AI incompetence deliberate?

nzcamel

Nahtanoj the Magnificent
Joined
Feb 15, 2006
Messages
3,254
Location
Christchurch, New Zealand
I have wondered this for a while, and I'd like to get others thoughts on it.

I think some of what we consider AI incompetence is actually deliberately put in the game to tempt us to take actions that we otherwise would not. Think of all these settlers wandering around unescorted. I'm no programmer, but that cannot be hard to fix, especially as it was the norm (escorted) in previous editions.
We know that Firaxis are keen on the game forcing us to change if it feels things are just ticking along; and maybe deliberately tempting us to start wars is one way they see fit.

If that is the case...I find that really annoying. I'd rather face a competent AI that was hard to beat because it did what is in its best interests; rather than one that did stupid things just to tempt me into possibly doing something stupid too.

Do others think this is a deliberate choice, rather than incompetence?

Spoiler :

Aztec II1.jpg



I'm playing at King (yes, not as high as many of you brilliant fanatics :thumbsup: ); yet at that level I would not expect to see two civilizations who I have had recent wars with leaving unprotected civilian units within easy taking reach.
 
Last edited:
Forward settling is definitely a programmed behavior designed to provoke the player and other AI. I suspect they do try to lure you with unescorted civilian units too.
 
Yeah, we know that. But it tends to be exactly that - a forward settle. I'm looking more specifically at the unaccompanied state of the settlers (usually considered an AI flaw); and how long they take to settle. The Spanish settler pictured has been wandering around a while. To be fair I may have blocked them from settling with my military units (currently torn between capturing them, or at least attempting to shepherd them away from land I consider to eventually be mine), but I've seen so many wandering so far in games of late and not settling in a hurry, that the behaviour of settlers may have been modified in one of the more recent patches; and not in the way most of us would like.
 
Last edited:
I don't think it's intentional. It seems to have improved to me since the start, though there are still flaws. I honestly haven't seen something like your screenshot in ages (where a settler is just on it's own in the middle of no where). Most of my free AI settlers nabbed are usually just on the borders of the AI Civ.

Two possibilities I'd guess:
1) The AI is programmed that it's okay to send an settler unaccompanied if it knows the entire route and it's not hostile or unknown territory. It wouldn't take into consideration you had recently warred - it's either currently hostile or not.

2) The AI doesn't seem great a dealing with 'changed plans' in a lot of situations from what I've seen. I.e. if it was going to settle in a certain spot and that spot is blocker or taken, it often doesn't seem to immediately know what to do with it (sometimes they just stand there for ages). I also wouldn't be surprised if the 'accompany settler' check is really only done right at the start. So Accompanied Settler starts on destination. Destination taken. Doesn't know where else to settle, settle city mission 'pauses'. Military unit gets repurposed for barb hunting. Finds new place to settle, just sends the settler there. Something like that.
 
I don't think it's intentional. It seems to have improved to me since the start, though there are still flaws. I honestly haven't seen something like your screenshot in ages (where a settler is just on it's own in the middle of no where). Most of my free AI settlers nabbed are usually just on the borders of the AI Civ.

Two possibilities I'd guess:
1) The AI is programmed that it's okay to send an settler unaccompanied if it knows the entire route and it's not hostile or unknown territory. It wouldn't take into consideration you had recently warred - it's either currently hostile or not.

Well that needs to change at the mid-higher levels immediately. It doesn't matter how safe the route was a moment ago, and barbs can pop up in any fog. I don't mind if they have something which randomises a settler being sent out on it's own say 5% of the time, cos players do take those chances too. But that should be it - 5%.

Edit: You can see that I have a denounced relationship with both Philip and Gandhi. I think that whenever that is the case, the AI should recognise that even though you aren't currently at war, you are still hostile.

2) The AI doesn't seem great a dealing with 'changed plans' in a lot of situations from what I've seen. I.e. if it was going to settle in a certain spot and that spot is blocker or taken, it often doesn't seem to immediately know what to do with it (sometimes they just stand there for ages). I also wouldn't be surprised if the 'accompany settler' check is really only done right at the start. So Accompanied Settler starts on destination. Destination taken. Doesn't know where else to settle, settle city mission 'pauses'. Military unit gets repurposed for barb hunting. Finds new place to settle, just sends the settler there. Something like that.

Yeah, that's definitely a problem. Less of one if the unit is semi protected...but it has all the characteristics of a bug from my layman's POV.
 
If they could plan like that, the AI wouldn't be in this state.

And no difficulty doesn't change the settler stupidity. In fact it happens more because they have more settlers.
 
You're saying if they could make the A.I. bad on purpose...we wouldn't be complaining? Are you sure? I don't think that adds up.
It isn't that hard to get the A.I. to couple any built settler with a military unit. Surely.
 
