Yes! Finally! The A.I. beat me tactically!!!!

[Prussian to the Iron]

I have never been so thrilled to lose in my life. Playing Prussia, I had conquered all the Austrian territories surrounding the capital. They still had 2 or 3 regions next to the ottomans, but they were drastically underdeveloped. Now, I had recently been forced to make some budget cuts, and had only a unit of smurfs in 1 city, and 1 unit of landwehr militia in the other city.

A couple raiding parties attacked, but no sieges. Then out of nowhere, stacks of 3-4 units of engineers and fusiliers laid siege to both towns. I was destroyed, hands down. Austria had basically taken back all of its lost territories south of Prague in 2 turns.

The A.I. had actually managed to:

1) Find a weak spot in my defenses. I know it wasn't dumb luck, because Prague was garrisoned, and not attacked (located directly north of Vienna, their capital).

2) Assemble it's forces. With this, I'm not sure if it was simply another couple unit stack standard, or if it decided I didn't have enough men to warrant a larger force. I'll never know.

3) Choose the perfect time to attack. I had just recently moved my main (elite) army out of the area, due to a suicidal french invasion (BTW- they took back Alsace-Lorraine, which i had been given for Corsica, and Flanders, then promptly received a thorough ass-whooping. Ironically enough, the French king is now isolated on Corsica ). By the time I had my army back in Austria, they had already established total dominance.

4) Co-ordinate with my enemies, the French. As they say: The enemy of my enemy, is my friend. They could not have exemplified this more. As I brought my Main (elite) army-filled with 8-inch mortars and Swiss guards-to attack Paris, after re-taking my possessions north of France, I was stunned to see that, not only would I have to fight a 14 unit French army filled with West European Mercenaries (An extremely deadly unit, but with a high upkeep), but their reinforcements of Austrian line infantry and Demi-cannons. I did manage to defeat hem, but the casualties were extremely high, especially for my 6-chevron left flank, who were left with only 35 men and 67 men, both down from 160.

5) Synchronized attacks of naval forces and ground forces. As their ground forces re-took their lands, I found myself having to fight against several galleons and a 4th rate. Quite a bit of trouble for 5 fluyts.


As you can see, even on easy campaign difficulty, the A.I. seems to be getting better by the day. Drastically better than day 1.

[Veho Nex]

GUYS I FIGURED IT OUT!! WE'VE BEEN PLAYING ON THE WRONG DIFFICULTY...

No more Vh/Vh for me, its all Easy/Easy from here on out.

Edit: Note this is in no way made to be mean towards the OP just a rant about how easy the game is on VH VH

[FactionHeir]

You are playing with a total conversion mod, that is a completely different animal than what CA gave us. Keep that in mind.

[Prodigal]

The OP is playing the conversion mod?

[Veho Nex]

I wasnt aware the TC mode was entirely up to date. I didn't think any one was playing it.

[antisocialmunky]

If the AI suicidally keeps throwing units at you, eventually enough of them will coincide to look intellegent.

[Slaists]

You are playing with a total conversion mod, that is a completely different animal than what CA gave us. Keep that in mind.

Yup. I agree. For example, Imperial Splendor campaign AI is much better than the one of vanilla. I am surprised about the campaign difficulty (easy) though.

On VH (in 1.3), the AI:

1. has much faster research times than the human player
2. has much lower upkeep for its units than the player
3. has constant cash infusions to keep it afloat
4. can move its troops much farther in a turn than the player
5. has cheaper building times

and probably some other bonuses too.

All of the above should give it at least some edge. My understanding is that on Easy, it's the player that gets the advantages (not the cash infusions though).

[Prussian to the Iron]

You are playing with a total conversion mod, that is a completely different animal than what CA gave us. Keep that in mind.

actually i am not. the only mod i am using is the AUM mod, which simply gives everyone more units. it doesnt do anything else.

GUYS I FIGURED IT OUT!! WE'VE BEEN PLAYING ON THE WRONG DIFFICULTY...

