What kind of AI do you want?

[sunsmountain]

Perhaps the CA can combine the best of MTW and RTW AI:

(MTW) * At the beginning of the battle use 1 of a set of formations that work, for say, a heavy infantry army.

(new) * Then choose 1 of a set of attack/defend strategies and translate this to the tactics level. Group units together [for 20 units, a max of 5 groups each with 2 (horses, archers) to 8 units (infantry)] and keep them together & coherent until engaged. Try to identify groups (again, max 5) of enemy units & apply the rock-paper-scissor knowledge to set up your line before engaging the enemy. Always engage the enemy as 1, if possible (ie unless fighting has already started, not with scouts though).

(RTW) * Once engaged, free all units of their groups and allow them to react on the spot & select their own targets. Prevent ParadingArchers/SuicidingGenerals as much as possible. Identify human outflanks and counter them.

* When losses do occur, retreat & regroup, start from the top, but skip the initial formation, go straight to the group level, which due to losses may be less than 5.

What are your thoughts? What kind of AI would you like?

[edyzmedieval]

I want a totally new one because of 2 simple reasons:

1. The MTW AI was too inflexible. It was difficult to use specified tactics in the battle.

2. The RTW AI can be beaten easily, by exploiting the AI problems.

So I want an intelligent AI, built from scratch.

[caravel]

I'd like to see an AI that, among other things, learns and saves it's own encrypted log files, which it can refer to later. So for example, if you start a campaign and always perform the same tactics, the AI wises up to it and eventually get's to know what you're doing. It could also borrow your tactics and use them against you.

[Furious Mental]

That would be awesome

[Rodion Romanovich]

I think learning AI adapting to how the player plays would be awesome, but such stuff is complex to implement and might not be possible with CAs budget for MTW2. What is possible though is a system where the AI maintains a good battle line, and makes proper decisions in a way almost similar to how the human player makes them about keeping or not keeping reserves, where to send in the cavalry charge, whether to outflank on both flanks or one, etc. An AI that also handles missile duels well, taking advantage of every range distance between missile units, using cavalry to chase away the player's missile units to temporarily keep them from firing to give the own missiles a temporary advantage etc.

I'd also like to see the skirmish function be more logical so units don't skirmish away from their own line to be isolated and slaughtered. They should always try to skirmish towards the own main battle line and once behind or inside it stop IMO.

[Dead Knight of the Living]

I want an AI that maintains freaking battle lines. In RTW the enemy army comes at you piecemeal and you can take out units one at a time. BORING.

Despite claims to the contrary, the same thing happens in Europa Barborum mod and the RTR mod.

Now, when I say I want them to stay in line, I'm talking first echelon only. They can maneuver the 2nd and following echelon forces in different ways to try and gain an advantage.

What I'm most disappointed in is that the AI does not coordinate Cavalry flanking attacks with its infantry and artillery/archer support. Yes, they maneuver cavalry to your flanks. But they'll attack before the infantry arrives or after you've already beat the crap out of the infantry.

Much more well coordinated attacks, timed well. IE, the cavalry should be hitting my flank slightly after their line engaged my line.

The AI should also use terrain to select tactics. If attacking at an angle on a slope, IE the their left flank is elevated on a hill while the ground slopes down to where the right flank is on level ground, maybe the AI could reinforce the flank on the high ground (left flank in this case) Try and crush my flank from the high ground while conducting limited engagement with the right. Kind of like Epaminondas (spelling?)

If attacking from the high ground, precede the main army with light cavalry and archers to use the range advantage. Something like that.

[Geoffrey S]

I'd like to see an AI that, among other things, learns and saves it's own encrypted log files, which it can refer to later. So for example, if you start a campaign and always perform the same tactics, the AI wises up to it and eventually get's to know what you're doing. It could also borrow your tactics and use them against you.

That's a lot to ask for. Designing a good AI is hard enough, particularly for a game as complex as this, and one that actually learns is probably beyond the means of most games right now. However, if it keeps coming across cavalry heavy armies it could decide to build armies designed to cope with that, such as many spearmen; that should be reasonable.

Basics need to be right, though, which they certainly weren't in RTW. Armies should stay in formation, no suicidal generals, and the AI should recognise weak areas and focus on breaking them; coordinated charges would help, particularly from cavalry. But having the armies stick together and not spread out, especially when attacking, would help immensely.