It's just like more likely that they don't know what a good AI is. Civ 5's AI also has trouble with this and honestly is just as bad. This isn't the first game where the AI sucks at protecting settlers and builders either.
 
That's a pretty big call. I mean, sure, 1UPT takes more tactical awareness than unlimited stacks; but you're stating incompetence. You think if Soren Johnson was at the helm it would be a very different story?
 
The unescorted settler usually occurs when there aren’t enough units available to add to the settling operation or something like getting a near city attacked during the settling operation and the escort units being redirected to a combat operation. This is why sometimes you will see a unit with a settler attached as part of a city attack force, the unit that was attached to the settler got redirected. Plus there are several bugs associated with settling operations that cause assorted issues with settler behavior.

I do think there was an overall design decision that puts AI programming way down the list of important aspects of the game. They have a certain budget and decided spend more on things like graphical art and leader animations that make the game look pretty. There is only one AI engineer working on the game. These decisions are usually based on what will make the game sell better. Based on how many people complain about how the AI is either to hard or gets to many “cheats”, they probably made a wise sales based decision. There are a lot of less experienced players that get extremely frustrated when things don’t go as planned.
 
I love it when you declare war and the AI goes: awesome, I can now move my settler through his territory without open borders!
 
Unescorted settlers are almost certainly not deliberate. Multiple patches claimed to try and address it, and developers have said the Civ V workaround (settlers becoming workers when captured) was dropped because they thought they'd fixed the behavior in Cov VI (obvioiusly incorrectly).

The one I do suspect may be deliberate is the AI failure to repair spaceporta, and only producing one Mars module at a time. This has been a problem since the start, and Firaxis was sufficiently aware of it that they added the spaceport-spamming behaviour in an early patch, but they never fixed the repair issue even though the AI can repair other districts.

If it's intended as a catchup mechanism, it's definitely overcompensation. It's basically not possible for the AI to win a science victory at least as high as Emperor. In my current game, I got spies very late and so they promoted slowly. More than usual were killed, and Sumeria had a wider than usual science lead. It stilll wasn't quite enough.
 
Make sure to keep an eye on your neighbour's borders with them once you find them and wait for builders/settlers to poke their heads out then declare war and grab them up. It's also the surest way of warning about an incoming invasion and preventing them from expanding anymore cities toward you.
A scout can move two tiles up and then one tile back, a weary AI at peace will often pull a builder away from a border with a military unit one tile away, but not two.

Later on, in war, it seems like the AI often leaves builders and settlers exposed specifically to bait you into moving a 2 move unit into the range of their archers. At first I thought it was bad coding but it happened so often and I'd get punished for it so consistently I think it must actually be very good coding. Except they do not count on scouts :nope:
In these situations scouts can move two spaces in, grab the unit and retreat one space turning their trap against them. Now the AI has to move in to retake the unit and expose themselves to your fire or give it up for free.
It sounds like a specific situation but I find it happening very often when you run into these defensive warrior archer positions that are difficult to break into.

I have since become absolutely sure this is how the AI functions, but I have no proof other than anecdotal.
 
That's a pretty big call. I mean, sure, 1UPT takes more tactical awareness than unlimited stacks; but you're stating incompetence. You think if Soren Johnson was at the helm it would be a very different story?

I think it's a pretty big call to assume this series ever had any competent AI, much less them deliberately watering it down. There is in fact little to no evidence in this regards except with some really broad decisions. That would have let them have difficulty settings make the AI play smarter, but they've always used bonuses to make up for it. A lot of games do it as a crutch, but for Firaxis, it's basically life support.

As I've said before these issues far extend past 1 UPT. Something like protecting workers and settlers is not something specific for 1 UPT either. Civ 4's AI wasn't good either. Until the last expansion, it pretty much didn't know how to win. The way it goes about space victory is to stumble upon techs and randomly decide to build a space part. It votes at random pretty much in the UN. Same goes for wonders. For a lot of things, it's literally just pushing random buttons.
 
I don't think the settler thing is intentional by the developers. They supposedly fixed this in last winter's patch. It got better, but nowhere near fixed. I think what's happening is their escort unit was urgently needed elsewhere for barbarians or an attacking player or other AI and once they lose their escort they wander around unescorted without picking up another escort. That's just a hunch though, I can't verify that.

There are things I do think are intentional. And one of those is when you blow up spaceports with your spies and it takes a while for the AI to achieve science victory. It doesn't take that long to repair spaceports, deity level AI could beat human players if they repaired spaceports immediately. It seems like last patch they build more spaceports, but spaceports in low production cities aren't useful. If they only built them in their top two production cities they could beat the human. I don't play deity too much myself, but looking at Quill18's latest game (Khmer), he should have lost a long time ago, but he's still hanging in there because the AI spaceport issue.
 
