Investigation of AI reassessment upon reload

Quote:

---

This was on medium campaign difficulty but, as I didn't make a move, you would be forgiven for thinking that that shouldn't matter. The only thing I can say is that the strategic element of RTW is some damn shoddy work.

---

Speaking of which (I forgot to mention this) all of my save games are on the hard/hard difficulty setting. I didn't know about the realtime battle behavior of the 1.2 patch when I first started using it...

The only thing I know of that could bring about that kind of unrest intentionally, and also what I witnessed in my own game, is a series of assassinations so that there was no Faction Leader. I actually wondered if the Brutii had done that. At any rate, this is going very far off topic...

I have AI faction expansion in my Julii SPQR mod campaign with no walls, and it is at the expense of other AI factions. Over 58 turns:

Factions which expanded:

Carthage started with 15 cities and now has 19.
Egypt started with 12 cities and now has 24.
Armenia started with 5 cities and now has 17.
Julii started with 3 cities and now has 14. (human player)
Britannia started with 6 cities and now has 12.
Brutii started with 3 cites and now has 6.

Factions which held their own:

Dacia started with 8 cities and now has 9.
Macedonia started with 9 cities and now has 10.
Germania started with 12 cities and now has 14.
Pontus started with 2 cities and now has 3.
Scipii started with 4 cities and still has 4.
SPQR started with 1 city and still has 1.
Selucid started with 7 cities and now has 5.

Factions which lost out:

Parthia started with 8 cities and now has 1.
Scythia started with 4 cities and now has 1.
Greek Cities started with 11 cities and now has 5.
Gaul started with 15 cities and now has 6. (this is due mostly to Jullii)

If I played this campaign with walls, I think the factions that expanded wouldn't have expanded as much, and the factions that declined wouldn't have declined as much because of the broken off sieges. Yes eventually I will have to fight most of the factions anyway, but I don't have to fight them all at once, and I'm in no position to take on either Carthage or Egypt yet. If I have to confront either Carthage or Egypt down the road it's going to be a big deal because they already have a stronger ecomomy than me, and they'll be expanding it. Carthage is twice as strong as my Julii faction in production and military power, and has 4 times the population. Egypt is twice as strong as Julii in production, equal in military power and has 3 times the population. Dacia, Macedonia, Britannia and Germania are all equal to Julii in production, military power and population.

Thought I'd show you the result of 80 moves without loadgames before I start attacking my own ideas.

This is the starting position in the year 163 BC (winter):

http://home.comcast.net/~b.brassfield/images/Start.jpg

This is the ending position in the year 123 BC (summer):

http://home.comcast.net/~b.brassfiel...es/80Moves.jpg

As you can see from the following territorial rankings, the Brutii did take one territory from Germania, as I reported, and not long afterward, Germania just "ceased to exist". All of its territories turned rebel; at that point, the Brutii began to eat up the formerly Germanic territories, and the rebel territories north of them. Although it was losing territories as it was gaining them, it managed to maintain a net gain until somewhere toward the last dozen moves or so:

http://home.comcast.net/~b.brassfiel...s/germania.jpg

I misreported Pontus a little; while the other factions were focusing on rebel territories, it actually took one Seleucid territory and one Greek territory; only the third was rebel. I wasn't paying as much attention in that part of the world and the color coding is pretty close. After that, Egypt managed to take three Seleucid territories -- whether they became rebel first I can't remember. But the last two definitely became rebel before Pontus took one of them. The other one was never claimed.

http://home.comcast.net/~b.brassfiel.../seleucids.jpg

The situation with Thrace and Macedon occured right at the very end of the 80 moves:

http://home.comcast.net/~b.brassfield/images/thrace.jpg

Like the situation in the north, most of the other acquisitions were rebel territories and the situation remained otherwise very static for the entire 80 moves. Some countries got smaller, others got larger, the Seleucids were beat by a war on two fronts, and the Brutii got lucky through mismanagement or something on the part of Germania, but aren't doing a much better job of managing those territories themselves.

However, if you'll look closely, you'll notice that there is a long period prior to the beginning of these 80 turns, of about 50 years or so, when all of these countries had no growth and no losses (except Germania, which was an altercation with Brittania and ended with Germania ultimately becoming Brittania's protectorate.)

It was during these 50 years that I did most of my expansion, and almost certainly never skipped a single turn without a load game.

If I go back to the year 213BC, here is what the territorial map looks like:

http://home.comcast.net/~b.brassfield/images/213BC.jpg

Now, at least you know that I'm not lying about loading each turn. Hehe. As you can see, with the exception of my target factions and myself, there have been no changes except for the altercation between England and Germany.