[Tahanaman]

I agree with Orda Khan- hello by the way! I would like to see an AI that would hand my butt to me for making mistakes on the battlefield

[Ironside]

Although mostly issues with the MTW AI, I would like to see proper a "horde" mode, that is proper swarming if the comp has weaker troops but larger numbers. Or the closly related "everything is going to hell anyway, so why don't try to take out as much of the enemy before biting the dust"-mode.

The MTW AI has big problems with the second one. Proper handling of reinforcements for both MTW and RTW (although different ways) and proper treating of a fast general unit if there's large forces still active (aka in RTW, if the first army is smashed retreat the remaining cav and especially the general to the other army and for MTW the general retreats to meat up with a new row of a full 16 units).

That combined with a great AI (MTW quality minimum) and with battles that last more than 10 seconds MTW2 would look very promicing.

[Orda Khan]

I agree with Orda Khan- hello by the way! I would like to see an AI that would hand my butt to me for making mistakes on the battlefield

Hello Tahanaman, I've not seen you in ages. Hopefully MTW II will prove to be worth the effort in MP

......Orda

[caravel]

That's a lot to ask for. Designing a good AI is hard enough, particularly for a game as complex as this, and one that actually learns is probably beyond the means of most games right now. However, if it keeps coming across cavalry heavy armies it could decide to build armies designed to cope with that, such as many spearmen; that should be reasonable.

Not really, it would not be much different to the logs kept and used as an AI script by programmed chess opponents.

[sunsmountain]

I'm glad with all the valuable responses. We all want units to follow Sun Tzu's Art of War, and CA have aimed for this since Shogun.

The AI in RTW however, fails to keep to the basics in the long run. We seem to all agree that these basics are more important to get right than which formation, strategy or tactic to use.
These basics (keeping troops together, sticking with decisions once made) go wrong exactly because of the following design decision:

How often do you want the AI to re-evaluate its decisions?

In Rome:TW, the answer was: Too often, leading to such horrors as the save/reload bug on the campaign map, where the AI evaluates what it's doing EVERY load. On the battlemap it re-evaluates EVERY 5 seconds, leading to those wonderful scenes with enemy units marching back and forth without engaging, or engaging flank first, or archers repositioning and never firing.... brrr

In Medieval:TW, the answer was: Twice, at the start of battle and just before engaging. No updates after that, only (smart) unit behaviour according to Sun Tzu rules. Mind you, they had less options/formations to choose from compared to RTW, and were therefore more likely to do things right. The MTW campaign map didn't re-evaluate its decisions at all, they always built and moved the same. The map was simpler too.

CA's weak point seems to be that once they have made a design decision, they're sticking with it. If only they could translate that determination to the battlefield :) yes it's ironic.

A better question would be:
When is it necessary for the AI to re-evaluate?

I'd like to see an AI that, among other things, learns and saves it's own encrypted log files, which it can refer to later. So for example, if you start a campaign and always perform the same tactics, the AI wises up to it and eventually get's to know what you're doing. It could also borrow your tactics and use them against you.

Ahh.... you were talking about campaign map AI... then don't say: Tactics. Campaign is all about strategy, not tactics, tactics is for the battlefield:

What it can learn, is what kind of armies you are building, ie heavy in Spearmen, swordsmen or archers, and assemble its armies to match that. But Rome campaign map AI already does this (while building all military buildings, not focusing, so it doesn't matter, but still).

Perhaps one the AI designers from CA could comment on their design decisions or shed some light...

[Lord Adherbal]

Not really, it would not be much different to the logs kept and used as an AI script by programmed chess opponents.

actualy the TW game rules - or any RTS rules - a a *tad* bit more complicated then those of chess

They always run into a strong front line and they rarely ever employ archers.

yeah this is one of the most incredible flaws (as in they would've ben so easy to fix). The AI never uses an archer advantage, atleast not when attacking. If he has better archers then his opponent, and his infantry/cavalry advantage over his opponent is not like 10 to 1, then he should sit back, and NOT attack with hs inf and cav until every last arrow has been shot.
But perhaps CA figured this would be too boring for the average 12 year old kid ?

