1. We have added a Gift Upgrades feature that allows you to gift an account upgrade to another member, just in time for the holiday season. You can see the gift option when going to the Account Upgrades screen, or on any user profile screen.
    Dismiss Notice

Report Questionable Behavior

Discussion in 'Civ4 - Better AI' started by Iustus, Jan 25, 2007.

  1. Iustus

    Iustus King

    Joined:
    Jul 18, 2006
    Messages:
    609
    Location:
    Sunnyvale, CA
    The purpose of this thread is to report questionable behavior on the part of the AI.

    If you have found a specific bug, that is clearly a bug, then please report it on the bug reports thread.

    On the other hand, if you have seen something which falls more in line with a judgement call, please report it here.

    Did the AI do something which you think may be incorrect? Would you like to see more defenders? Less defenders? Did the AI declare war at a time you thought was inappropriate? Did the AI build a wonder that did not make sense? Vote inappropriately for/against a UN resolution? This is the correct place to report such things.

    Please try to include screenshots or even better, save game files. You want the save before the questionable behavior occurs, if possible.
     
  2. Iustus

    Iustus King

    Joined:
    Jul 18, 2006
    Messages:
    609
    Location:
    Sunnyvale, CA
    example questionable behavior report

    Version: 1/16/07 build

    When automating exploration, the unit went southwest, to a flat forest, rather than going west to a hill, which I think would be better.



    Save game is enclosed, right before the move.

    To duplicate, select the warrior to the east of the city, and click the explore button.
     

    Attached Files:

  3. Iustus

    Iustus King

    Joined:
    Jul 18, 2006
    Messages:
    609
    Location:
    Sunnyvale, CA
    --- reserved ---
     
  4. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    I'm in the middle of my first game with the 1/25 build. And honestly, IMO the unit spamming is still there. Here is a screenshot of a nearby Egyptian city.

    There are 50(!!) units just sitting in it. And they've been there for at least the last 20 turns just sitting. Meanwhile, Egypt is running dog on tech and development....

    There is just nothing good about having the AI have that many units just sitting idle. I dont believe they should be building stacks that size at all, but if you are going to make the AIs do this, then they need to do SOMETHING with them. Otherwise they are just crippling themselves to little effect. And unfortunately if the AIs are going to continue to do this then the only recourse is to change the nature of the game to something akin to 'always war' (since the only way for those insane piles to pay off is to conquer with them).

    I also fought a war with the nearby Celts. And he too had 15+ units per city on the border. And with the right amount of Cannon and a few Grenadiers, I was able to take those cities with very minimal losses. I just dont believe that having piles of units like that really adds defense in proportion to the cost of building and maintaining them (or even remotely close). The collateral damage quickly renders those units ineffective. The only hope would be defending with massed siege, but even that doesnt do much unless it fully up to date (ie, the Egyptians 15 Catapults arent going to do much to a properly put together attack force of Rifles/Grens/Cannon).

    I still firmly believe that the mod is moving in the wrong direction. Massed units do NOT make the AI better at all. Its weaknesses are just as exposed, but the ability to actually tech and move ahead are diminished. The AI is never going to be able to handle proper usage of collateral damage and dispersal to prevent the same on your own troops. So with the AIs having more troops and forcing the player to build more troops its actually skewing things against the AI IMO.

    I would note that with the same customized difficulty settings, the 2.08 AI still seems to perform better as a whole than this build. Yes, I can conquer a neighbor on occasion with little effort, but I can still do it here with ease with the only difference being that all of us are investing more in troops than anything else. But in the end, its not really any 'harder' to take those same cities.

    FWIW, I still believe that the number of units being built by the AI should be cut by at least half again. Having only 2-3 defenders is not enough, but having 10-15 (with the occasional 50+) is just a complete and total waste.

    I definately appreciate the effort for the build, but the results (IMO) are not getting better as the builds go by. I think it 'peaked' somewhere around in December and has been moving away ever since.
     

    Attached Files:

  5. Mexico

    Mexico TR senior programmer

    Joined:
    Dec 2, 2005
    Messages:
    578
    Location:
    Slovakia, Kosice
    totaly agree..i was implemented this into our TotalRealism mod, but from some revision (don't remember #) there is still too big stacks which do nothing, too much (or too few) defenders.... i am sorry, but i start thinking to remove this (or make this as optional DLL) because players noted this AI as very annoying
     
  6. Roland Johansen

    Roland Johansen Deity

    Joined:
    Apr 29, 2003
    Messages:
    4,292
    Location:
    the Netherlands
    It seems as if this AI build an attack stack and when it was finished, it noticed that the potential targets had gotten to strong and aborted the attack.