I pressed "End Turn" with no load games only 16 times, and here was the political landscape in the year 205BC:

http://home.comcast.net/~b.brassfield/images/205BC.jpg

Pontus made all the exact same moves, but 50 years earlier. The Seleucids expanded a little, and it's obvious that Germania is not about to cave in on itself so easily. Almost all of the rebel territories were claimed immediately, as with the 80-turn test. It's harder to just press "End Turn" for a long number of moves at this point, but I think it's safe to say that with 164 turns remaining until 123BC, things would turn out slightly differently than with the 80 turn test. (Except, of course, for my own acquisitions from Gaul, Spain, and Numidia.)

My immediate targets (Gaul, Spain and Numidia) benefited only slightly from rebel (and other) acquisitions -- most importantly, Spain's acquisition of the last remaining Carthaginian stronghold of Corduba which took them 11 turns, and which they succeeded in occupying not long before the same year that I invaded Spain. Considering how badly I plowed right through Spanish territories when I was actually playing, I can't see how this small amount of extra unit production would have helped them very much. Numidia also acquired a rather inconsequential village at a small oasis in the Sahara.

I similarly question how much these changes would affect my long-term targets...at least in terms of the relative strengths of different AI territories. The Brutii are going to find it harder to take a unified Germania than their rapid grabfest of rebel territories, and the Seleucids might be a little stronger but they are still caught in a battle on two fronts with Pontus and Egypt -- not that the Julii care much about that corner of the map. Although those conflicts might take longer to resolve under these conditions, they will probably end up about the same. And I suppose Thrace will probably still wait 60 turns before attacking Macedonia. Who knows?

In any case, I think it also makes a strong argument for the powerful affect of the AI acquiring rebel territories. Because each faction can use its rapid rebel acquisitions to strengthen itself, individually. That will help when they're forced to ally against me.

The AI seems to do a fairly good job of maintaining a balance between the AI factions and the human player. One must realize that chess has been around for thousands of years, and that the algorithms and math studying the game had been in existence long before the first computer program.

In fact, if you ever study the field of combinatorial optimization, you will discover that the Hungarian method of two-dimensional bipartite matching was proposed and published by Kuhn all the way back in 1950. In a game as complex as RTW, the number of combinations of possibilities further complicated by rule-based hierarchy or even simple topological relationships, far exceeds the complexity of even three-dimensional bipartite matching. Since there is no game exactly like the TW games, and the developers cannot turn to a text book for canned algorithms, but must nevertheless write code which causes the AI to select its moves in a reasonable period of time, I think they've done a fairly admirable job of maintaining some sort of game balance and general satisfaction on the part of the average player. Like almost all computer games of its kind, it will fall to the player once they have achieved mastery over the AI; just as Chess Masters were able to consistently beat early chess programs. Despite its weaknesses, I consider the AI to be quite good compared to the plethora of other computer games which do not attempt to rely on AI to create a challenge for the human player whatsover. I also consider lambasting the AI to be in poor taste.

The thread has been cleaned up and I'm trying to formulate a tentative conclusion concerning the siege behaviour. Should you disagree with any of the following, please explain why.

• It is certain that after loading a quicksave (has standard save been tested?), the AI will most of the time break up ongoing sieges in cases where it doesn't do that without save/load.
• It seems that the objective to take the previously sieged city is not lost entirely, as the AI will usually try to reengage the siege the turn after the load.
• As a consequence, saving and reloading will affect the development of AI factions, at least in the beginning of a campagne.
• It is not yet clear to what extent the course of a campagne is influenced in an undesirable way by this. AI factions do conquer new territories at a slow pace, but it is not clear whether this is due to the save/load issue or general weakness of the AI. There are three possibilities:
  
  • Above the weakness of the strategic AI, this issue is irrelevant as its effect is only noticeable under extreme conditions.
  • Saving and reloading often does noticable affect gameplay, but doesn't make the game unplayable.
  • The effect is so pronounced that it can be called a "game-breaker".
• A clearly noticable aspect is that, when saving and loading often, AI factions will not take the opportunity to capture rebel territory to the same extent as it does in continued playing.

Saturnus, I will respond this way. Some of them are not a yes or no.

1. It is certain that after loading a quicksave (has standard save been tested?), the AI will most of the time break up ongoing sieges in cases where it doesn't do that without save/load.

Yes, it is CERTAIN that MOST of the time they will break ongoing sieges. ~;) (Yes, I've tested the standard save.)

2. It seems that the objective to take the previously sieged city is not lost entirely, as the AI will usually try to reengage the siege the turn after the load. Not only that, armies enroute to siegies will not lose their objectives.

I agree. However, you need to qualify this so that other readers understand that even after multiple reloads the target cities are never abandoned, which results in a state wherein cities are being besieged and then having their sieges relieved.

3. As a consequence, saving and reloading will affect the development of AI factions, at least in the beginning of a campagne.

