums.civfanatics.com/showthread.php?postid=1176803#post1176803

What do I think of it? I really don't know.
I must admit that I have always assumed (as have many other people....see HERE), that bombard works just like combat does.
I.e. a bombard of 8 has twice the chance of hitting a unit with a defence of 4 than a unit with a bombard of only 4.

Could it be that we are all wrong, and that this only applies to OFFENSIVE bombardment?
Is DEFENSIVE bombardment always a 'fixed' amount, no matter what the actual value?
Do Catapults have EXACTLY the same chance of causing a defensive bombardment hit against say an attacking Swordsman as a Radar Artillery unit has?
Logic says no....but I am begining to have doubts! :crazyeye:
 
Kryten,

Thank you for the comment.
I must admit I was surprised after completing
the experiments.

Rocoteh
 
Originally posted by ozymandias
A slight digression --

Has anyone played around with the "Dora" unit, or for that matter any >2 tile range units and noted specifically how "well" the AI uses them for bombardment?

Thx,

Oz
Click to expand...

OK, I will run a test tomorrow: Defender (human) will
have 25 infantry. Attacker will have 25 infantry +
5 heavy artillery 250 Bomb. Range 3 ROF 6.
There will be no infrastructure that AI can shoot at.


Rocoteh
 
Originally posted by Rocoteh


OK, I will run a test tomorrow: Defender (human) will
have 25 infantry. Attacker will have 25 infantry +
5 heavy artillery 250 Bomb. Range 3 ROF 6.
There will be no infrastructure that AI can shoot at.


Rocoteh
Click to expand...

:thumbsup: Thank you!

:) ,

Oz
 
Experiment 1:

I added 5 infantry for both sides (France-human, England-
Computer) and also changed the map since AI avoided
combat and tried to outflank me. Each side had 3 starting
cities. The English artillery had 2 movement 10 defence
250 Bombard 3 Range 6ROF + ZOC.

When AI saw it could not outflank me it started a good
old WW1 war. All my infantry-units were fortified in a
continous line.
AI attacked more than 10 turns (without any artillery
support) Then it had lost its 30 infantry units.
My own losses were 15 infantry.

I now counterattacked and there (you guessed it) in
the cities were the heavy artillery.
It did not use its Bombard cap. against me.
When I made a ZOC to ZOC move I lost 1 step-point!
Remember these units had 250 Bomb. 6 ROF.

I terminated the experiment.

ozymandias,

Any ideas for experiment 2?


Rocoteh
 
Experiment 2

I added 5 heavy bombers for England.
Heavy Bomber 100 Bombard 6 ROF 40 Range.
The is map is explored (so it was also in experiment 1).

Turn 1- Turn 3 AI makes 13 out 15 air-strikes against
my cities. Then on Turn 4 after losing 13 out of 30 infantry
AI changes strategy. Nearly all air-strikes are now
directed against my infantry with devasting effect.

Still AI loses 4 more infantry and all remaining infantry
are now directed to capture one of my cities.

I send 50% of my infantry to to defend the city.
All AI air-strikes are now directed against that city.
AI fails to capture the city and have now lost all
its infantry. My infantry now start to march against
the English cities. AI responds with directing all its
airstrikes against my advancing infantry and
here the experiment ends. The English Heavy Artillery?
Its sitting in the English cities all the time.

Remember in both experiment 1 and 2 there were no
infrastructure. In experiment 3 I will add a great railroad-system,
since AI really like to bomb railroads.


Rocoteh
 
Hi Rocoteh,

It sounds like 2 very well designed and executed experiments, even if the results were (to put it mildly) disappointing.

Conclusions and observations re:

Exp 1:

(i) The AI is indeed clueless re: artillery.
(ii) Kryten's (was it he who made this observation??) is correct, that artillery units are essentially defensive in nature and designed to only hinder an assault by inflicting minimal casualties (so much for WW1 ...)
(iii) Assuming that artillery will be used offensively almost exclusively against infrastructure, the "art of modding" becomes working out ratios of # of bombardment units and their BF/ROF vis-a-vis city improvement defense factors ...

Question: were the artillery within range of enemy cities and, if so, did they bombard infrastructure?

Exp 2:

(i) The AI uses aircraft as strategic air warfare was envisioned as WW2 before the blitzkrieg: that is, it obviously will attack infrastructure first and units only if desperate (I know, this is an extremely obvious one from reading your results -- I'm repeating it (a) as a sanity check and (b) to make cutting-and-pasting our results into a "definitive" set of rules easier)
(ii) The AI model makes sense from a gaming point-of-view: the AI builds bombers which will ultimately cripple your infrastructure unless you build fighters etc.
(iii) Given the limited "playing field", it seems that the AI (given the above assumptions) actually use aircraft rather well ... ?
(iv) A guess: the Conquests use of land unit AA factors will be employed to offset the "human intelligence" advantage of using airpower against combat units.

*sigh*

I'll give some thought as to how to test unit flags re: build preferences; I'm beginning to fear for ACW etc. re: naval units that the best approach might be to have an improvement a la the Knights Templar which simply periodically turns out different types of ships ...

Yeah, the results were disappointing, but still --

Great work! :thumbsup:

-Oz
 
There were no infrastructure since I wanted to
force AI to attack units. In the next experiment that
I will start 10 minutes from now infrastructure will be
added.

