New patch: AI walled city attack experiment results

[Jaybe]

@Gort, you mention in your experiments that you build a single city and there is only you and the Aztec. What if they are reluctant to take your city because of potential warmongering penalty (DESPITE there being only the two of you).

Also, were (some of) the rebels in Test 5 potentially Aztec because of your destroying so many of their units?

--
BTW, I am re the potential for AI air forces so I can discover the potential of anti-air support units, not to mention the use of fighters (and their promotions) for anti-air use. My games are epic (speed) on huge and haven't started a new one yet.

 

[Gort]

Also, were (some of) the rebels in Test 5 potentially Aztec because of your destroying so many of their units?

I think most of them were my rebels in the end, I was on -9 amenities that time.

 

[CivCube]

Fantastic write-up, Gort. Sounds like there's actually a game to play, now!

 

[cazaderonus]

Interesting tests ! I didnt get to read the pre patch results though, what were they ?

If those are an improvement, it seems warring on deity still remains easy as hell.

 

[Leyrann]

Interesting tests ! I didnt get to read the pre patch results though, what were they ?

If those are an improvement, it seems warring on deity still remains easy as hell.

Pre-patch Gort gave up because they didn't manage to conquer the city.

 

[leandrombraz]

Is it possible that the AI sometimes just don't want to take the city, that there's an instance where they are programmed to invade and pillage but isn't aiming for the city? Like on test 1, maybe the AI wasn't trying to take the city into turn 328, when they send a bigger army and actually finish the job. Before that it wasn't a conquering army, the AI was merely messing around and trying to hurt you as much as possible while aiming for other objectives.

 

[V. Soma]

TALE OF TWO CITIES

let me post here a test with two cities:

My leader: Cleopatra
Difficulty: Deity
Game Speed: Online
Map Type: Pangaea
Map Size: Duel
No City States
AI Leader: Montezuma
Start Era: Information Era
Resources: Abundant
Start Position: Legendary
Domination victory only
No turn limit
No barbarians

METHOD
Build capital, find Aztecs. Build a 2nd city close to Aztecs.
Then at the same turn declare war. Disband all units.
Do not attack Aztec at all...

208 game starts
214 2nd Aztec City
216 I settle my 2nd next to Aztec border. It is 11 tiles from my capital.
I delete all units, DoW on Aztecs
There is an A ranger near my 2nd city
218 That lone A ranger attacks my 2nd city
219 That lone A ranger attacks my 2nd city
221 A gets 3rd city
222 A is nearing 2nd city with 1 infantry corp and 1 m-gun corp
223 That lone A ranger attacks my 2nd city, another inf is nearing
224 That lone A ranger attacks my 2nd city
the other units stop 2-3 hexes away from my 2nd city
227 That lone A ranger attacks my 2nd city
the other units stay 2-3 hexes away from my 2nd city
228 A build Aerodrome near my 2nd city, has a
mech corp, inf corp, inf withdraws
229 That lone A ranger attacks my 2nd city
a chopper and an inf corp appears
230 inf corp is now army, stays away with the chopper
233 a settler appears north of my city
now there is an inf corp, inf army and chopper 2 hexes away from my 2nd city
235 now threre are two rangers, one attacks my 2nd city
two inf armies, one inf corp, one chopper dances around city
237 rangers form corp and attack
two inf armies move awy and go towards my capital (?)
other inf army stay near my 2nd city with the chopper
settler is in my territory
238 ranger corp attacks my 2nd city
239 first A bomber is built (I can see the aerodrome)
240 ACTION!
chopper and inf army moves in and attack my 2nd city
city is nearly destroyed to zero (21/200 and 29/200)
241 chopper and inf army retreats (???)
242 A units heal
243 ranger corp attacks my 2nd city
244 A has 4th city between my two cities
ranger corp moves away (to the new A city?)
246 3 unescorted builders go to the 4th A city
248 …and an unescorted settler
252 chopper attacks my 2nd city
there is one inf army and one mach-gun army near chopper
253 chopper attacks my 2nd city
mech-gun moves to city
254 A TAKE MY 2ND CITY

it took him 38 turns since DoW

264 chopper appears near my capital
265 chopper moves to city
(chopper stays)
276 inf army appears near
277 mach-gun army and two mil. Engineers appear near
278 inf army moves to city, now there with chopper
279 A TAKES MY CITY - Game over

it took him another 15 turns to finish it.

in total: 53 turns since my DoW

 

[PYITE]

An AI behavior that I have noticed a couple of times that kinda falls in line with your test results:
Sometimes the AI is getting ready for a city assault it will not attack until one of their have been attacked. Almost like the first shot from the city wakes them up and they commence with the attack.
A couple of times when at war I wouldn't notice that the AI is amassing troops outside of a city that I didn't expect for a couple of turns. I fire in my turn, then they start to attack next turn. Via auto save reloads this behavior seems to follow regardless of which turn it is. They always seem to begin their assault the turn after they were first attacked.

