Why Civ 6 AI is terrible - Part 1: Wonders

[Kwami]

That is a valid point, but if you haven't already, check out the Civ5 Vox Populi combat AI. If Civ6 combat AI could even be half as good, I would be happy.

How are turn times with VP compared to Civ V without the mod?

 

[Ryika]

It pushes you ahead relative to where you were, but not relative to competitors who are using their resources more efficiently.

So?

The relation of strength between an AI who is getting a production bonus and a player who is playing well, depends entirely on the size of the production bonus and the skill of the player.

An AI that is making bad decision will lose against a bad player and a good player.
An AI that is making bad decisions, but has a <small> production bonus will win against a bad player and lose against a good player.
An AI that is making bad decisions, but has a <large> production bonus will win against both.

Obviously that's not ideal as I said, but it's an easy way to compensate "behind the scenes".

Here you're making an artificial distinction. The worst result of bad combat AI is that the AI loses units, which requires production to compensate by allowing it to build enough units to make up its losses. Just as you say, that's the same as needing production for extra buildings.

Not true at all. Bad combat moves are visible whenever you're in combat with an AI. It's there, right in your face. That's simply not true for production priorities, it's "hidden" from the view of the player. The only time you will see it, is when the AI builds wonders that have almost no effect, or when it Sim City's like it doesn't understand the basic concept of adjacency bonuses.

And as you argue, the extra cities the AI has provide more slots for unit production, especially as AIs love to spam encampments. The entire reason Civ IV was better at combat than Civ V was that spamming excess units was a surer route to victory with stack combat, while in Civ V/VI simple space requirements on the map give diminishing returns for having too many units in play at once. That's almost purely a result of production bonuses rather than combat ability.

Precisely BECAUSE spacing on the map limits the usefulness of large amounts of units, the focus should be on combat, not on production priorities. Because the AI will not be able to make use of the latter due to the problem you mentioned - no space to even make use of the many units - but it WILL benefit from being better at fighting with these units.

My point was that the existing production boosts help it do one or the other, not both, and that it's simply easier to correct for poor combat AI by spamming extra units than it is to correct for poor building AI by allowing it to build more stuff, when that stuff will often be unnecessary holy sites, excess spaceports in low-production cities, aerodromes and air units they won't use, or extra Wonders. There's far more of this wasted effort than just situational Wonders (and many more situational Wonders than you're crediting. How often do you see Jebel Barkal built by civs that have no need for any of tourism, faith or iron, say?). When you have an AI that's bad at both building and combat it's pulling the production bonus too thinly, since it needs to compensate for both, and is putting all its bonuses into treading water rather than actually increasing the difficulty for the human opponent.

The reasoning here is not: "The AI has all of the extra production, but it does not make use of it, so that's not a solution!", the reasoning is: "The AI is bad, so we have to give it extra production so it can produce silly things without taking too much damage from it." You can almost scale that however much you ant, if the AI doesn't manage to do a good-enough job to be a challenge, then Firaxis simply hasn't pushed that number high enough.

I mean, you can try it yourself, right? Quadruple the AI production bonuses in the game files and see what difference it makes.

 

[PhilBowles]

So?

The relation of strength between an AI who is getting a production bonus and a player who is playing well, depends entirely on the size of the production bonus and the skill of the player.

An AI that is making bad decision will lose against a bad player and a good player.
An AI that is making bad decisions, but has a <small> production bonus will win against a bad player and lose against a good player.
An AI that is making bad decisions, but has a <large> production bonus will win against both.

Obviously that's not ideal as I said, but it's an easy way to compensate "behind the scenes".

It's a very difficult way to compensate behind the scenes, since to balance this effectively you need an idea of the average efficiency of both the AI and 'good' players.

If an average player gets, say, 350 production-equivalent back for every 700 spent, and an AI gets 300 back for every 700 spent on average, evidently the AI will fall behind. It should in fact be easier simply to code a calculation of this sort into AI priority setting ("if I look ahead, how many resources do I get over X turns after completion for spending Y resources?") and positively weight the results that produce the greatest efficiency.

Calculations and futurecasting based on deterministic features like tile output are things AIs are very good at, so this should be easier than simply coding useful production bonuses while producing superior results.

Not true at all. Bad combat moves are visible whenever you're in combat with an AI. It's there, right in your face.

