Why AI combat will never work in Civ 5

[gonzo562]

Nah I'm pretty sure the AI is actually horrible. They are easily baited, easily outmanuevered. They will suicide into set defenses all day every day.

I wouldnt say easily baited they got to go through my defense to take my cities. I would not say the AI is bad because their suiciding into my defense. Hell their forcing me to defend if i were to attack i would lose most of my units and they could take my cities

I been reading on the forums, that domination is easy, combat ai sucks, game to easy. I agree for king and lower difficulty setting I was leading in technology, building most of the wonders and had the largest military. I had technology advantage in warefare. Mech Infantry vs rifleman and lower.

Immortal Difficulty has been tough. This is one game that it will be hard to win by domination. England and Persia have bigger armies but luckily for me their not my neighbors. And the AI is either keeping up with me in technology or is slightly ahead of me

Anyways i am enjoying my current game

I am being forced to defend. If i were to go on the offensive it would be suicidal Also the minute men special unit is really bad. They move at 1 movement point for all tiles. They do not get the movement benefits from using roads

 

[ElephantRider]

a 20 year old strategy game that needed the only CPU available to do all the other in-game stuff isn't exactly the best example of what is possible with AI

I was just using that example to show that if they could do it then, they can do it now.

 

[Mîtiu Ioan]

So, CiV's AI will never be competent because some other game with the same format had different gameplay goal?

I'd like to be contradicted at this point ... but seems that actual status confirm my analysis/feelings ... also Heroes of Might and Magic suffer the same problem ( AI almost always used to have alot more units/skills in order to win a tactical battle ).

Sounds like the opinion of an expert.

Well ... if this is intended to be a sarcasm ... I never pretended to be an "expert".

 

[pi-r8]

The AI is actually pretty decent when it first declares war on you. It produces a big attack force and sends it over all at once.

It gets worse and worse as the war goes on, since its units tend to wander around your territory without attacking. It also tends to focus on attacking cities while ignoring your units in the field which is... not the best choice.

 

[Mivo]

To me The combat AI is not as bad as everyone make it seem to be. Levels king and lower was really easy. But immortal and deity I don't think its that easy

Giving the AI massive benefits and advantages doesn't make it "better". It's still bad, just can throw a lot more at you so that it may seem decent.

 

[Gismo]

I mean Ideas are great.. but it all comes down to implementation and it seems to me they didn't put near as much thought into that as they should have.

This one sums it up perfectly. One unit per tile "was" a nice idea but Firaxis totally failed at considering the consequences for AI or at least to counter that by putting "a lot" more effort into developing a good tactical AI.

What I'm wondering about: Did they intentionally ignore this issue? Hell, one official message from 2kGreg or anyone at Firaxis that they are going to heavily improve AI with next patches AND next expansion pack - I would be so much less disappointed with this game.

 

[klokwerk]

I think the tactical AI can be greatly improved.

We probably need using pattern recognition for unit placement (predefined patterns including opponent's units and player's units and defense values for hexes). The numbers of patterns needed are clearly huge. They should differentiate between melee and distant units too.

One of the benefit of using pattern recognition is we could get a good result without a huge CPU time between turn.

People who talked about a "chess like" AI and alpha-beta like algorithms are dreaming. You can't apply any "brute force" algorithm here or assess lots of future battle results, there are simply too many possible moves, it would take forever.

There are also several things which would improve the efficiency :
- focus attacks a bit more on units and less on cities.
- adding more naval help to take cities
- retreat individual units better and faster
- avoid city attacks unless you've got a mix of melee and distant, better yet a siege unit.

I certainly hope the BetterAI guys will work on that

 

[gonzo562]

Giving the AI massive benefits and advantages doesn't make it "better". It's still bad, just can throw a lot more at you so that it may seem decent.

I think the tactical AI can be greatly improved.

We probably need using pattern recognition for unit placement (predefined patterns including opponent's units and player's units and defense values for hexes). The numbers of patterns needed are clearly huge. They should differentiate between melee and distant units too.

One of the benefit of using pattern recognition is we could get a good result without a huge CPU time between turn.