 

[bbbt]

For test #5 - where the Aztecs declared war on you - I wonder if it's some dodgy logic involving warmonger penalties that are involved. I.e. he has a warmonger penalty from declaring war and doesn't want more of one from physically taking the city (I know that makes no sense from our perspective with only one opponent, but I don't know if that sort of additional logic would be in there, versus a generic "warmonger penalty too high check")

 

[V. Soma]

Another test, this time with island map

My leader: Cleopatra
Difficulty: Deity
Game Speed: Online
Map Type: Island Plates
Map Size: Duel
No City States
AI Leader: Montezuma
Start Era: Information Era
Resources: Abundant
Start Position: Legendary
Domination victory only
No turn limit
No barbarians

METHOD
Build capital, find Aztecs.
Then at the same turn declare war. Disband all units.

208 game starts, I settle (to have my city open to 3 water tiles) I have a rather big island
213 I find Aztecs, delete units, DoW
220 A ranger corps ship close
222 A ranger corps ship gone
225 A inf corps ship close
226 A inf corps embarks next to my city
227 inf corps attacks
228 inf corps and ranger coprs next to my city
229 ranger corps attack, A has city on my island
231 inf corps move away
ranger corps attack
233 ranger corps attack
234 ranger corps attack
237 ranger corps attack
238 ranger corps attack
239 ranger corps attack
240 ranger corps attack
inf corps ship and wait at A city near
also a mech-gun ship
244 ranger corps attack
245 mech-gun corps embark next my city
246 mech-gun corps moves away
247 ranger corps attack
248 ranger corps attack
repeated ranger attacks
254 my city wall is destroyed
repeated ranger attacks
257 a chopper gets next to my city from the sea
258 and a mech gun army ships close
259 mech gun army embarks next to city
ranger corps attack
260 mech gun army attacks
chopper is on land
261 mech gun army attacks
262 another chopper comes too
my city is taken… game over

since DoW it was 49 turns - no ships used by AI…

 

[Yzman]

I am not sure how long the AI takes to take a city is the biggest factor to really look at. I mean, we are really looking at one facet of a very involved game. I really consider the AI not allowing you to win as the true test of this game, not how fast they themselves personally win it.

I understand the idea in general but I am just thinking that we typically don't consider deity hard because of how fast the AI beats us, it is because we can't win. Though them winning is of course one way to prevent us winning.

 

[Funababbitt]

Thanks for that. Yours was the all important question for the new patch. I haven't played Civ 6 now for months. I too wish your results were more encouraging but I'll give it a try to see if the game is worth playing yet. I'm still hovering on the border asking myself if Civ has a future or whether it's a franchise that's just run out of time.

 

[Tabarnak]

They're encouraging compared to the results before the patch.

It's still way under the more abysmal abyss.

 

[Art Morte]

Anyone care to run the test with the AI+ mod (or similar if there are such mods)? Would be interesting to see if any fan-made mods made a difference in a test like these.

 

[PYITE]

I did some comparison tests between AI+ and vanilla. Seems like the patch induced some minor behavioral bugs in the AI+ mod. Vanilla combat was better than the mod most times. Vanilla was frequently using ranged attacks to bring down walls. Ranged units were continuously attacking cities even after they had no defense points until a melee unit shows up and finishes off the city.
Side note: alert function seems to be working perfectly and if you enable auto end turn it will prompt you to use city ranged attack.

 

[Victoria]

I have played 4 dom dei games now.... Tougher at the start.
The tests are superb for showing how the AI behaves taking a city. Its a little bit better defensively but still got some issues. I woukd like to hear some real games out there.

 

[Gort]

Anyone care to run the test with the AI+ mod (or similar if there are such mods)? Would be interesting to see if any fan-made mods made a difference in a test like these.

The maker of AI+ suggests that you shouldn't use his mod for now:

If you're starting new games I would probably recommend starting without AI+ right now just to be safe and so you can see the changes they made.

-----

Regarding the usefulness of these tests, I considered the pre-patch inability of the AI to take walled cities to be inarguable proof of "bad AI" - something you can point to and go "This is a mistake, right here", like a spelling mistake in a book. It's also a handy metric in that it's something you can measure with numbers, and test repeatedly. Obviously there's a lot more to a game of Civ than just conquering cities, but when it's the basis of warfare in a game where warfare is highly important, it's important for the AI to be able to do it well.

In the five tests I've done I think I've identified three main things that are wrong with the AI right now:

* Occasionally gets into a state where it won't move its units. I saw this in test #1 where the AI just sat in range of the city with a force capable of capturing the city for about fifty turns. Presumably it was waiting for a second siege unit to arrive before attacking, or something like that, but it should at least know enough to sit outside of city strike range.

* Targets encampments and cities that no longer have any walls or health for ranged attacks.

* Too reluctant to send units to pillage encampments or attack cities with no walls.