Sure, it's visible, which is why it gets all the complaints, but that doesn't equate to it being more of a fundamental problem. Why are those combat moves bad? Because they result in the AI losing more units at the cost of fewer of the player's - when you get down to the fundamentals that's all that matters, so upping unit production is an appropriate fix. If combat AI were to reach some hypothetical optimum, that's the only calculation it would ultimately affect. A production boost is not an ideal fix, any more than a production bonus to construction is ideal, but enough to produce good results.

The AI in Civ VI is presently better than it was at release mostly because it now builds more units - there have been some improvements to its combat performance, but they're minor in comparison. While Civ V certainly had superior combat AI it was as much better than Civ VI's as it was mostly because it simply produced more stuff. However well the human player uses units, they're going to take damage and enough enemies suiciding into them will overwhelm them. Civ VI at release was far too passive because it was concerned with preserving its units, and it remains reluctant to attack even cities it can take if it may cost it a single melee unit. Civ V would just suicide things at you until you were dead - which among other things opened up space to use those extra units.

The reasoning here is not: "The AI has all of the extra production, but it does not make use of it, so that's not a solution!", the reasoning is: "The AI is bad, so we have to give it extra production so it can produce silly things without taking too much damage from it."

I appreciate what your reasoning is - it's simply not good reasoning. As above, to adequately compensate for poor AI taking this approach requires much greater knowledge of play patterns than arbitrarily changing numbers until you hit something that happens to work in a playtest group.

 

[Ryika]

It's a very difficult way to compensate behind the scenes, since to balance this effectively you need an idea of the average efficiency of both the AI and 'good' players.

It's not hard to balance that at all. You just put a percentage there and see what happens, that's what difficulty settings are for. Players easily beat Deity? Increase the percentages, and there you go. Did you overdo it, and now almost nobody can beat Deity anymore? Well, that's not even a problem, after all, there are difficulties below that.

It should in fact be easier simply to code a calculation of this sort into AI priority setting ("if I look ahead, how many resources do I get over X turns after completion for spending Y resources?") and positively weight the results that produce the greatest efficiency.

That's nonsense. First of all, how do you even calculate that theoretical yield if it involves working tiles? What if there are stronger tiles available, do you factor those in? Or do you just assume that the AI is going to work all of the wonder tiles? Do you consider tiles that are going to be part of the city soon? If so, do you factor in border expansion speed? Do you consider the size of the city, and that it might not want to work the yield type that the wonder (or building) adds right now? What if it's a wonder that will become really good in that city, but only after it has grown?

You will simply not get "good" results without making extremely complicated calculations, which have to be done again and again and again for x buildings in y number of cities. You could get "better" results by making the AI check the tiles that it has available, but you would not get "good" results. Production bonuses would still be necessary to make up for suboptimal solutions.

Sure, it's visible, which is why it gets all the complaints, but that doesn't equate to it being more of a fundamental problem.

It totally does. Gaming AIs are bad and always do stupid stuff, it's all about hiding it so you don't notice.

 

[V. Soma]

Maybe it would be easier to define reasonable requirements for wonders than try to teach AI to understand complex game mechanics like wonder effects. Define the requirement in a way that you can only build a wonder in a city when you get an advantage from it.

Example : Chichen Itza
Make the requirement to have 3 or more Rainforest Tiles in city area.

Nice general concept for a mod...

 

[Gort]

I always wonder if Firaxis would be better off just abandoning the concept of an AI that plays the same game as the player. Civ 7 just has the player alone on an empty map, then 1206 AD hits and Genghis Khan spawns East of their empire with 40 screaming hordes.

 

[PhilBowles]

It's not hard to balance that at all. You just put a percentage there and see what happens, that's what difficulty settings are for. Players easily beat Deity? Increase the percentages, and there you go. Did you overdo it, and now almost nobody can beat Deity anymore? Well, that's not even a problem, after all, there are difficulties below that.

If it were that easy Firaxis would have done a better job with Civ VI than they have. These games don't even have external playtesters, so the best this arbitrary approach can do is find a difficulty that beats game devs who don't actually play the game beyond the amount necessary to tune it.

That's nonsense. First of all, how do you even calculate that theoretical yield if it involves working tiles? What if there are stronger tiles available, do you factor those in? Or do you just assume that the AI is going to work all of the wonder tiles? Do you consider tiles that are going to be part of the city soon? If so, do you factor in border expansion speed? Do you consider the size of the city, and that it might not want to work the yield type that the wonder (or building) adds right now? What if it's a wonder that will become really good in that city, but only after it has grown?

