The AI. How to fix it

[SPAMbuca]

Had quite some discussions about the AI by now. Not just after the launch of this game, but in general about the AI of about all Total War games. In general, the AI was found lacking and easily exploitable in most games; the eldest games in particular. In the eldest games, it was necessary to in some way exploit the AI to counter their massive cheating ways. For example, in medieval 1, it was possible to beat full stacks with a 1 peasant army if you were defending a fortress by running out and running back in right after. The army would collapse on the walls.

Same story in Rome 1. It was possible in a siege to run 1 unit outside of the walls and then back in to make the entire besieging army run around the settlement walls to be slowly picked off by the automatic firing towers. Like in any game so far, it's possible to exploit the battlefield edges; Best example would be to have hoplite units camp in a corner with their Phalanx stance active. If you'd put 2 hoplite units "into" each other which would create a double pike wall, literally nothing could penetrate it.

And in Medieval 2 and other parts, it was possible as a defender to take a cavalry unit and have it run around the map, Benny Hill style just to avoid defeat.

My point is that so far, in any Total War game, the AI hasn't been great/good and depending on the situation/difficulty, you'd have to exploit the AI. This is a game with a rather thin line between "cheating/exploiting" and "strategy".

Note however, that I don't want this topic to be about Rome 2 and the AI and it's flaws. We'll see whether that get fixed later and there's plenty of opportunities to discuss it already. I would instead like to discuss on what kind of AI you would like/expect, preferably in a realistic and constructive way. In order to do that, you need to take in account the tons of variables, shortcuts, exploitability and balance. There's also a difference in AI behaviour regarding on the battle map and on the global map, in both peace and war times.

I also think that if you let the AI be so far, it has a general idea of what unit should go against what. They have spearmen chase cavalry, cavalry avoid spearmen, skirmishers targeting heavy infantry etc. After a certain while, no AI partner wants to make any deals with you unless they are in a very tight spot themselves or unless you pay a lot, which indicates some intelligence (code). I also got the idea the AI (in war situations) got more agressive since Empire on the map as in most cases, you could just wait and see and possibly face a stack at your settlement once in a while if you didn't do anything.


I would like most changes in the AI to be on the campaign map. I think a bit more dynamic and adapting wouldn't hurt there and would be possible to code. If you grow too large, I wouldn't mind seeing the AI trying make deals with most of your neighbours like it did to some degree in some games. Shogun 2 just made every faction hostile to you the moment you hit a number which is lazy coding and could be exploited to a degree too. Yet, so far, I haven't seen the AI do anything with it; I would like the challenge of seeing a joint attack from the AI in the next game or this one. I wouldn't mind seeing them trying to attack with a real plan instead of turn by turn. So far, the AI would only attack adjacent territory (except in crusade situations) which makes the AI predictable. Wouldn't we all be surprised if an AI faction and 2 allies would send 3 combined stacks straight for your main city? (Rome for example)

AI factions can get big, but rarely exploit it as they are still not too hard to bring down. That's partially because of their passiveness or mismanagement (the AI so far in most games has been unable to properly cope with larger factions). In Third Age Total War, some factions couldn't be allowed to grow too big or they would simply spam stack after stack. Of course, that was also because of some financial script that had a lot of complaints too.

AI's should also see the value of fighting a defensive battle as so far, they give a human player a distinct advantage. So far, it is way too eager to attack a player while he had prepared for it.

In battle, I would like to see an AI give up favorable terrain less quick as it has been possible to lure them out way too easy. In sieges, it should see the value of a bottleneck and use it.

At the last (for now), I would like to see the AI less exploitable in diplomacy. It's really easy to confuse the AI each time by blitzing a few settlements and then sue for peace (along with some extra money) and then start again once you're prepared. Also, in some cases, they are simply willing to pay way too much things a player needs as well, like trade rights or map information in the past.



Your input? But please do keep it realistic and don't expect the AI to react differently in each situation as it's nearly impossible.

[Stars-N-Stripes]

could I get the consolodated version of what you said please?

[SPAMbuca]

TL;DR version:

AI never has been perfect. What would you suggest to improve it. Keep it realistic and don't expect any AI to ever react to a specific situation. Then I gave my suggestions.

[Hav0k]

it's actually a topic i could talk about for a long time, to me the MOST important thing is the battle AI and how good or bad it is, generally speaking, Total war AI is not very good and usually requires mods to improve it, but with rome II, they have changed the way it behaves, and opens up the possibilty to make it the greatest AI in any total war game, In past games the AI Would use 'scripts' to move its army around and engage in battle, offering little diversity and a lot of repetiveness, in rome II the ai is none scripted, meaning it reacts to your actions rather than having it set out and be told what to do, it basically reacts to what you do rather than do what it has been told to do, so it is intriging to find out what modders and CA can do with this AI in future. at the moment though it is not the best but certainly not the worst

[SPAMbuca]

Indeed I forgot to mention all the AI's were repetitive and predictable and I actually gave up hope on that part. I'd just expected to discuss scripts that could put into place to give us more challenge.