People who talked about a "chess like" AI and alpha-beta like algorithms are dreaming. You can't apply any "brute force" algorithm here or assess lots of future battle results, there are simply too many possible moves, it would take forever.

There are also several things which would improve the efficiency :
- focus attacks a bit more on units and less on cities.
- adding more naval help to take cities
- retreat individual units better and faster
- avoid city attacks unless you've got a mix of melee and distant, better yet a siege unit.

I certainly hope the BetterAI guys will work on that

This picture has everything your asking for in my current game.
except for the retreating thing.
focus attacks a bit more on units and less on cities
- adding more naval help to take cities[/B]
avoid city attacks unless you've got a mix of melee and distant, better yet a siege unit.

Also the AI need massive benefits because say if i were playing King or prince level. Probably right now it would be my Infantry, or mech infantry or artillery crushing their weaker units. Which is not fun when my Mech Infantry is just crushing everyone because of my technology lead. I usually have a technology lead in difficulty's king and lower. At immortal were almost equal in the technology race.

But the types benefits the AI receives can be improved though

They AI might take my city but doubt it. Cannons are really strong for defending. I can almost one shot their rifleman and one shot their archers and take out 3/4 the health of their cannons. I just researched rifling so im gonna upgrade my rifleman and I am probably gonna buy some cavalry and take out their cannons. Plus policies and world wonders boost the strength of my units


I still think the AI is not perfect and can be improved. But i dont think the AI is extremely terrible as many people claim. Well the AI is terrible as difficulty king and lower
And pfft I doubt Firaxis can develop an AI that can crush me military wise. I am struggling to win this game but no way the AI will take all my cities and capital
My Brain > AI military tactics . I will probably lose this game via space race though. Persia way ahead of me in the technology race

 

[Slowpoke]

I agree, while the computer will never get good at mix of arms, there are a few very important easy things they can go to fix the A.I.

A) VERY SIMPLE: If X fraction of their army is defeated, without taking any cities:
1) call off the war if possible
2) play defensively: send all units to border until ready again
3) build archers/seige units and keep them in their cities

The A.I. cannot understand how to use a small force correctly, so the answer is to never use a small force.

B) VERY SIMPLE: Don't build non-siege ranged units for an aggressive force unless it is that civ's speciality (e.g. longbowmen). Quite simply the A.I. will never use these correctly. In a rock paper scissors system, rock and scissors is enough to counter. Alternatively, never use cavalry either: Horses get countered too easily since the A.I. can't handle spearmen. Siege and infantry should be enough.

C) VERY SIMPLE: Never retreat ranged units and siege units period, i.e. if there is a unit in range, fire. This is probably the biggest problem. Better to lose a unit or two than never make one friggin' attack.

D) Kinda simple: Send no less than X units to attack a city, where x is based on age and army size (typically very high). I see all the time the enemy spreading a bunch of weak junk attacks over 3 cities and failing miserably. This might be hard to code depending on how they did it, but it doesn't sound hard.



I think Firaxis simply focused too little on the obvious. These are mostly easy to code, and would help a lot. They should've assumed the A.I. was dumb, and coded workarounds.

 

[Mîtiu Ioan]

We probably need using pattern recognition for unit placement (predefined patterns including opponent's units and player's units and defense values for hexes). The numbers of patterns needed are clearly huge.

That's clearly the only acceptable ( in term of inter-turn time ) way to go .. and developer more likely implemented something like this already ... however this is exactly my point to refer to PG series : it will be very dificult ... maybe impossible to adapt those "tactical patterns" to a "strategical map" - how to value an excellent tactical defensive position, which rely on short lines, against a weaker one but which allow to protect also a strategic/luxury resource for being pillaged ? And so on ...
BTW - I suspect ( from my programming knowledge ) that "already in use patterns" are responsible for decent manner of troop placement when AI starts a war.

There are also several things which would improve the efficiency :
- focus attacks a bit more on units and less on cities.

Some sort of preference for "decisive single battle" as von Clausewitz theoretized ?
This could lead to an exploit in which human players will mass their troops in easy defendable position ( neglecting defense of cities/resources ) and AI will blindly throw most of its armies against such heavily-entranched defense ...

