 |
|
Dominae
|
|
I'd help out alexman, but AU202 put me way back in my schoolwork. I'm eagerly awaiting the results though (my AU202 game could have benefitted from a Killer AI...).
Edit: I just thought of a simple test. Put 2 equally aggressive civs on a very small island and play with the OCN (some tests at around 5, some at around 50). The differene between the overall AI behaviours should become apparent...
Dominae
Last edited by Dominae on 10-12-2002 at 05:21
|
|
| |
|
|
|
Tharsonius
|
|
alexman: how did you test it? Is there a way to set up a game with only AI-players?
|
|
| |
|
|
|
Dominae
|
|
alexman, here's an improvement on your methodology: put a heavy object (a small paperweight will do) on the Enter key. Although this may not improve results, it could actually allow for more tests to be run (due to lower boredom factor). Hope this helps.
Dominae
|
|
| |
|
| |
|
| |
|
|
|
Theseus
|
 |
The warmonger formerly known as rpodos. Gathering Storm! Apolyton Spirit!!
Apr 2002
time: 00:24
|
|
Great discussion guys!
(And thanks for chiming in Soren, but you need to be a little less opaque  )
My first efforts at generating killer AI civs, based solely on game settings, were ultimately predicated on the theory that differing starting conditions would create 'rich' and 'poor' AI civs, and that there would be a self-reinforcing loop where the rich get richer and the poor get poorer.
Two comments, given all that's been learned since then:
1) I think the changes to AI civ behavior to date, including both ours and Soren's, will contribute to this taking place. Better mining, for instance, will increase the performance gap between an AI civ gifted with shielded grassland and one stuck with jungle.
2) But that doesn't get the AI civs to exploit relative advantage... or at least we don;t know what does. The whole impetus for me originally was the frustration at discovering a new continent equally divided by however many AI civs, when one of them clearly could have pulled an Arrian Deception (  ).
Hmmm...
What are the drivers for AI civ aggression / expansion:
* Resources (I almost always play 3B under the theory that clumping will somehow help)
* Aggression rating
* Military strength
* OCN and room (don;t forget this interplays with map settings too)
Soren at the time suggested giving the AI civs room, and thus time, to more fully develop in order to become killers. I'm guessing that he thinks increasing OCN in the same situation will just result in increased REXing.
With all due respect, Soren, I think you are missing what we are trying to accomplish. It's not that we are looking for more *successful* civs... I think our mutual improvements will already do that, especially for those blessed with good starting environments.
It's OK with me if an AI civ REXes more, in fact I'm sure it's better for the ones with good starting environments... I just want to know that when the rich meet the poor, the rich kick some f-cking ass. To me, that says:
AU Mod and PTW 1.14
125% OCN (maybe even 150%)
60-70% water
1 civ less than max
3 billion
Wet
|
|
| |
|
|
|
Catt
|
 |
California - SF Bay Area
May 2002
time: 21:24
|
|
quote:
Originally posted by Theseus
Regarding testing, could someone play with this idea:
* OCC for the human
* Squarish main continent with 3-4 AI civs
* Design the map to give each civ wildly different starting conditions, including terrain and resources (i.e., rich and poor)
* Variables to test would be # of useable land tiles per civ and OCN.
Set beer can on Enter.
There's got to be a balancing point between the 'success' of the AI civs and the rich taking advantage of their relative strength. |
Well . . . [looking down one's nose], we won't be setting our beer can on Enter. We might be persuaded to occasionally tap enter with the fine rounded end of our ale bottle or pint. [/looking down one's nose]
My thoughts are running the other way -- with a random start, or with a more or less equal start, I'd like to get a sense on whether increased OCN does actually produce a greater proclivity towards "killer-ness" or at least towards more aggressive exploitation of advantage. The self-reinforcing loop should naturally follow (although use of "should" in a complex environment is risky in and of itself). Part of the reason I am disinclined to test with starting positions whose characteristics are clearly disimilar is that I think you were on to something with the 3B world maps -- clumping of terrain features necessarily means that starting positions offer different inherent values -- and that the advantageous start is more likely to produce a Killer AI than other tweaks we've played with. But map features themselves haven't seemed to produce the "killer instinct" needed to exploit relative advantage.
