The AI

**bbad89** · February 03, 2011, 04:44 PM

Does this mod come with it's own AI improvements or are there any compatible AI mods I can use with it?

**Hartassen** · February 03, 2011, 05:46 PM

TATW had some of the best AI you can find. If I'm not mistaken it uses both AI from the developers as well as outside AI improvements.

**Scouten** · February 03, 2011, 06:42 PM

even still it suffers from total war AI and anyone with tactical sense can destroy the enemy with minimal casualties

**Hartassen** · February 03, 2011, 07:00 PM

Originally Posted by Scouten

even still it suffers from total war AI and anyone with tactical sense can destroy the enemy with minimal casualties

Of course, no mod in m2tw can produce good quality AI. Modders are limited by what they can do themselves within the engine.

**bbad89** · February 03, 2011, 09:12 PM

God...CA screwed it up THAT much?

**Makrell** · February 04, 2011, 09:28 AM

Originally Posted by bbad89

God...CA screwed it up THAT much?

CA never screwed up the AI is good, and considering NO other game is like total war series they have to do mny changes in the AI, cant just copy what AOE did

**baz44331** · February 04, 2011, 03:31 AM

No im sure it can be inproved some how.

**Ariaci** · February 04, 2011, 04:11 AM

It has improved compared to vanilla. Further, the "AI" gets some bonusses etc.

My personal opinion is that you do not want an AI that would beat you as a player everytime, that would not be fun either. Anyway, AI is not perfect but decent in my point of view and has improved.

**Hartassen** · February 04, 2011, 05:52 AM

The problem with the ai is that it lacks tactical sense. Therefore it needs to be buffed with larger treasuries and garrison scripts and spawned armies etc.
It's sad to see the ai that has an inferior army to you just stand stll and take arrow fire after arrow fire. Or an AI that fights you on the bridge at osgiliath running in all his troop in one little bridge and being bombarded by 15 archer units while 5 gondor spearmen hold the bridge gap.

Or even worse.. an AI that loses it's first ram and then brings up the second, breaks the gate and only sends in one unit and tries to move the rest of his 10 units up the ladders.

My point is you can improve the AI's capacity to produce troops and troops of higher quality and the mix of their troops could be of better quality. But the campaign AI they have to "cheat" a little bit and also the battle AI is just retarded in most cases.

**baz44331** · February 04, 2011, 06:14 AM

yes nice points Hartassen. i totaly agree.

**Whitewalls** · February 04, 2011, 07:02 AM

Yeah every once in a while the AI will pull off something smart, but usually they make dumb mistakes. Seems to me that there's a problem when I can win with 1-4 odds, and I'm very much not a great tactician. Then again if it was 100% realistic I'm sure only the best players would actually win battles. .

**Dave85** · February 04, 2011, 08:34 AM

I can't believe that it is an issue with the engine...

i've seen many games such as Battle for middle earth series, the warrior kings series (evern older tha TW, and known for its intelligent AI) where the AI is quite clever...

Just some good programmers should sort it out!

Tis sad...

**cahaya** · February 04, 2011, 03:15 PM

Originally Posted by Dave85

I can't believe that it is an issue with the engine...

i've seen many games such as Battle for middle earth series, the warrior kings series (evern older tha TW, and known for its intelligent AI) where the AI is quite clever...

Just some good programmers should sort it out!

Tis sad...

Having a background in computer science and artificial intelligence, I can tell you this is more difficult than most people realize. There are three tracks in AI that could apply to wargaming.

One track is the expert system (ES), where human knowledge (a knowledge base) is recorded and then searched for according to situation. For example, cavalry are good vs archers but bad vs spearmen. Thus, the AI will choose to move the cavalry towards the archers instead. But should it move at all if there are spearmen nearby the archers? Maybe if the spearmen can reach the archers within say 15 seconds? Thus, it can be very complicated encode all of these 'rules' of engagement. As far as I can tell, this is the most common wargame AI method, an ES with an algorithm that drives it. (I'm not an expert on this specifically, so I could be wrong.)

Another track is heuristic best-estimate searching, the method similar (but not quite same as) Deep Thought in chess. What the computer does is try out a very large number of different moves (by both the computer and the opponent) and then evaluates the position after these moves. If the evaluation shows a line (set of moves) that is weak, it will heuristically search for a better line. For example it tries out the cav charge vs archers and gets a positive score of say +5 and it tries out the cav charge vs spearmen and gets a negative score of -5. Maybe a charge on swordsmen will get a score of +2. Maybe if it sits still while everything else goes on, it gets a +1. The hardest part is how to determine the evaluation method and score. Then the AI will delve further into all of those lines with the best positive scores until it finds an (not necessarily *the*) optimal move. Deep Thought in chess, as it turns out, does a full blast search of all possibilities, so many moves deep, so it does not apply heuristics to search down the more promising lines. In a real time like TW, I'm not sure a PC could keep up with all of the changes in the battlefield and at the same time apply heuristics to it. It already requires a fairly powerful computer to do it competently for a simple turn-based move game like chess.

The third track is the 'self-learning' system. It may blunder badly in tactical battles for a while (oops, cav vs spearmen leads to a bad result, don't do it again!), but has code designed to keep track of what works and what doesn't and then it improves its decision-making over time. Theoretically, this type of AI could be done in the lab prior to releasing the game, and then the self-learning results data/algorithms included within the game on the market. Note that the results are both data (information) and processes (algorithms), in the sense that it is a self-improving program while at the same time utilizes the data for this purpose. This type of AI is probably the most difficult to program and can be quite unpredictable in its results. Self-learning systems is still very much in its infancy, but this is ultimately the route I think would optimize wargaming AI.

