AI Horror Stories

[chickenhawk]

There are two VERY basic AI mistakes that I think the programming to fix should be doable, in addition to the vast number of specific issues that are, in many cases very difficult.

The first is the strength calculator that the AI uses to assess whether it can accomplish a given goal. The A.I. should be far more conservative about how many troops it needs to conduct a given operation, whether it is attack, defend, or siege against the the player controlled faction. It is my understanding that not only is this portion of the AI less than brilliant to put it mildly, but that it is almost inaccessible to modders. If CA can't fix it, at least let the community try. Sending a stack with less combat power than the garrison to attempt a siege is only the most obvious of these errors.

Second, and this is related to the passive AI bug. the AI should be able to assess whether it is winning or losing the missile fight based on overall missile casualties per minute. Again this is not complicated programming, just missile casualties inflicted vs. missile casualties received per unit time. If it is losing badly in this comparison it should do SOMETHING, either charge or retreat or something. Standing there and letting Retinue Long-bowmen empty their quivers at you is not a strategy, it is suicide. Staying in range of the wall towers is even worse for any goal except a reasonable chance of breaching or seizing those walls is even worse than suicide, it makes for a boring game.

The idea that a sally can be conducted against a completely unprepared force in daytime is also just silly. The AI does it badly and the player does it far too effectively. This just needs to be completely rethought.

For that matter the whole concept of siege engines needs to be completely rethought. Except for the very highest tech cannon they simply had no battlefield application in this period. The rate of fire and the time it took to aim/move them was completely prohibitive. Sieges should be conducted on a points system where the siege engines simply give more points per turn. A given level of walls/towers would then take a certain number of points to break down. The more points a besieging army accumulated the more breaches it would have when assaulting on the battle map. this would also increase the level strategic decision making in terms of waiting to lift a siege as the chance of a successful assault would keep going up due to the gradual degradation of the defenses. Finally if their were no siege engines on the battle map the AI could not do silly/suicidal things with them.

Good Lord make me stop.....

[Gaius Terentius Varro]

Weird seems like the same people that complain about the AI failures are unable to stop using the exploits. Then again pretending that Kasparov is on the other side of the chess board is a bit unrealistic. It's just never gonna happen... The passive missile bug is still present in RTW after all the patches, the very same game we all loved to put down in comparison to MTW's AI and now look fondly upon... So is this a worse game? Hell, not for me. It's way more immersive and the true challenge of being pitted against a 10 star general is right there just try the multiplayer.

[Darkgreen]

I can’t stop people from posting here, and there has been some good stuff here and some stuff related to the original topic, but for the guys at CA it might be easier for them to sift through this stuff if we stick to the original topic. I started this thread because one of the guys from CA said:

"it'd be interesting to read a thread of 'ai horror stories' actually, the ai behaviours which people exploit the most to gain easy wins."

If you have other comments not related to this, I would suggest (I know I can only suggest) you put them in a different thread. It would probably make things easier for everyone.

[TevashSzat]

This has happened four times now in Baghdad. I am Venice in a migration campaign and have a nice garrison in Baghdad due to the nearby citadel, but I am constantly besieged by the mongols. Every time, they send a ram and it breaks down the gate so their whole army charges in, but I have 6-7 silver chevron venetian infantry standing right near the gate which holds their charge so I send my feudal knights and my general out back and charge them in the ass causing insta rout. I have a general and 4 feudal knights all with full exp due to this, it really is pretty sad.

[Crazy Larry]

There are two big flaws the A.I. has when sieging the town that need to be fixed (at least they are present in 1.1). First, a very small army can defend a settlement against an A.I. siege if the A.I. has not brought siege equipment that can batter down the walls (i.e. Ballista, Catapults, etc.), as long as the settlement has walls that they can walk on. This is because once the A.I. has ladders at the wall, if you can engage the units pushing the rams and seige towers, the A.I. will forget that they have them and try to send everyone up the ladders, where you can rout their army very easily due to the huge disadvantage units scaling walls via ladders are at. While I've never been in the exact situation, I would not find it surprising that a garrison of 4 or 5 Town Militia could beat an entire stack this way. All you have to do is send one or two units outside the wall, wait until the ladders are up, and then attack the rams and/or towers. After that unit routs or dies the entire enemy army will try to go up the ladders. I've racked up some amazing victories this way, with minimal loss on my part (often beating the enemy by more than a 10:1 ratio).

Secondly, when an A.I. spy successfully opens the gates for an enemy army, the A.I. ignores this except as a last resort. They still try to send towers and ladders to the walls, and will wait until you've killed 50%+ of their army before they walk through the gates. This is somewhat offset by the fact that certain city layouts consider the "main" gate different for attackers and defenders. Thus, I've had a few times where the enemy has had a spy in a settlement and no time to prepare and ended up with 90% of my army on the opposite side of the city. Now, I understand that a spy is supposed to be part of a sneak attack, but it seems odd that a single spy could cause my entire army to camp out at one gate when the army is building towers and rams 500 feet from a different gate.

[Slaists]

Passive AI "Fix"

Here is a minimalistic horror story, which, unfortunately, is rather common after the passive AI "fix".

I am the attacker;
My army: two hobilar units;
AI's army: one unit of armored swordsmen;
Terrain: steep hills, a flat valley and deep woods;
The AI starts in the woods (makes perfect sense in this scenario);
My army starts in the open.

Instead of sitting in the woods, which would give the AI the forest advantage + disadvantage to my already weak cavalry, the AI starts chasing my hobilar units in the open; get's charged from two sides (after running for some time chasing my cavalry) and runs...

If the AI was just sitting in the woods and waiting for my cavalry to approach, I would stand no chance.

[FactionHeir]

The AI would do that in 1.1 too. Basically because it thinks that it outnumbers you (in terms of battle odds)