AI cheats and strategies to counter them

[Willem]

Originally posted by Sir Ralph


I checked it and it worked like it was described. Btw, still playing the Zulus Aeson?

But I think it doesn't prove that the AI knows, that there are resources, but only, that there must be something valuable. Probably, the AI has a function "EvaluateCitySite()" or so, and this function (not the AI itself) has the ability of seeing the resources and giving higher results with resources than without. But the AI actually doesn't know, what it is exactly.

I suspect you're right about that one. It has to be more than just the presence of the resource, other wise it would try to monopolize every resource on the map. And once it has secured a supply, the value of any subsequent supplies will go down. Which would explain why it's quite often ignoring what to me are ideal resource sites.

[bsklepzig]

Originally posted by Sir Ralph
But I think it doesn't prove that the AI knows, that there are resources, but only, that there must be something valuable. Probably, the AI has a function "EvaluateCitySite()" or so, and this function (not the AI itself) has the ability of seeing the resources and giving higher results with resources than without. But the AI actually doesn't know, what it is exactly.

I don't want to seem flip, but it feels like there is this impression that the Firaxis team has somehow managed to create true Artificial Intelligence. The "AI" for Rome or the Aztecs or anyone else isn't the HAL 9000 or Data from Trek. It isn't a discreet entity, where one part of it's computer brain knows something and the other doesn't. I doubt that even Deep Blue (the IBM hardware/software combo that beat a human Grand Master at chess (Kasparov?)) had discreet subroutines that shielded it's knowledge from the knowledge of the programmers. (And that effort (I'd wager) had significantly greater resources dedicated to perfecting the AI. For starters, it ran on a Super Computer or Mainframe, I think.)

The AI IS the collection of functions programmed by Firaxis. It doesn't even make sense to speak in terms of what the AI "knows." That's a very human perspective. The series of calculations that include calculating the value of a tile based on the presence of aluminum (for example) IS the AI. Nothing more.

I REALLY doubt there is a discreet AI personality for the Zulus. I'd bet the farm it's a series of IF ... THEN ... statements (or some equivalent, depending on the native code environment).

IF ZULU
THEN ... (a bunch of variables change)

When the program evaluates land for settlement on "Zulu's" turn, a calculation happens.

IF TILE ... (has aluminum)
THEN ... (a variable changes to indicate the tile has good resources)

Other variables considered probably include proximity to water, food resources, terrain, etc. All things a human would consider. The difference is the program knows all resources from the first turn. Not really a huge problem for the human, since we have so many adaptive advantages over the program. We settle where resources will LIKELY appear (desert for aluminum and mountains for uranium), then go to war or culture bomb if our guesses were wrong.

Just don't give too much credit to the "thought process" or "decision-making" or personality of the AI. It's just a bunch of variables in a series of calculations.

(And before anyone chimes in with "that's all a human brain is," humans are exponentially, infinitely more complex, with tangential, inductive and deductive reasoning that still baffles biologists and philosophers. The best AI can't begin to touch what a human's thought process is, and -- with all due deference to a fine team of programmers that developed a terrific game -- Firaxis is not on the cutting edge of AI research.)

[Aeson]

It's possible that the AI calculates the value of unseen resources when determining the worth of an already founded city, but not in founding the city, or making terrain improvements. What the AI "knows" is defined by the programmer in any given function. If the EvaluateCity function adjusts for unseen resources, then the AI knows that they are there at that point in the program. If the EvaluateCitySite function doesn't use the value of unseen resources, then the AI doesn't know they are there when finding a city site. It really doesn't add much complexity to the program to make a tech level inquiry when determining the value of any given tile.

if(Tile->Iron)
{
if(Civ->Ironworking)
{
TileValue += IronValue;
}
}

I'm not very experienced at programming, but even I could figure out how to include this in one area of the program, but not another. Not to say this is how they did it, just that it could be done.

humans are exponentially, infinitely more complex, with tangential, inductive and deductive reasoning that still baffles biologists and philosophers. The best AI can't begin to touch what a human's thought process is

Can't help but 'chime in' here:

It is just impossible for any "program" to understand itself completely, humans included. To achieve understanding, there is required change in the data or fuction of a program. What you get is a program that may understand what it was, but not what it has become in that process. We are the best AI that we know of, even if(SelfImage > Reality){makes that hard to accept}.

[WFHermans]

Having read this subject I conclude:

The horrible cheating advantage of the AI is then that it knows where everything - units, resources, land - is. And its Galleys can cross the ocean. And it can trade during our turn. And it can get Invebtions for free.

A way the AI uses this advantage of all-knowing everything is by going for an undefended city, even with a very weak attacking unit.