My hope is that a test from random conditions (with some degree of natural variation, of course) together with an increased OCN will produce a proclivity towards "killer instinct." If true, then map variations, combined with the "killer instinct" mod, should produce a Killer AI through the strengthening of the self-reinforcing advantage of a favorable starting position.
Now I will open another ale and proceed to press Enter repeatedly (and then play some AU 202).
Catt 
EDIT: Completely forgot to make clear my main concern  . Given Soren's concern about our OCN idea, I would like to isolate whether it actually produces more aggression or not. If yes, then I think it adds to the likelihood of creating Killer AIs (building on natural advantages). If not, then we need not go further, nor be confused by a "non-Killer AI" simply using its inherent advantages to look "killer" in a random (not necessarily repeatable) fashion.
Last edited by Catt on 10-12-2002 at 08:20
|
|
| |
|
|
|
alexman
|
 |
Hunt Valley, MD
Jan 1970
time: 00:24
|
|
Jaybe, for some reason no wait at the end of the turn doesn't work. Maybe because if it did, since I was producing Wealth, I wouldn't be able to do anything to stop the game until it ended!
I don't believe PTW 1.14 changed the AI's desire to irrigate in general. The irrigation is partly a result of the AI's inability to use the luxury slider and needing more food to support entertainers, so it wouldn't be to its advantage to stop irrigating anyway. What 1.14f fixed was that the AI no longer unnecessarily irrigates grassland tiles in Despotism. This improves its production big time in the Ancient age.
I ran another test last night (beer or no beer on the enter key, it's still boring, and takes way longer than 20 minutes when the AI are at war BTW), and the higher OCN didn't seem to increase aggressiveness this time.
It seems that the AI becomes more aggressive (because it wants to use all the units being built instead of settlers) once it runs out of room to expand, or when it reaches the OCN, whichever comes first.
If the map is such that the OCN is reached when there is still room to expand (which is rarely the case when you play with the maximum number of civs), the AI will actually become more aggressive by reducing the OCN, as Soren suggested. If room runs out before the OCN is reached (which is usually the case when you play with maximum civs), increasing the OCN looks like it doesn't affect the AI's aggression.
If this is indeed the case, I still think that the AI would benefit from increasing the OCN because a) the game has less corruption than when Soren first established that the AI should stop expanding at the OCN, and b) the AI rarely expands enough to win by domination.
However, under Theseus' settings for a Killer AI (one less civ than max) the increased OCN would make the AI spend more time REXing instead of building units and infrastructure to attack its neighbors. Is that what we want?
|
|
| |
|
|
|
Soren Johnson Firaxis
|
|
Hunt Valley, MD
May 2001
time: 00:24
|
|
quote:
Originally posted by Dominae
Why?
Perhaps because the AI remains in its peaceful "REX" phase (or something like it) until the OCN is reached. Once its empire is fully "fleshed out", it enters "production" or "military" phase, and looks for reasons to pick fights. Thus, the OCN and AI aggressiveness may be inversely related (at least indirectly). Just a guess, though.
Dominae |
more or less, this is true. Once the land grab phase is over, the OCN's biggest influence on the AI is whether they decide to raze or capture cities (which is also influenced by other factors...)
|
|
| |
|
|
|
BRC
|
|
quote:
more or less, this is true. Once the land grab phase is over, the OCN's biggest influence on the AI is whether they decide to raze or capture cities (which is also influenced by other factors...) |
A question for Soren: If an AI is trimmed back to half the OCN later on in the game, and there are still open city spots, will the AI revert to it's REX or will it continue in it's "production" phase?
If someone else knows the answer, just throw it down. Don't need to waste Soren's time.
|
|
| |
|
|
|
Dominae
|
|
quote:
Originally posted by BRC
A question for Soren: If an AI is trimmed back to half the OCN later on in the game, and there are still open city spots, will the AI revert to it's REX or will it continue in it's "production" phase?
|
The "REX" and "production" phases are just guesses, by the way (the AI always seems to be looking for new land to colonize, although maybe this has something to do with excess Settlers). To answer your question, an AI civ that gets knocked down to a few cities will expand again if land is available (or else Domination would be a lot easier!). I have no idea if this has anything to do with the various phases.