Anything can be made unnecessarily complex - that's not a meaningful argument. The point is to get a better approximation than the current "I have nothing better to do - time for a Wonder!" approach, something that works as a general approach for any type of production (other than units). It literally only needs to consider tile output with the building vs. without - that means it will sometimes overestimate the value because it won't be able to work every tile, and sometimes underestimate because border expansion will usually only increase the potential output.

You will simply not get "good" results without making extremely complicated calculations, which have to be done again and again and again for x buildings in y number of cities. You could get "better" results by making the AI check the tiles that it has available, but you would not get "good" results. Production bonuses would still be necessary to make up for suboptimal solutions.

Yes, they probably would, but improved priority-setting means that production bonuses will more often be actual bonuses, not just bootstrapping the AI to a point where an average player would be.

It totally does. Gaming AIs are bad and always do stupid stuff, it's all about hiding it so you don't notice.

Obviously stupid moves are frustrating to watch, and in some cases reduce the challenge (Oh no, my city's at risk! Oh well, the AI lost interest), but ultimately the AI needs to be a challenge for peaceful victories. The way domination works in Civ V and VI means that the AI can literally not win a domination victory (another AI failing - vacillating repeatedly over preferred vicvory types and then choosing ones it has no ability or takes no action to pursue - "X is aiming for Domination Victory" followed by no wars or attacks being especially obvious), so as far as actually winning the game is concerned, there's an upper limit to how far it can get with units however much the combat AI is improved.

 

[Ryika]

If it were that easy Firaxis would have done a better job with Civ VI than they have.

I think there just isn't much effort put into creating a high difficulty cap. It's not relevant for most people, what is relevant are obvious flaws that take them out of the experience, or that allow for cheesy tactics (such as, capturing unescorted settlers).

If they did care, they could have tinkered with the modifiers, but I think they didn't at all after release. That tells you everything you need to know. Firaxis doesn't care about the difficulty cap.

Anything can be made unnecessarily complex - that's not a meaningful argument. The point is to get a better approximation than the current "I have nothing better to do - time for a Wonder!" approach, something that works as a general approach for any type of production (other than units). It literally only needs to consider tile output with the building vs. without - that means it will sometimes overestimate the value because it won't be able to work every tile, and sometimes underestimate because border expansion will usually only increase the potential output.

Which means you'll still end up with wonders that have 1 tile that can actually be worked.

I mean yeah, obviously it will make the AI "better" at evaluating when a wonder is good and when it isn't, but as long as it doesn't make it "good" at evaluating those cases - and it won't be "good", unless you factor in tons of possibilities - you will still have an AI that builds wonders that are of low value to them now and in the future.

What you're basically suggesting is to increase the tiles an AI needs to builds a wonder from 1 to X. Sure, that helps and makes the problem less severe, but it does not "solve" it. You still need a production modifier to equalize the playing field.

 

[orasis]

Lol @ 'Chicken Itza'

'The enemy is on their way my liege! What shall we do?' ... 'It's time to unleash the power of Chicken Itza!'

I believe the AI reasoning is more like...

Has the player character built it?
Can I build it?
Let's build it.

Even though many human players seem to ignore it the AI does care about the score and will do as much as possible to make sure your end score, no matter what victory, is as low as possible.

I think part of the AIs problem is it prioritizes a high score and only seems to care/think about victory types once you surpass its score(s).

 

[splungey]

Besides wonder-building, district/city placement, combat and policy cards.. I've noticed how awful the AI is at governor placement. I've watched Scotland cede their capital to me via loyalty simply because they didn't place a governor in it. They only had 2 cities so they certainly had enough governors to put one there and it would have been enough to stop the city flipping.

What's more, they're constantly moving their governors. Aside from the loyalty boost, they never get chance to make use of any of the other bonuses because they're always in the 5 turn establishing phase.

 

[KandAffar]

Besides wonder-building, district/city placement, combat and policy cards.. I've noticed how awful the AI is at governor placement. I've watched Scotland cede their capital to me via loyalty simply because they didn't place a governor in it. They only had 2 cities so they certainly had enough governors to put one there and it would have been enough to stop the city flipping.

What's more, they're constantly moving their governors. Aside from the loyalty boost, they never get chance to make use of any of the other bonuses because they're always in the 5 turn establishing phase.

Maybe the 5 turn delay should just be eliminated for he AI. Asking the AI to pre-plan things like Magnus chops seems a bit much.

As for putting governor's in to stop loyalty, that might degrade the user experience. What is the point of a new mechanic that has no effect because even the AI can stop it?

 

[splungey]

Maybe the 5 turn delay should just be eliminated for he AI. Asking the AI to pre-plan things like Magnus chops seems a bit much.