There, I do disagree. As a consequence, reloading a game will have an immediate affect on all rebel territories that can be easily acquired. This phenomenon exibits itself througout the game, not just at the beginning, and is the greatest effect of the phenomenon, rather than speculative projections on its effects throughout the course of gameplay. It can potentially also have an effect on the controversy between different AI factions, which will result in the occasional transfer of territories and an overall game balance for when the human player is ready to confront them. Such controversies are partially "locked in place" by the AI based on their alliances with each other, and can be further influenced by the human player.

I will respond to the last one after a while. I'm having some rather strange feelings about all this, since so many people debate against me and no one has responded in almost two days. I almost feel boycotted...

Quote:

---

Originally Posted by A.Saturnus

The thread has been cleaned up and I'm trying to formulate a tentative conclusion concerning the siege behaviour. Should you disagree with any of the following, please explain why.

---

• It is certain that after loading a quicksave (has standard save been tested?), the AI will most of the time break up ongoing sieges in cases where it doesn't do that without save/load.
  Agreed. I'm also happy that quick and normal saves create the same problem, as my first every test (not referred to in this thread) used normal saves.
• It seems that the objective to take the previously sieged city is not lost entirely, as the AI will usually try to reengage the siege the turn after the load.
  Agreed. This may just be a consequence of the reevaluation coming to the same conclusion though (i.e. it is possible that the objective is lost and we just don't discern a difference).
• As a consequence, saving and reloading will affect the development of AI factions, at least in the beginning of a campagne.
  Agreed. A worrying development arising from some of the long campaign tests run is that the AI may be reaching a critical mass at which it just can't manage its empire in a halfway sensible manner (independent of the loadgame issue). That though is for a different thread (i.e., for the purposes of determining whether the AI is adversely affected by the loadgame bug we'll have to assume that the AI is functional).
  RB's point is correct, in that if you defer the start of the campaign (as far as the AI is concerned) by loading every turn, then the initial Rebel land-grab is deferred. It's further correct in that, assuming the AI is competent, this issues affects faction on faction sieges in the late game as well.
• It is not yet clear to what extent the course of a campagne is influenced in an undesirable way by this. AI factions do conquer new territories at a slow pace, but it is not clear whether this is due to the save/load issue or general weakness of the AI. There are three possibilities:
  
  • Above the weakness of the strategic AI, this issue is irrelevant as its effect is only noticeable under extreme conditions.
  • Saving and reloading often does noticable affect gameplay, but doesn't make the game unplayable.
  • The effect is so pronounced that it can be called a "game-breaker".
  
  Assuming that the strategic AI is competent, this effect is extremely relevant if you play sessions of 4 turns or less. If you play for 7 or more, it probably doesn't have a huge effect. As some people really do play for less than 4 turns, it will be a game breaker for them (assuming that the AI is otherwise sound).
• A clearly noticable aspect is that, when saving and loading often, AI factions will not take the opportunity to capture rebel territory to the same extent as it does in continued playing.
  Disagree (semantic though). It will take the opportunity, in that it will march armies out to get the cities, but it will not succeed as it will be forced to break off. The same applies to trying to capture other AI factions, although diplomacy, walls, more coordinated defence, etc. mean that it needs even longer per city than with an average Rebel city.
  So, from my testing, it seemed that the AI was able to reasonably expand if the load interval was 4 turns. However, the Scipii were not able to do well in Sicily with less than an interval of 5 turns. Hence the supposition that there is something that makes taking other AI faction cities harder for the AI than taking Rebel cities (which seems like common sense, but it's nice to have it confirmed).

Quote:

---

There, I do disagree. As a consequence, reloading a game will have an immediate affect on all rebel territories that can be easily acquired. This phenomenon exibits itself througout the game, not just at the beginning, and is the greatest effect of the phenomenon, rather than speculative projections on its effects throughout the course of gameplay. It can potentially also have an effect on the controversy between different AI factions, which will result in the occasional transfer of territories and an overall game balance for when the human player is ready to confront them. Such controversies are partially "locked in place" by the AI based on their alliances with each other, and can be further influenced by the human player.

---

Actually, I don't quite understand how you can disagree with that, given that it is more or less a logical conclusion of the first proposition you agreed with. How can the development of a factions possibly be the same when the AI cannot bring most sieges to an end? Note that this point says nothing about the size of the effect. I only said that save/load makes a difference at all.
From my understanding, you argued above that the effects others found at the beginning of a campagne are an artefact of an extremum. You aknowledged the fact that the AI will, for example, not succeed in conquering Sicily early on as Scipii if loaded every turn, but you objected that this has no impact on the later development. If this is a correct description of your position, you cannot disagree with this point.