    How long does it take the AI to build such an attack stack? It seems like a long term project. Maybe it should reconsider the option to build an attack stack periodically. In that way it could decide that the attack stack is not going to be effective enough before it has finished it.
     
  7. Wodan

    Wodan Deity

    Joined:
    Jun 3, 2005
    Messages:
    4,867
    Location:
    In transit
    Agreed on the stacks. Seems like there are 2 decisions the AI needs to make:

    1) Yes, let's do Dagger or Crush or whatever, so start building units instead of growth in terms of infrastructure / settlers / workers. (You can literally see this happening... last game, Ragnar is sitting there with just his capitol while other AIs have 2 or 3 cities. It doesn't take a genius to figure out what he's doing.)

    2) Actually begin the attack.

    It seems that somewhere between #1 and #2 the AI sometimes gets stuck. Maybe it rechecks the power graph, maybe a shared religion has spread so that relations are much better, or maybe it looks over the border and sees a bunch of spearmen or something. In any event, the net result is that the AI is left with crappy infrastructure, fewer cities, and a bunch of obsolete units.

    That said, sometimes the Dagger strike works, and it's wonderful to see.

    Wodan
     
  8. Wodan

    Wodan Deity

    Joined:
    Jun 3, 2005
    Messages:
    4,867
    Location:
    In transit
    One other thing I've noticed, on several builds. Was wondering if it would still happen with the 1/25 build but I'm seeing it this morning too.

    The AI has a ton of units sitting there. Fine, we've been discussing that.

    BUT at the same time, barbs are having a field day, coming in left and right. So, one possibility with all those units is to teach the AI to do a better job at fogbusting. Just a thought.

    Wodan
     
  9. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    I just played out my game for a while longer but gave up in boredom. I'm now over 1000 points ahead of the next leading AI and continuing to pull away. And again, with the exact same difficulty settings I can rarely pull that off against the 2.08 AI.

    It feels like we are back to the days of being able to get an early lead and then just cruise to victory. As long as you keep up your Military Score you are fine. So its just a matter of having a few cities continually pumping the latest troops while you build up some reserve cash for an occasional upgrading of the older guys.

    But the AI is never going to stage any form of comeback like it used to be able to do because it is following the same predictable pattern of trying to keep up militarily even when I have 2.5x the next highest AI's production. It seems to be 'obsessed' with building those troops but regardless of how many it builds, I can build even more and continue to move ahead with my now superior econ.

    FWIW, 'Aggressive AI' is not turned on. There has been quite a bit of scrapping between some of the other AIs and Ragnar has vassalized Rome and was working on Egypt when I stopped playing. But he is hopelessly behind me in tech and his GNP and Productivity are laughable in comparison.

    For the AIs, it just takes too much investment to gain militarily with these builds. Even if it wins a few wars, its still going to be backwards so in the long run those Civs become less and less of a threat. I could smash him out of hand despite having relatively close military levels. And the reason is because I have massed, modern siege weapons which will easily dispatch his ridiculous stacks with little loss. That is another legacy of these massed unit builds...giving even more power to siege weapons.

    Anyways, I hope after 1.0 is released that you can perhaps find time for a 'sub mod' that simply corrects the bugs and problems with the 2.08 AI. I understand that trying to maintain multiple mods would be difficult, but perhaps just a 'patch' for 2.08 to catch the bugs rather than adding in the other routines that I feel arent really making the AI 'Better' anymore.

    Again, thanks for all the effort, but I believe that the project is going farther than the I would have considered as the original scope. Gameplay IS being altered quite a bit simply by the ways the AIs behave. And the game is significantly more military focused than I think is good for the way the game is designed.
     
  10. Wodan

    Wodan Deity

    Joined:
    Jun 3, 2005
    Messages:
    4,867
    Location:
    In transit
    Personally I don't feel it's at the point of throwing the baby out with the bathwater.

    2.08 had/has some huge problems with the basic AI algorithms. So huge, and so fundamental to AI behavior, that any adjustment is without a doubt going to result in massive swings in how games play out. To my thinking, that's a matter of fact and that outcome was to be expected all along. So, I guess it seems to me that pointing to this outcome and using as a justification to backpedal is a bit gratuitous (unreasonable).

    Anyway, without a doubt some huge progress has been made, with worker actions, war management, and more. Yes some more adjustment needs to be done with the huge garrison stack generation, but to me that feels minor. It will have a big effect on the way games play out, but the actual adjustment itself is minor. I'd be disappointed if that adjustment wasn't made before the BetterAI release, but Blake and Iustus will do what they think best.

