Combat Cheats by the AI

[arretium]

TheNiceOne wrote:

"The AI doesn't cheat in combat. Simple as that. There have been some accusations, but those has been proven to be false:"

To further discussion of this topic, I'm creating a new post. Here's my response:

(Let me first preference this comment with the following consideration: All of my games are played at Emperor level or higher, which may be skewing my own results as well)

I disagree. I've got countless save games showing a definite skew in the advantage of AIs. For example, on my most recent turn in a game, one chinese modern armor took out 3 of my modern armors at full strength. One Aztec Modern armor took out two of my modern armors. Meanwhile, it took two modern armors to take out the Chinese modern armor. It also took three modern armors to take out one chinese mech infranty. Meanwhile, one of the chinese modern armors took out two of my own mech. infrantries.

You'd likely respond to my evidence provided with the following counter response: "That may be true, however, if you consider the battles/wars in the long run, you will find that the AI does not have an advantage." Again, I disagree. In the current example, I provided, the turn mentioned above was followed by another turn in which one aztec modern armor took out my modern armor that was defending a mountainous terrain. A chinese modern armor took out two modern armors that were supposed to be receiving a radar tower bonus. In reality these examples following at least 10 turns of war of the same thing happening over and over again. Their units taking out more of my units per number of offensive units used.

My experience with AI cheats shows that the combat advantage varies according to two principles: the stronger the civilization of the AI perhaps relative to your own strength and whether or not I have gone significantly to war with that particuliar civilization in the past. It may also be dependent on the relative strength of your own civilization.

My experience has lead to a few general conclusions:
The advantage by the AI starts out strong, but as you wear down their civilization, their advantage diminishes.

If you have not had much war with them recently, they start out with a stronger advantage.

There has been much debate about combat cheating by the AI. I've concluded that that the game does cheat in the AI's favor, especially when factoring in elements such as terrian type, city size, use of city bonuses (civil defense in PTW/walls, radar towers in PTW), and unit attacking and unit defending.

TheNiceOne's conclusion about it evening out in the end just does not logically follow when I consider the fact that cheating bias occurs turn after turn and it is only when I finally have broken the back of the opponent civ that the other civ's advantages appear to have diminshed. For these reasons, I respectfully disagree with TheNiceOne, and other colleagues who hold that combat cheating does not occur.

 

[stevenx]

well... i did see more of my units died in diety level then in monarch level... so maybe u r spoton on this one... at least i got another thing to curse other than bad luck

 

[Toast]

I don't know about combat cheats, but the AI definitely cheats a little. I'll see if I still have the savegame, but I recently played a game where, in my attempt to cripple my Russian adversary before suing for peace, I opened the round by surrounding a worker with paratroopers while my tanks took his nearby cities. Much to my surprise, his worker managed to get a way, despite the fact that his only avenue of escape had no roads, and he couldn't move more than a single square. I had a spy steal his plans, so I know for a fact he didn't just disband his worker.

Also, I notice that AI galleys don't sink, even when spending four turns crossing an ocean (I sank his escort ironclad, and shadowed him all the way back to his continent).

 

[Strider]

The AI does not cheat in combat PERIOD

There may be many reason's for this:

1) General tile defense's: Your units may be defending themselves on flood plains... which are awful for defense... where has in most cases I might have to use to modern armor to take out just one on a grassland tile. Most desert/floodplains I can take out with only one unit (and leave it a a pretty good shape). The AI knows the terrian... and prefers the better defense terrian over the others. Whereas human players tend to forget about that fact and move to a less defendable terrian.

2) Rivers: I have big problems with not noticing the little blue lines called rivers. On several occasions I had whole fleets of units wiped out before I even noticed that I was attacking accross a river. Most attack/defense units of the same level are close (defense is close to attack power), but the +25% bonus to defending across a river turns the tide more than some people may believe.

3) Improvements: The Civil Defense/ Wall improvements can be a life saver, or a huge pain. A city with Civil Defense (+50% I believe?) has a huge advantage. If a mech Infantry is fortified in a mertropolis with a civil defense(grassland tile). It's attacked by an modern armor.

Mech Infantry Defense: 18*.50= 9+18 = 27 *.50 = 13.5 +27 = 40 *.10 = 4 +40 = 44 *.15 = 6.6 +44 = 50.6

50.6 defense
compared to
24 attack

That means that on average it will take atleast 3 modern armor to take down one mech infantry guarding that city.

Edit: Added metro bonus

 

[Strider]

Originally posted by Toast
I don't know about combat cheats, but the AI definitely cheats a little. I'll see if I still have the savegame, but I recently played a game where, in my attempt to cripple my Russian adversary before suing for peace, I opened the round by surrounding a worker with paratroopers while my tanks took his nearby cities. Much to my surprise, his worker managed to get a way, despite the fact that his only avenue of escape had no roads, and he couldn't move more than a single square. I had a spy steal his plans, so I know for a fact he didn't just disband his worker.