[Robayus]

In normal open field battles I've found the Ai to be fine and they normally try to outflank me. Somtimes they'll try to find favourable terrain to set up after deployment. When I'm on seige defense though the Ai is really bizarre, like there's a missing line of code or somthing. The Ai will just rush forward untill they get to the city, then crossover their units switching left and right. Somtimes they'll take more than one lane but they always rush their units at once into my strong defense. Even when they could go around me easily. When I have walls they all try to rush through one gate at the same time.

[DCBenton]

The closest I have ever come to making any kind of AI, was playing Carnage Heart on the PSone. So I am not very experienced in that area. :D

The issue I have had with the AI, is that it seems to be too focused on the flags, even to the point of being suicidal trying to pursue them, making it easy to exploit. It should only be caring about them as a last resort.

[SPAMbuca]

Yeah, that needs to be fixed. But I'd rather keep the topic about the Total War AI in general.

[Kalvix]

Slightly off topic, my favorite example of bad ai/bad balance was in medievil 2... I can't remember when I realised it but at one point I found out a army of ONLY generals was the best, with about 200-300 men worth of generals you could slaughter 4000 men. They were just beasts

[SPAMbuca]

Yup. I actually found them more overpowered in Rome as these factions had less access to spearmen and the heavy cavalry wasn't as vulnerable in close combat. All I needed in Rome in the Julii campaign were a few stacks with 3 Roman generals in them.

[HerrPanzerCardinal]

first things first would be to make the Ai determine a flavour of tactics before the battle even begins based upon unit composition. Now I haven't played Rome2, I have been waiting for footage of Ai capabilities first. But it looks to be the exact same Ai in every other TW game. If the player has an archer heavy army, one that is really too range heavy for its own good, the last thing the Ai should decide upon is being passive. If the human is cavalry heavy the Ai should be more passive and careful to not open itself up. Yet this is what I see in multiple videos. The Ai divides his forces allowing them to be outflanked by cavalry or the Ai attacks piecemeal or doesn't move at all and human controlled archers just shoot them in the face until annhilated. I won countless battles in Napoleon with my standard army, not that is good, but the most interesting to watch, of 12 cannons, 5 infantry, 2 cavalry, and 1 general. Now coming into the battle the Ai should "decide", if I charge all at once, I might take heavy losses at first, but once I come hand-to-hand, the human can't surivive. Instead the Ai marches his forces forward, takes his time about it, occaisionally shifts units left and right and then sets up to volley fire in front of my cannon raking him with cannister shot. I once took out 5 full stacks in 1 battle with 1 full stack of this composition. If the Ai had just charged, my cannons would have maybe had 1 or 2 volleys of cannister and then he would have had my infantry pinned and units swarming over my guns with the first stack, let alone 4 more stacks afterwards.

I have seen Rome2 videos of the same thing but with archers instead of cannons. Full stack Ai armies of melee troops either taking their sweet time attacking, attacking piecemeal, or not attacking at all. A simple calculation at the start of battle would have the Ai being uber aggressive and attacking en masse to either engage my archers, or force them to run away while the human infantry is annhilated. It's not that a skirmish heavy army shouldn't be able to defeat a heavy infantry army, it's that the Ai doesn't even try the most basic attempt to win. Hoping the human controller archers run out of ammo is not a valid strategy.

[SPAMbuca]

Hmm, I get a different impression from the Empire game. Of course, you mentioned "Napoleon", but I reckon it to be the same game sort off. All I had to do was fight defensive battles, wait for the enemy to rush in to be shot to pieces by a line of canons and line infantry right behind it. I could ignore the enemy canons as all they did was use the standard shots.

[Kalvix]

Originally posted by Acelondoner:

The reason the AI is so bad this time round is cause CA have stepped up from having scripted AI as seen in all previous total wars to now having Dynamic AI.

Having dynamic AI means the AI will no longer be predictable but means it can suffer problems very easily.

The way to improve the AI is just to tweak it, believe it or not, the AI isn't actually broken, It's just getting confused. Tweak tweak tweak and this AI will be the best AI we have ever seen in any total war game, trust me.

hmmm if this be so it could actually be the best AI to date when they do finally fix it. Well thats what I'm going to keep telling myself anyway....

[SPAMbuca]

I've heard this before, but could someone explain me the main differences in layman's terms? I know the differences between AI's, but I'd like to know how the new adaptive AI works/calculates or should work.

[Kalvix]

Originally posted by SPAMbuca:

I've heard this before, but could someone explain me the main differences in layman's terms? I know the differences between AI's, but I'd like to know how the new adaptive AI works/calculates or should work.

Basically the idea is that it will react to that you do. So if you use a phalanx it'll try to flank it, if you use horses to flank it'll try and take them out then take advantage of you not having them. So more or less its still pre-programmed things it does BUT it choses which one to use based on what you are doing. THATS where it gets tricky as it the AI needs to be able to know what you are doing and thus how to counter it to work properly

Very hard to do and will most likely take a while to work properly but it is definitely a step in the right direction, would be so awesome if they could make it work properly with patching in this one but most likely it'll take a few iterations to work properly

[SPAMbuca]

Yeah, I got what it means to have an adaptive AI, but I'd like to see how the AI would calculate such choices. For example, when will it forfeit a profitable ground and when will an AI decide to retreat to safer ground or attack the human player.