Dominae
|
|
| |
|
|
|
BRC
|
|
quote:
To answer your question, an AI civ that gets knocked down to a few cities will expand again if land is available (or else Domination would be a lot easier!). |
Most of the time, the AI overexpands at the beginning of the game, and is ripe to be hit. This is why archer rushes are so effective. I understand when you say they will expand again, but at what cost. Is it a full blown REX, comparable to their Ancient Age expansion, or is it a more controlled settling of the land? Basically, do they spread their defenses too thin and make it easy to overrun them again??
|
|
| |
|
|
|
Dominae
|
|
quote:
Originally posted by BRC
Most of the time, the AI overexpands at the beginning of the game, and is ripe to be hit. This is why archer rushes are so effective.
|
Actually, I think the AI handles early expansion quite well: it escorts its Settlers, produces defenders in each new city, and does so at a reasonably fast pace.
Archers rushes are effective because us humans know the AI's habits (and the AI doesn't know ours!). Rushes are also effective against humans, by the way. Striking a fine balance between expansion and defense against a possible rush is not easy. This is primarily why I think Expansionist is amazing.
quote:
Originally posted by BRC
I understand when you say they will expand again, but at what cost. Is it a full blown REX, comparable to their Ancient Age expansion, or is it a more controlled settling of the land? Basically, do they spread their defenses too thin and make it easy to overrun them again?? |
This an interesting question, but purely academic. If a civ is knocked down to half its cities, "spreading its defenses too thin" is irrelevant, because it's going to lose in the long run. Even if the AI did "regroup" and solidify its defenses, it would just prolong the inevitable. The point of the 'Killer AI' effort is to teach the AI to "go the distance" and wipe out such defeated civs.
Dominae
|
|
| |
|
|
|
BRC
|
|
quote:
Actually, I think the AI handles early expansion quite well: it escorts its Settlers, produces defenders in each new city, and does so at a reasonably fast pace. |
I agree. From reading Vel's earlier threads, they were amazed at the speed that the AI expands. I shouldn't have said "overexpanding", but I feel that this is one of the best times to hit the AI.
quote:
This an interesting question, but purely academic. If a civ is knocked down to half its cities, "spreading its defenses too thin" is irrelevant, because it's going to lose in the long run. Even if the AI did "regroup" and solidify its defenses, it would just prolong the inevitable. The point of the 'Killer AI' effort is to teach the AI to "go the distance" and wipe out such defeated civs. |
Here is where they tie in.
Let us say that on another continent, the Iroqouis are mauling the Aztecs. The Aztecs lose many cities. Do they start building Settlers instead of Military Units? If the answer is yes, then the Iroqouis should continue to run them over.
Now from the other side.
The Iroqouis are punishing the Aztecs. They have not yet reached OCN, and so are still building settlers pretty fast. Wouldn't it be more advantageous for them to keep producing military units so that the attack doesn't bog down?
Depending on how it works, the Iroqouis could end up as a "Killer AI".
What I want to know is if the Aztecs prevent the Iroqouis from becoming a Killer by locking down and fighting a slow war of attrition? This would not be a good thing.
|
|
| |
|
|
|
Dominae
|
|
quote:
Originally posted by BRC
What I want to know is if the Aztecs prevent the Iroqouis from becoming a Killer by locking down and fighting a slow war of attrition? This would not be a good thing. |
I think I understand your point. My comments:
1. The AI is rarely so focused as to build only Settlers (or a majority of Settlers). The sole exception is at the beginning of the game, where the AI does appear to have a definite REX phase. I think what you're asking is whether it can be knocked back into this phase during war. I would guess: usually not. If enemy units are threatening their cities, AI civs will tend to build defenders (in my experience).
2. The major problem in AI behaviour that I see with the "Iroquois mauling the Aztecs" scenario is that the Iroquois will be too open to a peace treaty. Unless the Aztecs have almost nothing to give, the Iroquois will stop their onslaught at a moment's notice. The difference between humans and AIs is therefore not "Am I close to the OCN?" but "Am I almost done defeating this civ?". Unfortunately, we cannot change this particular AI behaviour with the editor.