But I think it's only reasonable to ask Firaxis to remove this cheat if the humans will stop using their cheats like intelligence and creativity. And not to to mention reloading if something goes wrong! (Although I was tempted when I lost a 3 Horsemen Army to a Spearman .......)

And by the way, it also seems that the Barbarians go more after the human than after the AI. Not that I mind. Can anyone confirm this?

[solo]

Looks to me that barbarians give equal attention to the AI, but I have yet to see them defeat the AI in battles witnessed, including one where one of those stacks of 20+ horsemen went up against one Zulu spearman and lost! However, after taking into consideration the fact that the spearman was fortified and was located on a mountain, I concluded, "Of course! It always happens like this with such a good defensive position!"

[Aeson]

Originally posted by Sir Ralph
I checked it and it worked like it was described. Btw, still playing the Zulus Aeson?

I've been playing a very long game with the Zulu. I decided to play a Huge/Pangaea/Emporer with 8 AI to submit to the CivFanatics high score board. I did take a break to play through the GOTM for the past couple of days (English/Archipelaego/5AI) though.

I'm starting to think that the Zulu Impi/Horsemen combination is as powerful as the Iroquois Mounted Warrior. The Impi/Horsemen make for 2.2.2 stacks instead of 3.1.2 MW stacks, plus the Military trait is better than Religious for a pure pop rush conquest. I don't win as many of the fights, but I almost never have a defensive death. Impies are perfect for denying the AI resources as well. I was able to keep all but the 2 AI on the other landmass without Iron or Horses until I was ready to attack. I didn't have to use the Scout trick either. Plus I don't have to worry about running into the Zulu myself. Expansionist is the common trait, and by far the most important on a map like this. I got every Ancient Era tech, about 1000 gold, and a settler (usually would get 2+) from goody huts. It's 200 AD, and 4 more turns till I can upgrade to Knights. I only have 8000 gold though, so only 100 of my Horsemen will get the initial upgrade. I wanted to save my Elites anyways.

[WFHermans]

Another fun thing is when the 20+ stack of Barbarians appears next to your just built 1 small town....and after they cleansed it out almost ALL your money is gone...mmm I must have had the Royal Treasure in that little village.
Again not that I mind.

I never saw the Barbarians take an AI city but I suppose that is coincidence.

Oh and I just thought of another way humans cheat: they don't take their alliances as seriously as the AI.

[pchang]

I'm not worthy

I bow to the incredible patience and fortitude that is Aeson. 475 units?!?!?!?! I would never have the patience to deal with moving that many units. How long does it take to make each turn? How long do you have to wait between turns? My god! the humanity!

[Aeson]

AI Cheat:

Can "see" weakly defended cities anywhere on the map.

Player Counter:

Use purposely weak city garrisons to draw AI offensives away from strategically important areas. In some cases start a game of AI "ping pong".


AI Cheat:

Can "see" undefended bombardment units, workers, settlers.

Player Counter:

Leave these units (preferably bombardment units) out in the open, on high movement terrain. AI offensive units will be left out in the open for counter attack, and usually the captured unit will be reclaimed the next turn.


AI Cheat:

Can "see" resources that aren't visible for their tech level.

Player Counter:

Use the AI as your eyes, finding later resources by their willingness to part with cities.


AI Cheat:

Trading during the Humans turn.

Player Counter:

Not really anything you can do. The AI will give away techs for free rather than let the player make more trades. A very good reason for bloodthirsty tactics to begin/resume. Sadly the AI will never learn from it's mistakes, even when they know that my 100 Horsemen/Knights/Calvary/Tanks/Modern Armor are going to demand satisfaction for the AI's underhanded ways.


AI Cheat:

Free Starting units depending on difficulty level.

Player Counter:

Use the fact that the AI has 5-10 extra warriors/spearmen to support early on to your advantage. Wait a bit, and then hit them with Horsemen or Swordsmen. They won't have as many of these up to date units because their army is still comprised of many of their obsolete starting units. If this doesn't help, try a lower difficulty level.


AI Cheat:

Production bonus depending on difficulty level.

Player Counter:

Let the AI build cities and wonders for you. They will build them more efficiently than you could. Put the saved production into military that will be used for wonder aquisition. After the wonder is in hand, just keep on rolling with your military advantage over the AI. A nice side benefit is that a leader (free wonder) could be generated in the process. If this doesn't help, try a lower difficulty level.


AI "Cheat": (pre-patch)

Moving across ocean with galleys (with the Lighthouse).