Also, I notice that AI galleys don't sink, even when spending four turns crossing an ocean (I sank his escort ironclad, and shadowed him all the way back to his continent).

My last game I saw a persian galley sink in an ocean square. So they CAN and WILL sink.

 

[arretium]

Strider wrote:

----------------------------------------------------------------------
"The AI does not cheat in combat PERIOD

There may be many reason's for this:

1) General tile defense's: Your units may be defending themselves on flood plains... which are awful for defense... where has in most cases I might have to use to modern armor to take out just one on a grassland tile. Most desert/floodplains I can take out with only one unit (and leave it a a pretty good shape). The AI knows the terrian... and prefers the better defense terrian over the others. Whereas human players tend to forget about that fact and move to a less defendable terrian.

2) Rivers: I have big problems with not noticing the little blue lines called rivers. On several occasions I had whole fleets of units wiped out before I even noticed that I was attacking accross a river. Most attack/defense units of the same level are close (defense is close to attack power), but the +25% bonus to defending across a river turns the tide more than some people may believe.

3) Improvements: The Civil Defense/ Wall improvements can be a life saver, or a huge pain. A city with Civil Defense (+50% I believe?) has a huge advantage. If a mech Infantry is fortified in a mertropolis with a civil defense(grassland tile). It's attacked by an modern armor.

Mech Infantry Defense: 18*.50= 9+18 = 27 *.10 =2.7 +27 =29.7 +.15 =4.5 +29.7 = 30.2

30.2
compared to
24 (modern armor attack) "
-----------------------------------------------------------------------

I disagree. All of the battles discussed above in my examples occured on desert or grassland terrain OUTSIDE of a city, unless otherwise noted. I have already factored in the terrian, river, and city improvement considerations.

Desert and grassland terrain provides a 10% bonus.

Your river analysis is irrelevant. I am well aware of the 25% bonus, but others might like to know about it. . In any respect, there are no rivers in any of the terrain discussed above.

Your civil defense analysis is irrelevant to open combat considerations. Only the "radar tower" could be a factor but then the AI would have to build the radar tower on my own territory and then remove it before my next turn. Such a tactic would also be cheating, since the AI would have to move their own worker to my terrain thru some sort of "right of passage way activiation" without my consent, build the radar tower, and consequently destroy the radar tower before my own turn. I'd argue such a move by the computer would likewise also be a combat cheat. I suspect we'd agree that the AI's radar tower is not a factor in this discussion.

City bonuses depend on city size. A wall is equivalent in defense to a size 7-12 city. 13 and above receive a 100% bonus based on size of the city.

Your math is incorrect above. You discuss civil defense and then provide an example using mech. infrantry. Specifically, your mathematical computation regarding "29.7 +.15 " does not make any sense. Please provide an explanation for your different considerations.

I stand by my conclusion reached above. I've played this game for many years, in fact, more than a decade when you factor in all of the different civilization versions.

 

[Bamspeedy]

Most people remember more often the times when they got 'cheated' in combat, but don't remember the times they were victorious with a unit that shouldn't have won.

Most humans would never throw a warrior against a pikeman, so they don't know how often the human warrior would win, but they sure do notice it if the AI warrior beats their pikeman.

You have bad streaks, and good streaks. On the higher levels, the AI has a more respectable counterattack, simply because they can produce more units (and higher-quality units), so it will seem like you are losing more of your battles.

The only 'cheat' is that they know if your city has walls/civil defense, whereas the human would have to investigate the city to know that.

You ever hear of a combat calculator?
I'm in the process of downloading one to give you some stats, % chance of units winning vs. other units, just give me a few minutes.

 

[Bamspeedy]

Originally posted by Toast

Also, I notice that AI galleys don't sink, even when spending four turns crossing an ocean (I sank his escort ironclad, and shadowed him all the way back to his continent).

If the AI has ironclads, they have steampower, and then obviously have navigation or magnetism, which allows *all* ships to safely travel ocean tiles.

 

[Bamspeedy]

arretium- You aren't mentioning very often how many hitpoints your units have, their experience level (regular, veteran, elite), and terrain, all of which is very important in determining the odds.

All of the below responses are assuming all units are veterans, and fully healed.

one chinese modern armor took out 3 of my modern armors at full strength.

If the chinese MA was on offense, a Veteran MA has a 66.4% chance of winning (on normal terrain). Or 2 out of 3 battles. Does it seem too unlikely he could win 3 out of 3?