3. Confusing the AI is possible in some situations (thinking it's directly at war, believe now is a good time to build a Wonder), so maybe it is possible to trick it into entering a new REX phase. Again, I'm not sure.
Dominae
|
|
| |
|
|
|
BRC
|
|
quote:
I think what you're asking is whether it can be knocked back into this phase during war. |
Yes, this is what I'm asking.
quote:
I would guess: usually not. If enemy units are threatening their cities, AI civs will tend to build defenders (in my experience). |
Ok. I just didn't know.
quote:
2. The major problem in AI behaviour that I see with the "Iroquois mauling the Aztecs" scenario is that the Iroquois will be too open to a peace treaty. Unless the Aztecs have almost nothing to give, the Iroquois will stop their onslaught at a moment's notice. The difference between humans and AIs is therefore not "Am I close to the OCN?" but "Am I almost done defeating this civ?". Unfortunately, we cannot change this particular AI behaviour with the editor. |
Yeah.... The Iroqouis will stop. Is this where aggression factor could kick in? Does aggression affect how likely the AI accepts a peace treaty?
|
|
| |
|
|
|
Catt
|
|
California - SF Bay Area
May 2002
time: 21:24
|
|
Too much beer, not enough civ.
quote:
Originally posted by alexman
I ran another test last night (beer or no beer on the enter key, it's still boring, and takes way longer than 20 minutes when the AI are at war BTW), and the higher OCN didn't seem to increase aggressiveness this time. |
Ain't that the truth. I am up to about 1700 AD in my "standard" game. One AI has been at war for over 1000 years. The other continent has been at war off-and-on for 700+ out of the last 1000 years. Much more than 20 minutes invested in this one (although I have found it much more interesting than alexman becuase this was the first time I had run a "debug mode" game and it's interesting to watch the AI development habits).
In my test game (only one sample) Egypt systematically destroyed 2 of its 3 continent mates and was well on the way to owning an entire (large) continent. It then began razing cities instead of keeping them, and wasn't settling the exposed land.
With Soren re-visiting to explain his concerns a bit more (and to inadvertantly provide an explanation as to why my Egypt razed and razed), I'm not really sure that continued "increased OCN" testing is worthwhile, at least in connection with a desire to induce more aggressive exploitation of relative advantage. What do you all think?
quote:
If this is indeed the case, I still think that the AI would benefit from increasing the OCN because a) the game has less corruption than when Soren first established that the AI should stop expanding at the OCN, . . . |
The current mod increases it by 50%, right? (For AI FP placement reasons, not corruption iteration reasons, of course). Let's hear some reports from AU 202.
quote:
. . . and b) the AI rarely expands enough to win by domination. |
With Soren's comment re: razing, I wonder if the AI will ever win via domination. Maybe another reason to increase OCN (aggression will still trigger with lack of elbow room)?
Catt
|
|
| |
|
| |
|
|
|
Fosse
|
|
U.S.A.
Mar 2002
time: 23:24
|
|
I agree that OCN should go up for the AI just to stop the razing habits that Catt's Egypt got into.
I played a standard PTW game in which Japan razed a gigantic subcontinent of Mongolian cities, and only kept cities with access to resources. While the resource keeping was smart, the giant whole in the map meant that all of the other civs put all kinds of distance corrupted cities on former MOngolia, and Japan's strength didn't go up nearly enough to relfect winning such a definitive victory.
If Japan (or Egypt) had kept all of the cities then they would have been in a much stronger position, even if many of them were corrupted.
Would it be to the AI's advantage, without being too much of a human helper, to reduce corruption levels with the slider?
That way AI's (and human) distant cities that they keep under increased OCN percentage would be even more profitable for them, assuming they build courthouses and police stations in them.
Since humans are better at dealing with corruption than AIs are, (better city placement, chopping forests for production, disbanding military units, etc) making corruption less crippling would, I think, help the AI out more than it would us.
|
|
| |
|
| |
|
|
|
Dominae
|
|
I'm confused: how often does the AI raze cities if it's already reached or surpassed the OCN? In my games, I've never seen a big AI leave land unclaimed. Rather, it always gobbles everything up (much like the human player). In 'Son of SVC', the Germans were nearing a Domination victory by the time I launched the Spaceship. Have things changed with PTW?