    I think your feedback is good, Uncle Joe, and I agree with the intent of most of it, just not with the premise of your conclusion. Just my opinion.

    Wodan
     
  11. Arlborn

    Arlborn Legendary Noob

    Joined:
    Jul 29, 2006
    Messages:
    1,939
    I think its a bit too much military yes..It is specially because of the defensive spam, that also cripples completely the AI(what also cripples is an unused dagger, but it was 'OK' before the defensive thing), but lets not forget that Blake is only working in this because YOU guys complained about the low number of defense in cities..

    I think the program definetely is improving. You must be wondering why I say it Uncle Joe? Because while they are trying to fix this spam problem, they are also making other improviments! Look at the logs of the versions! Finally Im using the automatic governament(in small cities because I prefer to look myself in big ones hehe)! Now the AI is even back to pillaging as far as my first game with the new AI goes!
    And if the AI decides to use that dagger stack against you...

    Well, I think I know why its hard to calculate how many defendes is needed of each type and make it balanced! Probably because the AI dont think in long term but turn by turn..
    But Im sure they will find a way.

    I think the problem with the economy is that the AI should build less dagger stacks, only build them when the difference in military between them and the aim will surely not drasticlly change while making the dagger up! AND disband units if they are gonig to sit in a city for 50 turns waiting a mass upgrate without any use while it!! Just build again a little before a military tech comes!
    What about it guys? AI start to build dagger when they start to tech(or going to the path of it) a important military tech? Then upgrate the units that were built before it(while it was being researched) and keep building some more turns and attack!!

    I think that with some smarter defensive tatics(meaning less units) and smarter use of dagger the AI will get better again in management of gold!

    They are trying to 'fix' it and eventually they will. The AI didnt defend good before, probably its not defending good now also, but the objective is make it defend better than vanilla, one way or other.

    '1 steps backward for 2 forward'. Or something like that :p
     
  12. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    But at the end of the day, 2.08 WORKS. It provides a competitive and fun game (given proper handicaps). It doesnt pigeon-hole the player into following one basic militaristic strategy. And the AIs feel far more dynamic. Are they easier to conquer? Perhaps. But that isnt necessarily a BAD thing. As long as there is a cost associated for doing so (in this case, potentially letting other AIs get ahead), then I think its OK.

    But overproduction of military creates cascading effects that ripple through the entire AI and game balance. This has continued through the last three builds with the stated intention of not reverting back to the 'easy to conquer' AIs. I dont believe that 'easy to conquer' was the problem per se.

    I think 2.08 with bug fixes would be a lot closer to the design parameters for the game. Civilization was never meant to be a 'conquest game'. I have dozens of games I can play for that. The current trend is to emphasize military (whether defensive or offensive) over other aspects of the game play in the name of making it 'harder'. But really, it only makes it 'harder' if you intend to win by conquest. If you are trying other types of victory, its not really any harder at all...its just more tedious.

    So, what I'm simply asking for is a quick-fix for the worst of the 2.08 AI bugs without all of the changes to the AI's military routines. I'm willing to accept that it will not play as 'smart' militarily because I believe that that 'smartness' comes with far too high of a price tag in terms of the way the game plays out. I would wager that I am not alone.

    Many of the other improvements are great, but since the underlying foundation is flawed (again, IMO), they arent contributing as much as they could if they were retrofitted to 2.08 base. I would fully understand if its just too much work to do this.

    Its becoming clearer to me that its not going to be possible for one AI project to 'fix' the game for all people. The warmongers want an AI that will match their military build up. The people who didnt try to overcome the AI will military force every game see this enforced military as 'annoying' or distracting or tedious.

    But I dont see many middlegrounds within one AI. Either its going to build enough military to try and oppose human conquests or its not.
     
  13. Wodan

    Wodan Deity

    Joined:
    Jun 3, 2005
    Messages:
    4,867
    Location:
    In transit
    You remember that this is a beta test of BetterAI, right?

    Giving feedback is great, and why we're here. Let's not jump to prejudge the end result, though.

    One of the biggest problems IMO with 2.08 (and prior) is that the AI has a weakness in war management. High level players fully admit that rushing and such are the main/best way to win on Emperor and Deity. Basically you have to exploit the AI weakness at war in order to compensate for the insane production (etc) bonuses the AI gets on those levels.

    So, you point out that BetterAI has a weakness here. That's no different from 2.08 IMO. Plus, I think that BetterAI can (and I hope will be) fixed/tweaked to stop the huge garrison stacks.