One Aztec Modern armor took out two of my modern armors.

Same as above. Who is on offense, what terrain is the defender on?

Meanwhile, it took two modern armors to take out the Chinese modern armor.

Need more info. What terrain was the Chinese MA on? (assuming the Chinese MA is on the defensive).
If he is on plains, you have a 66.4% chance of winning, but if he is on mountains, you only have a 34.7% chance of winning.

It also took three modern armors to take out one chinese mech infranty.

Veteran MA attacking a Veteran MI in a metropolis built on a hill has only a 19.6% chance of winning. If the metropolis has the civil defense or radar towers, your chances will be even lower.

Meanwhile, one of the chinese modern armors took out two of my own mech. infrantries.

If we take the previous example, and change it to either a walled town, or a city, and built on plains/grassland, the MA goes up to a 40.2% chance of winning.

the turn mentioned above was followed by another turn in which one aztec modern armor took out my modern armor that was defending a mountainous terrain.

34.7% chance of attacker winning. Certainly not too outrageous in a 1 vs. 1 battle.

A chinese modern armor took out two modern armors that were supposed to be receiving a radar tower bonus.

So what terrain where you on? If plains and a radar tower bonus (50%?), the attacker will win 50% of the time. Flip a coin and see how many times you get heads twice in a row.

 

[Speaker]

I get really sick of these kinds of posts. The AI absolutely does not cheat in combat. You don't complain when your 2-Attack Horseman takes down a 2-Defense Spearman, when odds are that should not happen. If your entire fleet of Modern Armors were defeated by that same Spearman, then maybe you would have cause to grumble, but when a 16-Defense Modern Armor takes out a 24-Attack Modern Armor, really cry me a river.

 

[arretium]

Bamspeedy -

All of the above battles, unless otherwise noted, occurred in the following manner:

The `person' who "took out" the other was the OFFENSIVE unit and was discussed as the SUBJECT of the sentence.

The terrain, as mentioned previously, was desert or grassland. Defensively they only provide a 10% bonus.

All of the units were at full strength at the point of the initial attack. In other words, the chinese MA that took out three of my own MAs was at 4/4 when it first started attacking. It was at 3/4 when it finished. All 3 of my MAs were at full strength. I find it highly improbable that it would lose only 1 hit point.

Radar towers (at least for the chinese) are inapplicable. Some of my units were supposed to benefit from the radar tower, but for purposes of our current discussion I'm leaving them out. At any rate, they would further strengthen my argument. Civil defense, walls, and city sizes are inapplicable.

 

[arretium]

Speaker wrote:

"I get really sick of these kinds of posts. The AI absolutely does not cheat in combat. You don't complain when your 2-Attack Horseman takes down a 2-Defense Spearman, when odds are that should not happen. If your entire fleet of Modern Armors were defeated by that same Spearman, then maybe you would have cause to grumble, but when a 16-Defense Modern Armor takes out a 24-Attack Modern Armor, really cry me a river."

Perhaps there is some truth to our claims if you are getting sick of them? I'm not complaining about a 16 defense modern armor getting taken out by a 24 attack modern armor, nor am I complaining about one 24 attack modern armor taking out three of mine (MAs). Nor am I complaining that it takes me 2-3 MAs to take out one of theirs (MA). Instead, if you read the post carefully rather than skimming the title and then posting your rant would you have discovered that I am basing my conclusion, perhaps only a hypothesis at this point, on past practices and experiences.

Here's a few more:

1. Territory - mine, grassland
3/4 china MA beats my 4/4 MA, remaining hit points of victor: 2/4


2. Territory - mine, grassland
My 4/4 MA beats China 2/4 MA, remaining hit points of victor: 3/4

3. Territory - mine, desert
2/4 China MA beats my 3/4 MA, remaining hit points of victor: 1/4

4. Territory - mine, desert
MY 4/4 MA beats China 1/4 MA, remaining hit points of victor: 1/4

In the following four examples, I was the AGRESSOR (offensive) and the chinese were the DEFENDERS. I'm not even batting 66.4% here, am I?

 

[wilbill]

Originally posted by arretium
In other words, the chinese MA that took out three of my own MAs was at 4/4 when it first attacking. It was at 3/4 when it finished. All 3 of my MAs were at full strength. I find it highly improbable that it would lose only 1 hit point.

Maybe it got promoted during the course of 3 winning combats and ended up 3/5?
I really find this subject tiresome. The conclusion you seem to be drawing, arretium, is that the game's designers programmed it so that the AI has not only an advantage in combat, but one that works on a sliding scale depending on relative strength of your civ vs. the AI, how long you've been at war, and so on. But, just to get a good laugh at us, they wrote the code so that if you use the editor to set up a test and run a couple of thousand combats - somthing that's been done to death - you get a result almost perfectly matching the expected probability.
Well, maybe in a couple of years the people at Firaxis will tell us that's what they did and get a good laugh at our expense. But you'll have to excuse me if I find it a little farfetched.