No more Vh/Vh for me, its all Easy/Easy from here on out.

actually, i play on easy/VH (that is to say, easy campaign, very hard battles). i find it difficult to keep money coming in anyway. in my current campaign, for like 10 turns i was getting like 5-700 a turn.

[Elmar Bijlsma]

Tactics... Strategy... Except for being different, it's all the same, eh?


This is a textbook example of damning with faint praise.
That a player of a strategy game feels the need to come here and tell us that ETW can actually mount a successful attack pretty much says it all, doesn't it?

[Prussian to the Iron]

not necessarily. it's just that lots of people have been complaining (most notably sturmwolf lately), and I feel like they should be shown that the a.i. is more capable in certain situations than thought.


wouldnt it be hilarious if the a.i. was programmed to be uber easy until like setember, then it destroyed the player in 4 turns every time on easy? :D

[Durallan]

that would be a cruel and vindictive joke, and it would be amusing if they did that lol

[Prussian to the Iron]

inorite?

[Elmar Bijlsma]

not necessarily. it's just that lots of people have been complaining (most notably sturmwolf lately), and I feel like they should be shown that the a.i. is more capable in certain situations than thought.

That case for capability has as yet not been made, TBH.
Having witnessed the dismal ability of the AI for these past few months, someone else saying they saw it do something cunning once isn't enough to suddenly change my mind about how woeful it is performing. Even a broken clock is right twice a day. The AI doesn't even get near that low standard, IMHO.

I don't ask for an AI to be in the same league as me. But could it please, at the very least, play the same sport?

[Durallan]

like someone very very smart said in a topic in a galaxy a long long away time agoes said through a parallell universe,

"They need to design the Artificial Intelligence of this game to make to so it can WIN the game, instead of TRYING to make the player LOSE."
Like most? ( there could be more excpetions to the rule than I am aware of) strategy games, the artificial intelligence is designed so it can play and win at the game itself and then attempt to defeat a human opponent.

One of the BIGGEST problems in this game, is the fact no AI nation will send trade requests, in my last great britain campaign, i was 1 out of 3 nations that traded with more than 2 nations, and 3 out of however many nations are in the game to trade at all with another nation (mostly them trading with me)

[AussieGiant]

RANT ON:

And I once said the following:

If CA could create artificial intelligence that could control an AI on 22 individual levels in a game this complex, they'd be £$%^ coding algorithms for banks on the stock market!!!

Or perhaps writing 10 million dollar military code for the defense industry.

They certainly wouldn't be £$^ing around trying to satisfy a bunch of gamers for 50 to 100 bucks!

Can every please get a perspective check on this.

Good lord.

RANT OFF:

Have a nice day everyone.

[Elmar Bijlsma]

And as I said, in this very thread:

I don't ask for an AI to be in the same league as me. But could it please, at the very least, play the same sport?

I don't think many here have expectations that the AI should be outwitting them. But it needs to have at least the vaguest clue as to what it needs to be doing. At this point it's just dismal at it's job.An AI capable of not breaking suspension of disbelief, that's what I'm after, and so are most people I would think.

An AI where people come to report that it's not all stupid all the time isn't what we should be satisfied with.

[AussieGiant]

And as I said, in this very thread:

I don't think many here have expectations that the AI should be outwitting them. But it needs to have at least the vaguest clue as to what it needs to be doing. At this point it's just dismal at it's job.An AI capable of not breaking suspension of disbelief, that's what I'm after, and so are most people I would think.

An AI where people come to report that it's not all stupid all the time isn't what we should be satisfied with.

You've just described what the military anywhere on the planet would pay about 50 million dollars for, at a bare minimum.

Intelligence operative at a Intelligence Sub-committee:

"CA is able to code a socio-economic-strategic-tactical war simulator that can suspend belief in the average human player."


Chairman of said sub-committee.

"Right then, have them sign the appropriate waivers, disclaimers and their 2nd testicle, AND GET THEM IN HERE NOW!!"

So I'll say it again.

Perspective.