    Whereas 2.08 AI is plain stupid. It can't be tweaked... it will always suck at war and the human will always be de-facto encouraged to go to war to beat it.

    Rather, I think that bumping up the AI military is to compensate for its current biggest weakness. The basic idea is that if the AI becomes harder to conquer, then human strategies other than war become more viable. This is in line with your stated goals/objectives for Civilization.

    The latest build decreased the odds that AI will choose Dagger. This is an improvement. Further, the BetterAI includes such improvements as rush production / whipping if invaded. Which gives the AI a suitable alternative to building a massive army to "oppose human conquests".

    Wodan
     
  14. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    I'm going by the recent trends. With this build, my understanding was that the 'unit spamming' was to be decreased. I see no measurable difference. The AI is still swimming with units and forcing the player to do likewise.

    What I'm getting at is that I accept that my perception of a 'Better AI' might be differing from yours or theirs or anyone elses. I think a lot of players enjoy the 'standard' gameplay from Civ and dont feel the need to see the military aspect accentuated.

    That doesnt make sense to me. If the AI is 'improved' to the point where it doesnt need massive bonuses anymore (where I think we were a handful of builds ago), then the need 'rush' the AI is diminished greatly. And if you are 'rushing' and AI while the rest of the newly improved AIs are teching and not building redundant troops, you will quickly fall behind. The massive bonuses were necessary as a result of poor opimization. That is largely gone, obviating the need for those bonuses (and as adjusted in the Better Handicaps).

    So whats left is that perhaps it was too easy to get a material advantage by attacking the AI early. That can be tweaked with a FEW more units. As demonstrated repeatedly, 12 Lonbows does not provide any more realistic defense than 6, but costs twice as much. There are SEVERELY diminishing returns at work here.

    But it doesnt work this way. Bumping the AI's military just results in the PLAYERS building more military. The AI is still at a disadvantage and still loses to a skilled player in warfare. All the larger militaries really do is slow the pace of the game down and ensure that constant attention is given to the military They dont make it much harder overall.

    The only way it becomes truly 'harder' is if the AI continue to have more and more units than a human player can produce. And to do that, it needs massive bonuses. So perhaps going this route removes the ability to 'rush' the AIs, but the whole of the gameplay suffers to prevent a few players from 'abusing' the AI. IMO, its not a worthy tradeoff. Others will differ on that, hence my request for something that simply gives some economic opitimizations whilst leaving the rest of the 'decision-making' intact. It may be flawed, but it allows the game to progress unimpeded unlike the enforced military buildups.

    It not an improvement if the AIs are still playing as if military were the primary goal. It just means that fewer will beat themselve to death on a neighbor, but the arms race continues.
     
  15. Wodan

    Wodan Deity

    Joined:
    Jun 3, 2005
    Messages:
    4,867
    Location:
    In transit
    You're advocating leaving the AI's military management static, while increasing its ability to do other things. This clearly leads to a situation where if you don't attack the AI, it will perform research (etc) faster and better than before.

    The whole point of attacking the AI (and exploiting its weakness at warfare) is a pre-emptive strategy, not a reactive one. (That's why high-level players swear by early rushing.)

    Does that make more sense?

    I agree.

    Clearly I didn't explain very well... when I was talking about "bumping" I wasn't talking about simply building more units. Rather, it's about unit management, effective movement, use of mixed unit types, having cats for counter-stack defense, etc.

    Simply building more units chokes the AI and is what everyone is complaining about right now. I think most everyone is agreed that this needs to be tweaked/changed.

    Wodan
     
  16. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    But if the AI's bonus is reduced to a more manageable level, then the need for that 'pre-emptive strike' is obviated.

    On another note, when playing multiplayer games, highly skilled players often 'rush' too. And other players arent getting any bonuses over them. Its just a preferred way to play for some people (wargame rather than builder game). Its not a bad way to play nor is it 'wrong'. But I feel that the AI is being catered to employ and counter that strategy far too much to exclusion of doing other things. My 2nd game is an indicator of that (see next post).

    My impression was that most everyone was in agreement before this build too. And yet the same strategies are being employed. Rather than pressing people to conform to my standard of what the AI's 'should' be doing, I'd just as soon see a few simply bug fixes for the worst offenders in the original AI. Then, if the wargame style AI is to be pursued there is at least an option for players who prefer the original (and IMO intended) styles of play.

    As I said above, I dont believe that its going to be possible for the AI to be all things to all people. Its either going to be optimized to fight the warmongers (and in turn become a warmonger) or else its going to follow the original path of a more varied AI with commeasurate weakness to a warmongering style of play. I dont say that in a bad way or to disparage the capabilities of the AI team at all.
     