So, I don't pity the AI. I just know that very likely thousands of hours went into programming and testing the gameplay AI and there's a limit in terms of time (release date!) and money (cost!) of resources that can go into it.

**Radzeer** · February 05, 2011, 10:19 AM

Originally Posted by cahaya

Having a background in computer science and artificial intelligence, I can tell you this is more difficult than most people realize. There are three tracks in AI that could apply to wargaming.

One track is the expert system (ES), where human knowledge (a knowledge base) is recorded and then searched for according to situation. For example, cavalry are good vs archers but bad vs spearmen. Thus, the AI will choose to move the cavalry towards the archers instead. But should it move at all if there are spearmen nearby the archers? Maybe if the spearmen can reach the archers within say 15 seconds? Thus, it can be very complicated encode all of these 'rules' of engagement. As far as I can tell, this is the most common wargame AI method, an ES with an algorithm that drives it. (I'm not an expert on this specifically, so I could be wrong.)

Another track is heuristic best-estimate searching, the method similar (but not quite same as) Deep Thought in chess. What the computer does is try out a very large number of different moves (by both the computer and the opponent) and then evaluates the position after these moves. If the evaluation shows a line (set of moves) that is weak, it will heuristically search for a better line. For example it tries out the cav charge vs archers and gets a positive score of say +5 and it tries out the cav charge vs spearmen and gets a negative score of -5. Maybe a charge on swordsmen will get a score of +2. Maybe if it sits still while everything else goes on, it gets a +1. The hardest part is how to determine the evaluation method and score. Then the AI will delve further into all of those lines with the best positive scores until it finds an (not necessarily *the*) optimal move. Deep Thought in chess, as it turns out, does a full blast search of all possibilities, so many moves deep, so it does not apply heuristics to search down the more promising lines. In a real time like TW, I'm not sure a PC could keep up with all of the changes in the battlefield and at the same time apply heuristics to it. It already requires a fairly powerful computer to do it competently for a simple turn-based move game like chess.

The third track is the 'self-learning' system. It may blunder badly in tactical battles for a while (oops, cav vs spearmen leads to a bad result, don't do it again!), but has code designed to keep track of what works and what doesn't and then it improves its decision-making over time. Theoretically, this type of AI could be done in the lab prior to releasing the game, and then the self-learning results data/algorithms included within the game on the market. Note that the results are both data (information) and processes (algorithms), in the sense that it is a self-improving program while at the same time utilizes the data for this purpose. This type of AI is probably the most difficult to program and can be quite unpredictable in its results. Self-learning systems is still very much in its infancy, but this is ultimately the route I think would optimize wargaming AI.

So, I don't pity the AI. I just know that very likely thousands of hours went into programming and testing the gameplay AI and there's a limit in terms of time (release date!) and money (cost!) of resources that can go into it.

Good points!

**muller227** · February 04, 2011, 09:22 AM

Originally Posted by Whitewalls

Yeah every once in a while the AI will pull off something smart, but usually they make dumb mistakes. Seems to me that there's a problem when I can win with 1-4 odds, and I'm very much not a great tactician. Then again if it was 100% realistic I'm sure only the best players would actually win battles. .

Depends on the faction you are using, 1-4 sounds about right when play say HE against OotMM. That is the Tolkien lore speaking if nothing else. No other BAI is trying to balance the lore driven issue of massive differences in troop quality between factions. Try those odds in reverse and you will likely get annihilated.

I tried changing recruit priorities for the OotMM to give them more wargs and archers this helped them against me but the Heavy dwarven infantry owned them in Autoresolve. This is the CA issue Heavy infantry dominates in AI vs AI. Yet elven archery reigns supreme on the battlefield when you fight the same battle yourself. Since these modders are going to extraordinary lengths to stick to the lore, change how you play bit and follow their lead it becomes much more interesting.

**Lemoniser** · February 05, 2011, 09:42 AM

This isn't quite off topic but it is close so sorry if anyone minds but ive been wondering for a while, does the ai in this get much money? and is it possible at all to bankrupt them? As I'm playing as High Elves and I've basically launched a seaborn invasion of Harad. Since I can't keep sending fleet after fleet down with men I took the southern most minor city and kept it and since then I;ve only been sending out small raiding armies to sack a city and destroy all the buildings i can and move on. By doing this I've kept my economy afloat during some rough patches, set their unit making capabilities back and I've been wondering if I've been hurting their economy? So would it be possible to bankrupt them through this or through this and say triggering the garrision script?

**mattgoby** · February 05, 2011, 09:48 AM

This isn't quite off topic but it is close so sorry if anyone minds but ive been wondering for a while, does the ai in this get much money? and is it possible at all to bankrupt them? As I'm playing as High Elves and I've basically launched a seaborn invasion of Harad. Since I can't keep sending fleet after fleet down with men I took the southern most minor city and kept it and since then I;ve only been sending out small raiding armies to sack a city and destroy all the buildings i can and move on. By doing this I've kept my economy afloat during some rough patches, set their unit making capabilities back and I've been wondering if I've been hurting their economy? So would it be possible to bankrupt them through this or through this and say triggering the garrision script?

The ai is nearly always bankrupt.
They recieve money from scripts everyturn so they can carry on building and recruiting whilst forgetting about upkeep.

**Lemoniser** · February 05, 2011, 09:59 AM

Originally Posted by mattgoby

The ai is nearly always bankrupt.
They recieve money from scripts everyturn so they can carry on building and recruiting whilst forgetting about upkeep.

damn. i was afraid of that. i guess ill just have to destroy harad the old fashioned way. thanks for the quick response.