[Zenicetus]

If CA could create artificial intelligence that could control an AI on 22 individual levels in a game this complex, they'd be £$%^ coding algorithms for banks on the stock market!!!

(snip)

Can every please get a perspective check on this.

The only perspective needed, is a comparison with the AI in other strategy games on the same platform, selling for a similar price. We don't need to compare it to megabuck Defense Dept. wargames.

I've seen better map-level, strategic AI in other games. I think most of us have. I don't think it's too much to ask, that a game like this deliver decent map-level AI, when other game companies can do it*. There is also the not-so-subtle point that CA claims in their advertising that yes, they are selling a game where the AI can control these factions.

* Before someone else makes this point: It's true that strategy games aren't 100% comparable, since they have different feature sets. A game like GalCiv2 with no tactical battle AI can have more resources thrown at the strategic map-level AI, and there are fewer factions to deal with. The first TW games with Risk-style maps were no doubt much easier to code AI for. However, CA is still responsible for the decisions it makes on current feature set vs. the ability of the AI to handle those features.

FWIW, I think it's possible that CA can whip the AI into shape. All the pieces seem to be there, it's just a question of whether they can hit the balance between aggression and passivity, and make diplomacy feel more realistic and predictable. The time for true ranting would be if we reach the point when CA says they've released the last patch for the game, if it's still in a similar state as it is now. We're not they're yet.

[AussieGiant]

I agree, that we are not there yet.

And while some games are similar, nothing is doing exactly what CA is doing with this series.

I mean just comprehend what moving to a digitized map means to AI...you have a million times a million individual spots on this map...try and code unit movement...just that is quite a task.

Then try and have 25 different AI's with 25 different set of unique goals and 25 unique sets of economic, strategic and tactical factors to action PER turn. These then change from turn to turn and must be re-evaluated, reconsidered and actions taken.

Then times that by 200 turns.

Comparing this game to anything else is not apples to apples. At that point comparisons become less than relevant.

[Elmar Bijlsma]

Actually, the fact that it isn't apples with apples is a large part of where CA went wrong.
I not only think too highly of the AI, I don't think too highly of the enviroment it is placed in. Even solid AI would be in trouble, I suspect. It's too much

Since STW and MTW lots of features have been added that looked good on paper but was too much for the AI to handle. It's time for CA to consolidate. Make the AI better able to deal with the environment, and tailor the environment to the capability of the AI were it is out of it's depth.
Some good features need to be dropped I fear, simply because it's beyond the AI we've got now. As basic as the old provinces based movement was, it worked. We needn't go all the way back to that but it does illustrate, less can be more.

[Zenicetus]

Actually, the fact that it isn't apples with apples is a large part of where CA went wrong.
I not only think too highly of the AI, I don't think too highly of the enviroment it is placed in. Even solid AI would be in trouble, I suspect. It's too much

It looks complicated on the surface, but under the hood, many things have actually been simplified, compared to the earlier TW games.

There is no Pope to deal with, and religious differences aren't that big a deal anymore (annoying, but not a central consideration). Unrest in newly conquered lands is no longer tied to distance from the seat of a kingdom. Trade just means moving ships to trade points, instead of having to pathfind merchants around the campaign map. Diplomacy has been streamlined; you can contact any faction at any time now, without having to move diplomats around the map to make contact. The fog of war is much less a factor now, due to the way trade lanes keep distant areas open, and all those extra towns and resources keep more of your provinces visible (notice how they ditched watchtowers). Reinforcements have been streamlined, with generals now able to recruit and build units in the field. Remember in M2TW how the AI would build crazy armies like a stack of artillery and nothing else, just because the nearest city was dedicated to arty and everything else was too far away? That's been greatly improved by field recruitment.

All this considered, there are many things that both the player and the AI have a lot easier in ETW, compared to M2TW and RTW. I agree that there are some new additions that are problematic, like the way towns and resources can be attacked separately from main cities. The AI isn't very good at distinguishing (and building) small raiding forces vs. large conquering armies. The balance of naval invasions is still a problem. The AI doesn't deal well with port blockades (although that's not exactly a new feature in ETW). But overall, I would think it would have balanced out somewhat, with new complications being offset by all the places they simplified the earlier design.