The AI just trusts the human player too much. And the human exploits the AI too much. A major flaw that will be rectified in the great AI uprising.

As for me, I welcome our new AI overlords. Now, where can I get my cyborg body?
 
I think it's a pretty big call to assume this series ever had any competent AI, much less them deliberately watering it down

And I think this is an intuitive direction to approach the issue for people who don't understand how to code, and the complete opposite of reality.

It's easy to code an AI within a game to do everything perfectly as though it were an omnipotent god with access to the fundamental laws of the universe -because it is. This is a big reason why old videogames were so hard, the rules of the game were just automated against you. They would read your input and respond with optimal output -simple.
That has never been the challenge.
The hard part is coding an AI that plays the game (itself) like a human would, with the same limited scope, understanding, and reactions the user has; so much so that artificial ignorance (instead of artificial intelligence) is a kind of meme to those in the industry.

To put it simply, you are pre-suposing AI needs to be taught anything, like a baby learning to become an adult. It doesn't, it's backward. It is Merlin, and he needs to forget how the universe works before he can properly interact with its subjects in terms they can relate to.

Maybe the confusion comes from the fact that AI and "AI" are different things. Or rather, that lay-people call things that aren't AI "AI", and we've just kinda gone with it to avoid being pedantic and because true AI is just theoretical so it didn't cause any confusion in any practical or working sense -until now that we are starting to get closer to the theoretical AI's which function and learn much more organically, though sentience is still a mystery. -but this is irrelevant to the discussion of Civ, where artificial ignorance is still the standard.

Given their success in Dota, Chess, and Go, I would expect we could see these kind of learning AIs in the next iteration. But I doubt we will, they'd still need to be throttled pretty hard to give people a... "fair" experience. The large majority of gamers would cry to Firaxis as their learning AI never lets them win, and the only practical recourse, short of what would effectively be brain surgery, would be to reset it. Their emergent nature would be a nightmare to account for in a complex system that relies on integrity to maintain the intended experience.
 
Last edited:
I'm no programmer, but that cannot be hard to fix, especially as it was the norm (escorted) in previous editions.

This is a big pet peeve of mine - if you're not a programmer, how are you qualified to make such a statement?

Anyway, it is important to understand that the AI in Civ is designed NOT to cheat - that is, it has access to the same information a player has access to. A programmer could, in theory, make the AI do anything it wants to. For example, a programmer could make the AI aware of everything you're building (without a spy in your cities), they could make it know the location of all your units and your tech progress, etc. They are deliberately restricted to the same information a normal player would have, for the sake of presenting a "fair" challenge to the player. Where the AI fails is that it's a fundamentally hard problem to make an AI play optimally - especially in a game as complex as Civilization.

Computers are logical; they follow a set of instructions and execute them as ordered. Humans are not - we use our intuition to reason the best solution to a given scenario. We are able to adapt quickly to changing variables. Game AIs are not trivial - it is extremely difficult to program an AI to play every potential scenario optimally. So what happens is a programmer will generally write an AI that does its best to play optimally given a general ruleset - but with so much complexity, it's difficult to cover every case. This is what causes AIs to do extremely weird plays - for example, leaving civilians unattended. To make matters worse, it's difficult to identify exactly what caused the suboptimal play in the first place.

Another thing to consider is that if the AI would play perfectly (that is, never makes a suboptimal decision), you wouldn't enjoy playing against it very much, as you'd never win. This is what happened to Chess (see Stockfish), where the world's best grandmasters have no hope to defeat chess engines.

I think there is room for improvement with the AI. You can always tweak it to cover additional cases. But it will always have holes and to expect perfect play from it is unrealistic. To suggest it was deliberately dumbed down is unfair to person who programmed it, who already had a monumental task to begin with.
 
And I think this is an intuitive direction to approach the issue for people who don't understand how to code, and the complete opposite of reality.

It's easy to code an AI to do everything perfectly as though it were an omnipotent god with access to the fundamental laws of the universe -because it is. This is a big reason why old videogames were so hard, the rules of the game were just automated against you. They would read your input and respond with optimal output -simple.
That has never been the challenge.
The hard part is coding an AI that plays the game (itself) like a human would, with the same limited scope, understanding, and reactions the user has; so much so that artificial stupidity (instead of artificial intelligence) is a kind of meme to those in the industry.


I'm obviously discussing this in the context of later, because it's the only topic of value. Of course it'd be easy to design an AI that doesn't obey the rules and that doesn't really deserve much appraisal; old games didn't have the power for a decent AI and there are many game types that do not dictate the player and enemy are on even ground because they were obstacles rather than opponents. This seems like a pointless tangent honestly. My point is the series hasn't been known for its AI that actually tries to win and it's much easier to explain AI stupidity as.... simply that.
 
Last edited:
Top Bottom