Dominae
|
|
| |
|
|
|
alexman
|
|
Hunt Valley, MD
Jan 1970
time: 00:24
|
|
Dominae, perhaps your Germans had very high culture compared to their victims in that game? I'm sure another factor that determines whether or not the AI razes a city is the probability of a culture flip of that city. Another is, as Fosse mentioned, access to resources or a Wonder.
As an extreme experiment, reduce the OCN to 1 and watch the AI never build a settler. I did it yesterday. Definitely not domination threat from such an AI.
Fosse, the AI's corruption can be changed without affecting the human, because in this mod the AI plays at a difficulty level that is not meant for the human. I say this is not the ideal way to help the AI, because the idea is to make it react better to game situations, not to make it cheat more.
Last edited by alexman on 11-12-2002 at 02:56
|
|
| |
|
|
|
Catt
|
|
California - SF Bay Area
May 2002
time: 21:24
|
|
quote:
Originally posted by Dominae
I'm confused: how often does the AI raze cities if it's already reached or surpassed the OCN? In my games, I've never seen a big AI leave land unclaimed. Rather, it always gobbles everything up (much like the human player). In 'Son of SVC', the Germans were nearing a Domination victory by the time I launched the Spaceship. Have things changed with PTW?
|
I'm only making inferences (i.e., wild-ass guesses) based on Soren's short post from above, the "debug mode" game I'm watching, and memories of other games -- but I think that OCN must be a reasonably important factor in an AI's decision to capture or raze (I would guess that flip chances, "terrain value" and the like also are included in the calculations).
I have never seen the AI win by domination. To be honest, I've never heard of the AI winning by domination.
In the "debug game" (again, admittedly only one sample), Egypt has taken a large portion of the largest continent. It is also a cultural powerhouse, having built a bunch of wonders and enjoying a strong infrastructure. In wiping out a couple of smaller civs, it seemingly refused to keep certain cities -- the terrain was decent (though no, or very few, resources), and the enemy culture was not a danger. Not only was it razing, but it wasn't building settlers to settle the land -- but it was building workers and other non-military units / improvements.
I have seen games where, if it gets down to an AI civ and the human, unless the human settles razed lands, they will remain open. Always assumed it was the AI's attitude towards OCN at work, but never tied that attitude to a decision to raze or capture. I briefly went back to "Son of SVC" -- I don't think your Bismarck was very close to domination. Compare your winning minimap to Dave McW's upon his domination win.
It might be worth the twenty minutes to play a human-AI "duel" on a small or standrad map, lots of land, with the human on a one-tile island, in debug mode. No possible war means only 20 minutes of debug mode (). The way would be clear for an easy AI domination win if the AI settled enough of it's large continent (on a small map, or even a standard map, would the AI continue to settle, or stop short of Domination?)
Catt
EDIT: Doh! Cross-posted with alexman. And his OCN=1 test much easier and simpler than my duel.  ("give me a santa hat"  )
|
|
| |
|
|
|
Dominae
|
|
Catt, you're right in saying that Bismarck was not going to win via Domination in my Son of SVC game right away. Probably he needed another 30-50 turns or so. But from the time I started building the SS to its launch, he almost doubled his land mass through conquest alone. He didn't raze many cities, as far as I could see. Maybe he didn't win earlier because he was in fact razing cities...I wasn't looking too closely at the time. But there's probably a good explanation for this (relative culture, etc.).
This is how (I think) the AI should behave: if it's above the OCN and has low relative culture, it should raze and rebuild; if it's under the OCN, it should always conquer; and if it's over the OCN and has high relative culture, it should always conquer. The second scenario isn't exactly optimal, but big civs are usually the stronger ones, so I don't think it happens too often. Now, the question is: does the AI do this?
Dominae
|
|
| |
|
| |
All times are GMT. The time now is 05:24.
Apolyton Time is 00:24. |
top of page
|
|
|
Forum Rules:
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
|
HTML code is ON
vB code is ON
Smilies are ON
[IMG] code is ON
|
|
|
|
|
|
10-12-2002 05:14
Plese don't hurt me! 
|