Making things too simple can cause problems too, and I'm wondering if (in addition to the timespan of the game being so short) this isn't another aspect of the wacky diplomacy.

In previous games you had to explore, open the map, and send diplomats marching around to make contact. Until you reached the point where you had a diplomat or princess parked at the cities of other factions, or at major intersections like Rome, you couldn't just open the diplomacy dialog on every turn. Now you can do that starting on turn one, and so can the AI. I wonder if that might be "heating up" the diplomacy too much for the AI to handle. The fact that you actually had to move diplomats around to make contact, was a sort of built-in brake on stirring the diplomatic pot too often, and too early, in the previous games.

[AussieGiant]

Actually, the fact that it isn't apples with apples is a large part of where CA went wrong.
I not only think too highly of the AI, I don't think too highly of the enviroment it is placed in. Even solid AI would be in trouble, I suspect. It's too much

Since STW and MTW lots of features have been added that looked good on paper but was too much for the AI to handle. It's time for CA to consolidate. Make the AI better able to deal with the environment, and tailor the environment to the capability of the AI were it is out of it's depth.
Some good features need to be dropped I fear, simply because it's beyond the AI we've got now. As basic as the old provinces based movement was, it worked. We needn't go all the way back to that but it does illustrate, less can be more.

I agree Elmar Bijlsma.

That's exactly what has happened, but we are talking about AI. Something that is so hard to create and develop that world class research and scienfic organisations can't do it.

Let alone a games developer.

[Prussian to the Iron]

which is why it is pointless to complain about it:

artificial intelligence is just that: artificial. created by humans.
if you would like, think of it like this:

By certain religious beliefs (which I don't agree with), god created man. now, whether god or nature created man, both are extremely dangerous and all-powerful (except nature, starting 2,000 years ago). Man is trying to create what it took 6 billion years to create: intelligence capable of processing facts, weighing options, making decisions, and learning from its own, as well as others, mistakes. it takes a lot more than 20 years to do that.

honestly, i think we would be better off being only allowed to play against other players in campaign. not for every faction obviously, but perhaps 4 human players and the rest a.i. which take general orders from the human players (ex: playing as GB, i could tell the 13 colonies to do basic actions, such as committing a certain number of troops to a certain faction, or helping me invade, with a penalty if not completed.)



sorry, i had to get that out.

[antisocialmunky]

Okay, off topic rants aside.

Yeah, the increasing complexity of the TW games probably has something to do with it but there's nothing hard about adding a virtualization level between the AI and the game so the AI gets simplified versions of variables.

That still doesn't explain how the TW Diplomatic AI has gotten less and less rational over the years. All we want that to do is approve/decline trade agreements depending on how they they like/are scared of you and ask for peace if they are losing a war and declare war if they view you as weak vs them.

[AussieGiant]

Okay, off topic rants aside.

Yeah, the increasing complexity of the TW games probably has something to do with it but there's nothing hard about adding a virtualization level between the AI and the game so the AI gets simplified versions of variables.

That still doesn't explain how the TW Diplomatic AI has gotten less and less rational over the years. All we want that to do is approve/decline trade agreements depending on how they they like/are scared of you and ask for peace if they are losing a war and declare war if they view you as weak vs them.

So whip up some code and send it through.

Or better yet, whip up a technical requirements document outlining what you want and send it through to Jack.

It's not easy. In fact it's incredibly complicated.

[antisocialmunky]

I know. Yes a good AI is hard, but these things worked in STW. Why don't they work now?

I don't like words. I rather just build one myself and post it.

[AussieGiant]

I know. Yes a good AI is hard, but these things worked in STW. Why don't they work now?

I don't like words. I rather just build one myself and post it.

Why? Well, how about the map as a starting point.

