I'm convinced the AI has a maphack.

[Blake]

If you know me, you know I like to test stuff.

For ages I've known the barb pillagers know about improvements that the AI state has never seen. But I could accept the barbs having "local map knowledge" or whatever.

But I can DEFINITELY manipulate the behaivours of AI marauders by changing garrison counts at various cities, they WILL go for the weakest point, and if there is no weak point, they might leave.

I'm pretty sure I'm right about this, I did a lot of reloads and puppeteered this AI chariot, and it consitently went after targets it hadn't seen.

It did however seem to be somewhat ignorant of areas that truly lay under fog, as in it didn't seem to want to pillage improvements in areas still under black fog, however this didn't hold true of cities, it could still sniff out weakly defended cities even if they were under black fog.

Now this is pretty subtle map hacking, it's difficult to detect without repeated reloads, for example if it seeks out a horse resorce in your territory and pillages it, you could assume that the AI had presumed you have made a pasture on the horse resource, since it's smart play to do so. And it seems that the AI sends units on "missions" and changing parameters during the "mission" wont effect the units behaivour, that is if it's bored of life and decides to attack a weakly defended city, it'll still attack that city even if you defend it to the gills, but if you reload to before the decision was made and garrison the city, it's mission will change to attack a weaker target.

I'm not going to point at the AI and scream CHEATER!, I suspect the fault lies in subroutines, like the AI asks for an improvement that it can pillage, and the game tells it where one is, or it asks the pathfinder how to route somewhere, and it's given a correct route, but the pathfinder happens to see more of the map than the AI does, so it's like the AI is a friend with benefits with the game engine, it doesn't exactly know the entire map, but it can get information that is inaccessible to the player (the player doesn't need to ask the game engine where a pillagable improvement it, players have eyes for that).

And overall, it is subtle enough to be handwaved away as "Good deduction" rather than "Omniscient", since AI's aren't known for their deductive skills I can live with partial omniscience (is that an oxymoron?). But still, I think the truth is important, the AI does have map knowledge that the player doesn't.

[LordShiva]

For the longest time I was like "What on earth is [maffack]??"

[CrapFactory]

Almost all games like this in any way has cheating AI.

RTSs and all of 'em! We're not good enough at making true AI so we let them cheat a bit. In WarCraft III they have a limited Knowlege of the whole map (take a huge army and move it up and down on the map and the enemy will do that too to match you quite often, you can use the reveal map cheat to see this). In Rise of Rome which is a very economy game the computer doesn't have bad things happen if they are losing 100000gold a turn.

Cheating is a norm for AI. I try not to think about it because it really kinda sucks.

[Kuciwalker]

I believe someone else also uncovered this, for naval resources IIRC. Soren said it must be a bug, and that the AI has no knowledge of things under fog IIRC.

[Enigma_Nova]

In Warcraft III, the Reveal Map code reveals it for the AI as well.
If you download a 3rd party maphack and install it, you will see the AI ignorantly creeping, as opposed to matching your forces.
Oh, this is off-topic.

[Blake]

I really don't see how it is offtopic, unless you meant your post, in which case I agree.

[Cort Haus]

Yeah, E_N likes his gratuitous +1s.

Good research again, Blake . When Sir Ralph documented the naval 'maphack' (and has never been seen since ) the word was that this was unintended and limited to sea-units not land. This data suggests otherwise.

[Arrian]

I think it was unintended that it be so obvious, both because it would annoy players and because it would be easier to exploit. But, IIRC, Soren never said the AI couldn't see units under the fog... some of the Beta testers might have mistakenly spread that.

In short, yeah, the AI knows where your units are...

-Arrian

[Blake]

I've noticed that only a few AI units seem to cheat, these being the marauder / lone ranger types.

A good example of this is a multiplayer game I was playing. My capital city was in the core of my empire, this being late game and I'd done a lot of conquering, it was supremely safe.

Now, my capital city was defended by a single warrior, it's something I do because it amuses me, EVERY other city in my empire had better defenses, and most others MUCH better defenses.