Player Counter:

Build the Lighthouse yourself, the AI will be stuck to ending it's turn on coastal tiles while you can go anywhere with your galleys. Also realize that once Astronomy is researched, the AI will be able to move onto sea tiles. With Magnetism, Galleys can go anywhere.

'Dirty' Lighthouse Use:

The AI will only cross sea tiles with galleys (pre-Astronomy) if they can end their turn on a coastal tile, making 2 sea tiles their max crossing. They also will not attack any ship in a sea tile. Use this to make invincible sea Galley blockades until Astronomy.

[Aeson]

Re: I'm not worthy

Originally posted by pchang
I bow to the incredible patience and fortitude that is Aeson. 475 units?!?!?!?! I would never have the patience to deal with moving that many units. How long does it take to make each turn? How long do you have to wait between turns? My god! the humanity!

Haha, I usually play 1 or 2 turns and then switch to another game. It takes about 20 minutes to get through one of my turns, the sad thing is it is only going to get worse! I produce on average 20 units per turn, and don't lose half that many, even on the worst turns. Luckily half my army is fortified in the northwestern portion of the map. They are waiting for the Galley fleet to be of sufficient numbers for a crossing. Also waiting for the Knight upgrade. The worst part of it is the 100's of captured Workers I have running around that aren't shown on the screen there.

I have named all my pop rush cities (everything outside the main northern circle of cities) to 1nationality# for easy recognition in the domestic advisors city list. I just switch to listing by population, and pop rush any of the cities that have reached size 2. I'm up to 1french15, 1indian17, 1babylonian12, 1persian19, 1chinese2 (been razing most of their cities because of cultural concerns). Also I keep steady at about 3 cities being settler disbanded for relocation each turn. Gotta keep the city pattern right! At least my 1.33GHz/512MB machine processes the AI turns in about 20 seconds, but it's early.

[Arrian]

Aeson, I read that military advisor screen and scanned the map about three times before my brain accepted it.

I'm at a loss. I mean, I understood that poprushing could be pretty powerful... but 155 Horsemen & even more Impi... at that point in the game... it's INSANE.

Let me ask you this - and I'm totally serious, no insult meant - is it fun? A quick look at the screenshot is all one needs to understand that it's effective , but is it fun?

-Arrian

[WFHermans]

You deserve to get the Nobel Prize for empire building mr Aeson!
What next, a 16 player monstergame?

And thx very much for your thorough analysis of the "AI cheats".
i especially love

AI Cheat:

Can "see" undefended bombardment units, workers, settlers.

Player Counter:

Leave these units (preferably bombardment units) out in the open, on high movement terrain. AI offensive units will be left out in the open for counter attack, and usually the captured unit will be reclaimed the next turn.

Use them for bait....use the opponents' strength against them. right out of Sun Tzu.

[Aeson]

I think it is quite fun. I look at is as playing several games at once, I have 2 "builder" games going (one around my palace, one just starting around the forbidden palace), and 2 military conquest games going all at once! I just take it slowly, as there is a lot of "fun" that could become tedious playing too many turns in a row It is nice that I can get a great leader just about any turn that I don't already have one. I just used one for Sistene's, and hopefully I can get one for Leo's. After that, none of the wonders should be in question.

I am dreading the period after the conquest is taken care of though, as I'll just have terrain improvements going on forever during the buildup.. not to mention figuring out how many tiles I can claim safely without triggering domination. That's part of why I'm sticking to a very strict spacing pattern, as I can more quickly count if each city claims a certain number of tiles. Eyeballing it, I would say that I can safely claim about 4/5ths of the main continent, with a couple colony type cities on the other for resource purposes.

The real power of pop rushing is that almost any corrupt city can still churn out 1 unit every 5 turns (with the pyramids) or faster. I was able to expand faster than 2 of the AI civs who started out near jungle areas, and so was able to demand cities from them very early on. This made me more powerful than a couple of the other civs, who I could then demand cities from. My first 10 or so "conquests" came this way, and most of my Impies were pop rushed from those cities, leaving my core to build peacefully. Coastal cities were able to rush Horsemen after a Harbor and Barracks. I also set up a few temp cities in really high food areas. 5 of these cities were producing a unit every other turn. Every 20 turns I was able to demand more cities, and captured others in the meantime. It just kind of snowballs out of control.

[WFHermans]

I am certain it is a stupid question, but what is "pop rushing" units? And it is something the AI can't or doesn't use?

[Aeson]

Pop rushing is when you get out your whip in despotism or communism, and trade population points for increased production. Each population point is worth up to 39 sheilds. This makes it very powerful in building up ancient era armies quickly. And it's a production method that corruption doesn't affect at all.