I agree with you that with regard to history there
is a point that AI should go after infrastructure

....but after 1940 nearly all offensives launched when
enemy had air-superiority have failed.

This should also be reflected in the game.


Rocoteh
 
Experiment 3

Maybe the most interesting: (at least for me, since it confirms
what I had suspected) I have now added a great railroad-
system. AI only attacks railroads until it have lost 17
infantry units, then 1 airstrike per turn is allocated against
my infantry. When AI has lost 27 infantry (out of 30) it
start to retreat. I advance with my remaining infantry.
From now AI only attacks my infantry, thus making
the counterattack meaningless.
Experiment terminated.

As I suspected AI will go for 1)infrastructure 2)cities
3)units with its air-units. I think that is clear that the above
is the priority.

...and also the Heavy Artillery were sitting in the cities.


With regard to NAPP: I think we should have a very
restricted number of artillery units if AI is not changed
in Conquests.


Rocoteh
 
Originally posted by Rocoteh
I am sure I have never seen a report like:

"The end phase of the game was tough. I had to
destroy the huge fleet AI had built".

Can there be something like a 10% rule for AI??
I mean : AI will stop building naval units when they are
let us say 10% of total forces?
Click to expand...

That just goes to prove you don't read all my posts with fanatical attention. :) I've mentioned in several threads on AI naval weakness that I had one game (vanilla, one of the earlier patches), in which the Iroquois built a huge fleet of Frigates and eventually Ironclads, which I was unable to beat at sea. Thanks to the AI being to dumb to block the sea-lanes, I eventually managed to get a force of Cavalry over to the Iroqois's continent and capture their harbours, thus ending the threat of their fleet.

I've never ever seen anything similar before or after, nor can I recall anyone else reporting a similar experience. I would estimate the Fleet to about 1/3 of the total Iroquoise war machine by numbers.
 
Originally posted by The Last Conformist
A point on RoF and ZOC: ZOC fire is always one shot, no matter RoF (this is possibly to allow RoF=0 non-bombardment units to do ZOC attacks), so Rocoteh's units only losing one HP is perfectly expected.
Click to expand...

The Last Conformist,

"ZOC fire is always one shot, no matter RoF" The Last Conformist

I know that, but I am not sure everyone knows it.


Rocoteh
 
Originally posted by Rocoteh


The Last Conformist,

"ZOC fire is always one shot, no matter RoF" The Last Conformist

I know that, but I am not sure everyone knows it.


Rocoteh
Click to expand...

Maybe I'm confused -- I knew about the one shot deal; what happens/happened with multiple artillery units per tile? More than 1 HP loss presume ... ? Which might make a "solution" a relatively high number of cheap ROF1 units an answer for some historical periods -- horse artillery companies in a game where most units are regiments; etc.

Abraxas,

Oz
 
Originally posted by Rocoteh
Experiment 3

As I suspected AI will go for 1)infrastructure 2)cities
3)units with its air-units. I think that is clear that the above
is the priority.
Click to expand...

... And thus is another "rule" discovered! -- Again, Rocoteh, i must say "well done", even if we are all (metaphorically, I hope :) ) tearing our hair out over the results ... :) :thumbsup: :cry: ...

...and also the Heavy Artillery were sitting in the cities.
Click to expand...

-- & again: *sigh* ...

With regard to NAPP: I think we should have a very
restricted number of artillery units if AI is not changed
in Conquests.
Click to expand...

Agreed -- maybe just give all "larger" units BF1/ROF1 + ZOC to simulate the artillery of the era ...

Best,

Oz
 
ZOC to ZOC movement, 5 artillery units with ZOC, how
many shots? I will check it out. My next experiment
(starting 20 minutes from now) will be a human
controlled armour force attacking at broad front
against a continues line of infantry, each backed up
with anti-tank artillery 300 Bombard0 Range 6 ROF.

Not very historical, since the principle of finding a point
of breaktrough (schwerpunkt) was central in armored
warfare, but still interesting.

Rocoteh
 
Experiment 4

15 human controlled Modern Armour Attack 24 Defence 16 attacks
15 AI controlled Infantry Attack 6 Defence 10.
Each Infantry unit is backed up by 1 Anti-Tank unit
Attack 0 Defence 0 Bombard 300 Range 0 ROF 6.

Outcome: 4 M.A. wins = 1 hitpoint lost 7 M.A. damaged=
Attack repulsed 2 hitpoints lost 4 M.A. destroyed

Now Experiment 4b
Bombard set to 30. Nothing else changed.

Outcome: 10 M.A. wins 4 M.A. damaged 1 1 M.A. destroyed.


Rocoteh
 
You must log in or register to reply here.

Similar threads

SayHayKid
Infantry Upgrade to?
Replies
2
Views
581
Predator145
Predator145
Predator145
AI changing strategy Flag?
Replies
7
Views
1K
Predator145
Predator145
A
How does the AI decide where to build cities?
Replies
0
Views
219
average_communism_enjoyer
A
T
AI League 2.0
Replies
12
Views
508
Thrasybulos
T
C
Can't seem to choose number of AI players
Replies
5
Views
553
Civvillian
C
Back
Top Bottom