  17. Uncle_Joe

    Uncle_Joe Prince

    Joined:
    Sep 9, 2005
    Messages:
    467
    1/25 build, second test game.

    After losing interest in the first game, I fired up a second with all of the same parameters (modified difficulty, Epic, no Aggressive AI, everything else standard).

    This time, Alexander went balls out attacking in the early game. He quickly eliminated Isabella (good riddance!) and shortly thereafter killed Hatty. Both of these conquests were completed during the early/mid Classical Age (ie, no siege weapons available).

    I, as Russia (Catherine) was somewhat slowed by a pretty crummy peninsular start, but was in 2nd place behind the now-bloated Alexander. He eventually DoWs me and sends in the hordes. My military is smaller, but I am holding my own. He sends stacks and stacks and stacks. Eventually he starts to bludgeon his way through. Its hanging in the balance when Monty hits me from the other side...game over (tip of the hat to Alex).

    What I observed here is that the AI is FAR better about conquest in the early game than later when their is prolific siege weapons available. The 'monster stack' is a huge liability against massed siege and the AI never reacts accordingly.

    But the problem with Alexander's approach was that was all or nothing. After I conceded, I looked at his set up and he was completely stagnant in tech. He had little to no economic buildings or other improvements (outside of Barracks). Yes, he would have beaten me down, but in the end, he ruined himself to do it. If he had been beating himself against another AI rather than the only player, I would again have easily pulled ahead and eventually outteched him to the point where he would no longer be a threat with his backwards piles of units.

    And that is folly I see in the strategy...its not going to possible to WIN the game that early (unless on Pangea). So mortgaging the future to win a few wars is not an effective strategy. You'll eventually run out of steam and be left in the dust.

    On another note, because he was going so balls-out on the attack, about the only way to survive that would be to equally be spamming units. Its evident that Hatty and Isabella didnt do that and they were QUICKLY dispatched. So the existance of this AI strategy again compels military buildup as the only viable strategy, even in early in the game (unless you want to play with fire).

    In order for this AI to work and not convert the game into a simple wargame, there is going to have to be a LOT of moderation put in here. The AIs should not commit to all-out attacks, particularly early in the game. It detrimental to gameplay and to the AI doing it even if it succeeds.

    That was probably the first time I've actually lost militarily to an AI in the early game. In some ways that is a good thing. But in the future, I know to start the spamming of my own military a little earlier if there is an aggressive AI player attacking randomly like that. Which in turn starts the loop all over again...build military, military, and more military. Even if you have no intentions of 'rushing' an AI, you still have to commit to the military route because unless you do, you'll have no chance against an AI that is soley pursuing a military strategy, heedless of the longterm results.
     
  18. jkp1187

    jkp1187 Unindicted Co-Conspirator

    Joined:
    Aug 29, 2004
    Messages:
    2,496
    Location:
    Pittsburgh, Pennsylvania
    What does "modified difficulty" mean?

    I ask only b/c I was positioned next to Alexander in a noble-level 1/25 game (small, techtonics, 8 civs, no aggressive AI) -- we shared a continent together. He did not attack at all...in fact, in the mid-game he actually looks like he is trying for a culture win. If anything, it looked like he hasn't been building enough units (the Mongols were able to launch an amphibious attack and burn one of his cities.)
     
  19. Blake

    Blake The AI Guy

    Joined:
    Feb 18, 2006
    Messages:
    342
    Location:
    Apolyton
    This feedback is gold. In 2.08 an AI would NEVER have been able to conquer two neighbors (at least not without massive luck) so the AI didn't have to worry about "when to stop and consolidate" - it never reached that position in the first place...

    Do you happen to have any saves from this game? Particularly during Alex's wars with the other AI's (I'd like to see his economic stats).

    By the way I'm thinking of tying unit training levels to aggressive AI... so aggressive AI would be militaristic AI.
     
  20. Quagga

    Quagga Former Dictator

    Joined:
    Aug 31, 2006
    Messages:
    663
    Location:
    Li'l Rhody
    The unconditional statement about committing to all out attacks is too strong. I think this after reading a large number of posts from players who wanted to see the AIs launch devastating attacks. Many players requested AIs that go to war more often. Sometimes the result is going to be a benefit to the warring party, sometimes not. Maybe more often not. However, this was what was called for by many posters.

    Speaking for myself, I find totally peaceful games really boring. If the only reason AIs go to war is because I've manipulated them, that's boring too. Sometimes you need that insane leader that goes to war, even though the poor result can be predicted. This happens in real life. :(
     

Share This Page