- adding more naval help to take cities

Take care what you wish !
IMHO Call To Power became quite "unplayable" due to increased significance of navy ... but that's my personal opinion, of course.

- avoid city attacks unless you've got a mix of melee and distant, better yet a siege unit.

Yes - but this issue seems to reflect more a general lack of "countryside pillaging" significance ...

Anyway - it's great to try to summarize some ideas and to analyze their pros/cons.

 

[salaminizer]

I wonder what would happen with AI if we removed ranged combat (or just switched every ranged unit to melee)

 

[r_rolo1]

I still think the AI is not perfect and can be improved. But i dont think the AI is extremely terrible as many people claim. Well the AI is terrible as difficulty king and lower
And pfft I doubt Firaxis can develop an AI that can crush me military wise.

Given that the Ai is the same in every level ( more or less handicap ) you are saying that the Ai is terrible unless you give it a huge starting advantage + cut down costs in science, inflation , upgrades and rush buy....

Basically you are saying the same as everyone else

 

[troytheface]

easy fix for the AI

have dumb units in front and siege in back

and move them correctly

it should work just fine. course i can have like almost no units and stop huge invasions right now-
but attacking is still hard

kinda as demonstrated by that screenshot

i need to do a screenshot

 

[Mahi]

The problem with the AI is mostly the offensive moves as it has already been claimed. More precisely the way it handles invasions. It seems to me that the AI leaves it's own borders with a fairly decent formation but if that invasion force encounters something unexpected, like a small military force or a complex bottleneck on it's way to the target the AI's formation and logic just crumbles into dust.

I too have my doubt that the AI will ever be able to be a strategic offensive challenge with 1UPT. It can cause troubles, yes but it does have a really hard time finishing the job and that's when it is destroying itself.

 

[ohioastronomy]

The problem is one unit per hex, period. It makes pathing impossible and neuters the AI advantage of superior numbers, as well as promoting completely cheesy and artificial moves. Even changing it to 2 would help tremendously by preventing complete logjams.

Until they fix this foolish and basic mistake nothing else will materially change things.

 

[TheLean]

The AI needs to be better at handling invasions. I had open borders with germany on Prince level. I trusted Otto, who is a bastard btw, and he attacked three of my cities with about three units for each city. Bad idea. No cities of mine were captured and I am kicking his ass now. Now the coward wants me to take all his cities except for his hiding place in Berling. Weakling.

 

[Iberian]

Currently no AI will not stand a chance against a human player in a strategic game with the number of options CiV has. At least not on a desktop computer with < 30 seconds to calculate.

Given that the AI will need a bonus to counter the human IQ advantage. I know we hate that the computer cheats and gets free units, can see the whole map, has free upgrades, but honestly it has to have something to help it compete.

Last the AI in CiV does suck. It doesn't change just because you change difficulty. The only thing that changes are the bonuses to my second point. Minus the map thing. Which they should give it to make it fair. If you want proof go play a diety 1v1 on a small map and rush the AI.

 

[jtwood]

I certainly hope the BetterAI guys will work on that

My heart breaks every time I see someone say that they hope modders will fix something.

We paid good money to a company, and here we are sitting around hoping that some unpaid people will make the game right out of the kindness of their heart. That seems inherently very wrong to me. Modders, in my opinion, should be out to make the game more variable: Want a WWII Civ5? Get some modders together. But fixing the base game? It's so wrong that we place that expectation on them.

 

[klokwerk]

@jtwood : The thing you don't see is there are only so many resources a dev can use on a game. There are time constraints and money constraints. The mod community has infinite people, infinite time.

There's not a single chance in the world that a AI developer (even talented) can do better than a community of thousands of people combined. Plus, Firaxis did a good job here. The AI, while not building enough units, is already competent. I saw a 15+ units attack on my city at Prince Level. I saw mistakes too, but it's definitely not broken at all.

It's just not possible for a single dev to do what a community of thousands of people can do... Humans can build great things when they're working together in thousands for a very long time.

 

[jtwood]

If that's the case, then the game should endure an extended beta where talented modders and expert players are invited to spend their free time offering free work.

It shouldn't be done after people have shelled out retail price for what should be product free of obvious problems.