Anyway, the AI didn't try to send a giant stack in against my capital. No, instead it sent a single knight in. I didn't actually notice this knight, with so many reports of troublemaking pillagers flooding in, and it successfully defeated the warrior and took my capital. (my response: "... not again" having lost my capital 2 or 3 times previously in the game, to units sneaking in - you'd think I'd learn and garrison it properly), actually I only speculate it snuck in, it could have teleported in for all I know. It's possible to pretend that the AI sent the unit in, on the offchance I'd left my capital undefended, but I know it actually cheats, for sure I've never seen a single unit sneak in only to find heavily defended cities - they only come in when there is weak point.

But the odd thing is, it seems that only the troublemaker units have this maphack, the AI's main invasion forces don't seem to.

So what I think is that the AI does indeed see the whole map, it just pretends it doesn't, but there are a few cases where it gets caught out - so it's sort of a bug, but also as if the AI is a player who has a maphack, but tries to disguise it and make it just look like good play/intution.

[Badtz Maru]

Are you sure the AI didn't have some other way to know? Remember that you can see what's going on in cities that have a religion that you founded - if your attacker founded Hinduism and you have Hinduism in your capital, they know exactly what units you have there. You can also suggest to an ally that they attack a specific city, so if the country invading you was friends with someone who had the holy city for one of your capital's religions, they might have told them to go for your capital - then they would send a knight to scout it out, and then find that it is virtually undefended.

I've never got the feel that the AI was cheating in this game yet, but there are a lot of legit resources for gathering intel.

[Spaced Cowboy]

Like spies for instance.

[Kuciwalker]

Originally posted by Blake
So what I think is that the AI does indeed see the whole map, it just pretends it doesn't, but there are a few cases where it gets caught out - so it's sort of a bug, but also as if the AI is a player who has a maphack, but tries to disguise it and make it just look like good play/intution.

That's how it always works - I mean, the map is in memory, so the AI could access it, but it's programmed to only consider things it can see. Then the bug comes in when it accidentally considers things it can't see.

[Diadem]

The AI definitely scouts out terrain a lot. I'm not convinced they are actually cheating.

They might be sharing maps with other AIs a lot, though?

[I AM I AM I AM MOBIUS]

If you know me, you know I like to test stuff.

For ages I've known the barb pillagers know about improvements that the AI state has never seen. But I could accept the barbs having "local map knowledge" or whatever.

But I can DEFINITELY manipulate the behaivours of AI marauders by changing garrison counts at various cities, they WILL go for the weakest point, and if there is no weak point, they might leave.

I'm pretty sure I'm right about this, I did a lot of reloads and puppeteered this AI chariot, and it consitently went after targets it hadn't seen.

It did however seem to be somewhat ignorant of areas that truly lay under fog, as in it didn't seem to want to pillage improvements in areas still under black fog, however this didn't hold true of cities, it could still sniff out weakly defended cities even if they were under black fog.

Now this is pretty subtle map hacking, it's difficult to detect without repeated reloads, for example if it seeks out a horse resorce in your territory and pillages it, you could assume that the AI had presumed you have made a pasture on the horse resource, since it's smart play to do so. And it seems that the AI sends units on "missions" and changing parameters during the "mission" wont effect the units behaivour, that is if it's bored of life and decides to attack a weakly defended city, it'll still attack that city even if you defend it to the gills, but if you reload to before the decision was made and garrison the city, it's mission will change to attack a weaker target.

I'm not going to point at the AI and scream CHEATER!, I suspect the fault lies in subroutines, like the AI asks for an improvement that it can pillage, and the game tells it where one is, or it asks the pathfinder how to route somewhere, and it's given a correct route, but the pathfinder happens to see more of the map than the AI does, so it's like the AI is a friend with benefits with the game engine, it doesn't exactly know the entire map, but it can get information that is inaccessible to the player (the player doesn't need to ask the game engine where a pillagable improvement it, players have eyes for that).

And overall, it is subtle enough to be handwaved away as "Good deduction" rather than "Omniscient", since AI's aren't known for their deductive skills I can live with partial omniscience (is that an oxymoron?). But still, I think the truth is important, the AI does have map knowledge that the player doesn't.

Have you tried editing the garrison counts using a hex editor like cheat engine and then observing behavior? You might be able to isolate where in the 'module sets' the emerging behavior is present.