 

[arretium]

Willbill -

Thank you for your response. You are free to disagree. You may even call my theory far fetched.

You mentioned that if "you use the editor to set up a test and run a couple of thousand combats - somthing that's been done to death .... expected probability." If you are alluding to you having ran some of those tests, then I'm sure you'd agree that if an armor was promoted, it would have ended with 4/5 or 3/5 hit points when it finished based on the following rationale. First, it started with 4/4. Two, when an unit gets promoted it receives an additional hit point in its currently available column and in its maximum allowed column. Three, a unit can be promoted no further than elite status. Four, once promoted to elite status, the unit does not continue to receive additional hit points in its "currently available column" until rested. Five, a unit can receive no more hit points in its "currently available column" than in its maximum available column. However, the example that you discussed in your reply occured without promotion.

 

[Bamspeedy]

In the following four examples, I was the AGRESSOR (offensive) and the chinese were the DEFENDERS. I'm not even batting 66.4% here, am I?

50% is close enough when you use such a small sampling (4 battles). When you attack with a 3/4 MA, that is like attacking with a regular MA, not a veteran. You are engaging in piece-meal attacks like the AI do to each other. Use stacks, so every single battle is with a fully healed unit. Don't use injured units, even if they have the possibility of making another attack that turn. I only do that when it is an obvious outcome like MA vs. spear.

 

[Hygro]

time to crawl out of my little hole:

I have found I cheat in combat. I win battles all the time I shouldn't. The closer the odds are to 40% that I'll win, the closer chance I have to 100%

No really though, just make sure to play your important turns on days with good astrology

also arretium, theres a button called "quote" on the topright corner of a post which you can press to quote people. It will save you time. Welcome to the boards!

 

[Hurricane]

@ arretium: no offence meant, but there have been dozens of people before you coming stumbling into this forum with the same accusations as you, and all of them have been shown wrong. I suggest you sit down and study the combat calculator and make sure you understand it, and then start to write down the results of each and every battle you fight. And this is important: DO NOT write down only the improbable ones. After a while you will see that most of the times the battles end as expected, some of the time you are lucky, and some of the time you are unlucky. That is called randomness. It's as simple as that.

 

[RegentMan]

Originally posted by arretium
I've got countless save games showing a definite skew in the advantage of AIs.

Looks like we've got another person who's met the Random Number Generator! I've seen many posts about "the AI cheating." It's just not true. Do you complain when your knight beats a musketman without losing health? No. Do you complain when an AI knight does the same? Yes. I always get ticked when an AI beats the odds or my performance. For example, I bombed an AI infantry down to 1/4. Then I used my own infantry (veteren) to attack. It wins, but is now down to 1/4. The AI counterattacks with a 4/4 infantry, and wins on the first round! I didn't get angry. It's just another one of those sigh moments. Whenever something goes my way, I just say, "Whoa!"

 

[Bamspeedy]

Here is a test save, where you can test all the MA vs. MA battles you want (295 battles).

http://www.civfanatics.net/uploads4/Civ_III_PTW_Autosave_4000_BC1.SAV

Must have PTW. (It's in the Debug Mode).

Have each of your MA attack the enemy MA right next to it. Fortify it if the unit wins, so you can go right to the new unit. Using a unit that isn't fully healed and/or is promoted, would skew the results.

Deity level. Entirely desert. All MA are regulars.
I only modded the game to give both players extra gold (for all the unit support), and I got rid of the deity starting units.

My first test on offense, my MA won 194 battles (65.76%).
Combat calculator says I should win 64.2%. I guess I beat the odds!

I could have a much higher kill ratio if I use the ones that got uninjured and/or promoted, but that isn't the test here. (and this is what the AI will do, during their turn if you let them-but they also use injured units).

 

[Mark Young]

Originally posted by RegentMan

For example, I bombed an AI infantry down to 1/4. Then I used my own infantry (veteren) to attack. It wins, but is now down to 1/4. The AI counterattacks with a 4/4 infantry, and wins on the first round!

I've had the same thing happen to me on numerous occassions and it normally ends with me calling the screen all types of names

But then just as you said, I will win a battle, usually a last ditch attempt to take a city, with a multi turn unit that is down to it's last 1 or 2 hitpoints and out of sheer frustration attack a unit that should be able to beat it and win the battle.
The random number generator can be your friend or your enemy or both

It's a game of swings and merrygo rounds.