[Cesare diBorja]

I would to see an AI that combines all of the features provided in the poll. A totally humanistic AI that makes a change when it is necessary but otherwise follows basic strategy and with a special knack for improvisation(sends chivalric seargents and crossbowmen to assist militia seargents in attacking a fortified position).

[Cesare diBorja]

I would to see an AI that combines all of the features provided in the poll. A totally humanistic AI that makes a change when it is necessary but otherwise follows basic strategy and with a special knack for improvisation..........

i.e. starts battle with the intent to engage with missile units and finds the enemy statica as well so seend the cavalry in to defeat enemy missile troops and supports them with feudal seargent and feudal knights. Upon defeating these forces, more towards the center the AI is confronted with the players stronger forces and sends in adequate reserves. The right side is player's right flank is particularly tough and so the AI sends militia seargents to assult a fortified position. When this unit begins to have trouble, the AI sends chivalric seargents and crossbowmen to assist the militia seargents in attacking a fortified position).

diBorgia

In short I would like an AI that tests defenses and modifies it course of action. I would also like it if the AI knew, in detail if it's position was tenable or not.

[Evil_Maniac From Mars]

An efficient, challenging, non-suicidal AI. Built from scratch.

[Zalmoxis]

Wouldn't it take a long time to create a brand new AI?

[Evil_Maniac From Mars]

Ah, but I'd play it longer. It would be worth it IMO.

[AwesomeArcher]

I liked RTW's AI, it was a good difficulty for me, challenging, but winnable. Although i would like more options that the user could define. Such as the user could set whether the AI uses move infantry or cav. Or works on building an army for attacking or fortifying for defense, you know stuff like that.

[Monk]

Building Ai is a complex thing, so i've been told. However, if CA is truely spending all this time polishing mtw2, i'd ask they spend this time on the AI. it needs a huge overhaul; anything better than an RTW AI will make me happy. and i'm willing to wait longer just for a challenging game

edit: because i sounded like a jerk. <3 CA

[redriver]

I've only played Rome and so my selection was the last one. I'm pretty sure that's what CA will endup doin' anyhow... I just wanted to point out that RTW AI isn't too bad. I've had my original game bugged by the offiicial patch fix and had to delete everythin' and reinstall RTW from scratch. applied patch 1.3 then 1.5 and noticed that there's lotsa files cleared up in the root directory and I think the game AI was better as well. didn't get the large wall beseigin' tower bug either.
ok the new AI should be true to the time period. the romans were masters of warfare compared to most other factions of the period. barbarians didn't even have proper formations or unit types. just one big hip of randomly armed/armored men chargin' the enemy lines was pretty much the norm of the time... thus less depth/AI vs MTW IMO. I too hope this reincarnation of MTW will be more than just graphics update :)

[Ludens]

ok the new AI should be true to the time period. the romans were masters of warfare compared to most other factions of the period. barbarians didn't even have proper formations or unit types. just one big hip of randomly armed/armored men chargin' the enemy lines was pretty much the norm of the time... thus less depth/AI vs MTW IMO. I too hope this reincarnation of MTW will be more than just graphics update :)

I don't quite agree with your starting points. Unlike the Romans would want us to believe, the battle tactics of the so-called Barbarians could be quite subtle compared to the steamroller tactics Roman armies usually employed. Yes, the Roman legion allowed for more flexibility than most other armies at that time, but many Roman generals did not have the ingenuity to use this. I am not saying that the barbarians were sophisticated tacticians, but the idea that barbarian armies were just a mass of men going for an all-out charge is simply not true.

[Sabuti]

I just want the Ai to be better on both the battle and campaign maps. The Ai is what will make or break the game for me. If I hear too many bad reviews of the AI on fan sites I won't even bother buying the game.

[sunsmountain]

After the last couple of months discussing the AI, my general sentiment towards the AI is still in the same direction: Don't buy unless you know they've taken us seriously. Will they stop giving more beautiful breath taking screenshots and start talking about the really difficult stuff? Or will that be too boring according to marketing directors and managers which care primarily for 1 thing, exposure and cash.

And what about the possibility of letting the AI Programmer Team continue long after the Art Team has finished their work??? Anybody from CA like to comment (probably not, but hey, i can always ask )