As for putting governor's in to stop loyalty, that might degrade the user experience. What is the point of a new mechanic that has no effect because even the AI can stop it?

That's a fair point and I think it raises a bigger question - whether the AI should be designed to spoof human players and compete at a similar level; or if they should be set-dressing, part of the environment. Guards in a stealth game, goombas in Mario, goblins in an MMO. I think the latter makes more sense, but is not how they're currently presented being as they have the equivalent play options as a player.

 

[Bad Wolf]

Maybe the 5 turn delay should just be eliminated for he AI. Asking the AI to pre-plan things like Magnus chops seems a bit much.

I would agree with this. The AI does need a lot of hand-holding...

But yeah they seem to focus mostly on loyalty, and they're actually pretty good at it. I've seen cities on the verge of flipping suddenly stabilize, and I can only assume it's because they put a governor there.

 

[Brian LeClaire]

Well the combat AI is now better than in Civ 5 so that's pretty sweet. (Yeah yeah Civ 5 combat AI isn't great but..)

But of course a big shocking headline creates more views on youtube.

I say that for most of players AI is _good enough_.
Next I think they should concentrate on getting it to build more air power.

Why the dig on youtube views? He's making a pretty valid point. The AI in Civ 6 is and has always been the worst in the series. Yes it still is. They've just now made it where civ's will even build ships, they still don't utilize air power. The AI is NOT better than in Civ 5 at all. In virtually every situation the AI makes stupid decisions, whether it's city placement, unit upgrading, unit production, or attacking cities. It only ever excels because of the bonuses it gets from difficulty. That's not good AI, that's a game mode.

 

[orasis]

Why the dig on youtube views? He's making a pretty valid point. The AI in Civ 6 is and has always been the worst in the series. Yes it still is. They've just now made it where civ's will even build ships, they still don't utilize air power. The AI is NOT better than in Civ 5 at all. In virtually every situation the AI makes stupid decisions, whether it's city placement, unit upgrading, unit production, or attacking cities. It only ever excels because of the bonuses it gets from difficulty. That's not good AI, that's a game mode.

Install Ai+ and the 'real tech tree' mods and your good. Firaxis should just pay those two modders.

 

[Kruos]

Why the dig on youtube views? He's making a pretty valid point. The AI in Civ 6 is and has always been the worst in the series. Yes it still is. They've just now made it where civ's will even build ships, they still don't utilize air power. The AI is NOT better than in Civ 5 at all. In virtually every situation the AI makes stupid decisions, whether it's city placement, unit upgrading, unit production, or attacking cities. It only ever excels because of the bonuses it gets from difficulty. That's not good AI, that's a game mode.

No. You are wrong.

Since R&F the AI IS better than Civ5, absolutely. It is not even debattable. Saying the contrary is lying and promulgating false information. Sorry to break your Civ5 honey moon dude. Play the game.

 

[PhilBowles]

No. You are wrong.

Since R&F the AI IS better than Civ5, absolutely. It is not even debattable. Saying the contrary is lying and promulgating false information. Sorry to break your Civ5 honey moon dude. Play the game.

See my earlier post. I pointed to multiple specific areas where the Civ VI combat AI remains inferior to Civ V's. In what particular respects, other than the two (one fixed) that I mentioned are you suggesting Civ VI's combat AI is superior, given that it remains objectively worse in every one of the areas I mentioned (at a bare minimum)?

As for the peaceful AI, LeClaire also made points about specific failings relative to Civ V - and in the broader sense, the AI remains incapable of achieving victory at the highest levels in the same timeframe as Civ V's, assuming no player interference (in Civ VI a Deity AI will win at about turn 350, at least 50 and perhaps as much as 100 turns later than its Civ V counterpart).

 

[Dustbrother]

Thanks for this video! Good work, I hope you make more, because the AI IS trash. It's really frustrating because i love every game design decision made in Civ6. It's like having the best board game in the world, but only being able to play against a 6 year old.

 

[splungey]

@Dustbrother maybe see it as a solo board game in which the AI are components rather than opponents

 

[ChocolateShake]

There's something I'm curious about regarding Vox Populi. Did anyone actually truly face difficulties fighting the AI? Or was it just a matter of making more units than you normally would to deal with them?

I think the Civ 6 AI is good at providing a challenge for non-warmongering play styles. In my games I rarely conquer other civs or even take cities from them (I usually only do so in pure warmongering games or when they really make me feel the need for revenge), and keeping up with them through peaceful means, while also roleplaying, is tough.