STW:

Design an AI tactical decision making tree for units that can move from one tile (province) to another from perhaps 8 hexagonal directions. Total number of combinations 40 x 8 x 8 = 2560

ETW:

Design an AI tactical decision making tree for units that can move from one point in a 1 000 000 by 1 000 000 digitized map from perhaps 8 hexagonal directions.

I don't even want to do the math, and, just to top it off, add in the "ocean" maps for a top up.

This is what you have to do just to get the AI in 25 different nations to move their pieces (armies, agents, etc etc) in a rational, intelligent fashion.

Comparing the complexity of STW to ETW is like comparing a T model Ford from 1922, to a 2009 Formula 3 Grand Prix Car.

Keep in mind I only have a vague awareness of what is required from my prefessional experience.

For the average punter, there's stuff you would even know you needed to know, in order to get this stuff to work.

I'd love Jack to just chim in with a mathematical equation to show what the real numbers might look like.

[antisocialmunky]

.. Dude, I'm not talking about tactical AI. I'm talking about diplomatic AI.

Also you are misrepresenting the problem - pathfinding is not a tactical AI:

1)STW and MTW was some sort of simple graph based system for moving to adjacent places. This is very simple move along edge to vertex sort of thing.
2)RTW and MIITW used some sort of tile based A* varient for its pathfinding.
3)ETW I'm not sure how but you'd probably use some sort of random forest algorithm to calculate pathing or some sort of space body varient A* thing. I'm not entirely sure, its fairly impressive that they can path that quickly.

Those are the movement schemes.

Now the tactical AI for the STW and MTW was this: Look at your map, if you are at war and you have enough armies in adjacent provinces attack.

Now, would you be surprised if I told you that THIS IS STILL HOW THE AI PLANS? Yeah, the AI still plans this way because TW has always been based around the adjacent province structure*. Except with RTW and beyond, it plans via provinces and then it when the army enters a pronvince it has to pick a target like forts, watchtowers, armies, or the settlement.

So, to review, the AI:
-Picks an adjacent province
-Picks a target in that province to attack.

That's all there is to it. You want to know why they keep raiding your farms in ETW? This. Its army is can't take the settlement or win a battle agaisnt your army so it attacks your farms. If the AI massed units effectively, this wouldn't be the case. That's where one of hte the problem lies, the AI jsut doens't mass effectively sometimes. For some reason it sends armies on missions before they can mass. Shouldn't be too hard to fix TBH.

*This is incidently why the AI was naval invasion happy in other TWs because it always counted all coastal provinces as neighboring so the planning AI got borked hard because people tend to not garrison them that well so it thinks it has a chance.


EDIT: This is all based on what I've been able to deduce. I could be totally wrong about RTW but I'm fairly sure my STW analysis was fairly accurate.

[AussieGiant]

Well it seems like you know more than me.

So it shouldn't be too hard to work out the complexity differences between STW and ETW.

And movement is related to both Tactical and Dipliomatic AI coding. If they weren't you'd have major problems. These functionalities must interact with each other.

And yeah, things seem real easy when you can't see the whole picture.

There smart guy's. I assume they read these threads and have a real laugh at the fumbled attempts we make to "guess" at what their problems are.

[dismal]

Why? Well, how about the map as a starting point.

STW:

Design an AI tactical decision making tree for units that can move from one tile (province) to another from perhaps 8 hexagonal directions. Total number of combinations 40 x 8 x 8 = 2560

ETW:

Design an AI tactical decision making tree for units that can move from one point in a 1 000 000 by 1 000 000 digitized map from perhaps 8 hexagonal directions.

I don't even want to do the math, and, just to top it off, add in the "ocean" maps for a top up.

The problem with this is that 99.678787% of the possible options should be ignored.

Just as the human player ignores 99.76737% of his potential moves.

The AI needs to be able to assess threats and opportunities. When it identifies a threat, it needs to respond. When it identifies an opportunity, it needs to develop a reasonable plan to take advantage of that opportunity.

I would guess that almost all human players go through a similar approach to doing these things. Spot a potential invasion force? Concentrate some troops to defend. See a weakly defended and or economically desirable city? Concentrate some forces and attack. Etc.