Quote:

---

Originally Posted by Bromley

• A clearly noticable aspect is that, when saving and loading often, AI factions will not take the opportunity to capture rebel territory to the same extent as it does in continued playing.
  Disagree (semantic though). It will take the opportunity, in that it will march armies out to get the cities, but it will not succeed as it will be forced to break off. The same applies to trying to capture other AI factions, although diplomacy, walls, more coordinated defence, etc. mean that it needs even longer per city than with an average Rebel city.
  So, from my testing, it seemed that the AI was able to reasonably expand if the load interval was 4 turns. However, the Scipii were not able to do well in Sicily with less than an interval of 5 turns. Hence the supposition that there is something that makes taking other AI faction cities harder for the AI than taking Rebel cities (which seems like common sense, but it's nice to have it confirmed).

---

I may have misunderstood something, but it seemed to me RB was saying that:

Quote:

---

The most obvious and pronounced effects are the way that AI factions will grab rebel territories -- several of them in just a few turns, if they can -- thereby significantly increasing their production power.

---

If I'm reading roguebolo's position correctly, it basically comes down to the premise that the AI plays so ineffectively normally, that the additional penalties caused by the save/reload problem are rendered moot.

In general, I agree with that position. Even with no save/reloads, the AI doesn't expand as much as would be necessary to challenge a human player. Empires end up being relatively stagnant.

However, I disagree with the conclusion that the save/reload issue isn't a problem in and of itself. Any attempt to make the AI more aggressive would be rendered ineffective by this problem. Basically, I'd conclude that the only reason it's not a serious problem is because there's another serious problem. Fixing just one or the other won't work nearly as well as fixing both.

Bh

Bhruic said it so well.


P.S.
Only reason why AI is good against rebels (without loads) is that rebels don't build new units in their cities.

3. As a consequence, saving and reloading will affect the development of AI factions, at least in the beginning of a campagne.

I'm only disagreeing with the wording because I think the description is incomplete, and because I think the development of the AI is most effected when numerous rebel territorories are available, which is true at the beginning of the game but might also be true in other situations. I suppose this degresses into semantics, so I will reverse my statement by saying that I agree with a number of qualifications.

Bhruic, the "other problem" extends itself to other computer games too. Most AIs in most games can eventually be mastered by a human player. Part of the problem has to do with the computing power of computers. "Combinatorial optimization" is a field of algorithmic study whereby you attempt to reduce the number of combinations that need to be analyzed to arrive at a conclusion.

It doesn't surprise me when I master the AI in a game that I once considered challenging; rather, it seems to be par for the course, even with simpler, more straightforward strategy games like "Massive Assault." I think there's a lot of room for improvement of AIs in general and in specific for any given game. However, upon mastering an AI I don't come to the conclusion that it is horribly scarred because I can beat it consistently. I might have some ideas on how it can be improved, but that's about it.

Quote:

---

Originally Posted by roguebolo

Bhruic, the "other problem" extends itself to other computer games too. Most AIs in most games can eventually be mastered by a human player. Part of the problem has to do with the computing power of computers. "Combinatorial optimization" is a field of algorithmic study whereby you attempt to reduce the number of combinations that need to be analyzed to arrive at a conclusion.

It doesn't surprise me when I master the AI in a game that I once considered challenging; rather, it seems to be par for the course, even with simpler, more straightforward strategy games like "Massive Assault." I think there's a lot of room for improvement of AIs in general and in specific for any given game. However, upon mastering an AI I don't come to the conclusion that it is horribly scarred because I can beat it consistently. I might have some ideas on how it can be improved, but that's about it.

---

That's going well beyond the scope of the discussion. It doesn't matter whether or not other games have issues. We're only talking about RTW, and the specific issues it has that are related to this bug.

One of the major points is that the AI is not sufficiently aggressive in general, so unlike MTW, there doesn't tend to be much empire growth. This fact, I believe, "masks" the problem that the save/reload issue causes.

If you choose to agree or disagree with that point, that's fine. I'd love to hear your explanation. But you do seem to have a tendancy to jump to off-topic issues that don't really apply to save/reload issue.

Bh

Quote:

---

I'm only disagreeing with the wording because I think the description is incomplete, and because I think the development of the AI is most effected when numerous rebel territorories are available, which is true at the beginning of the game but might also be true in other situations. I suppose this degresses into semantics, so I will reverse my statement by saying that I agree with a number of qualifications.

---

Ok, I don't insist on the wording. What is more important is point four anyway. Do I assume right that you agree with point five?

Quote:

---

Originally Posted by A.Saturnus

I may have misunderstood something, but it seemed to me RB was saying that:

---

I thought my post was quite good considering I was wasted ~:cheers: .

The point I was trying to make was that I basically agree with what you were saying in 5 (A clearly noticable aspect . . .). The disagreement was only with the wording, specifically, "will not take the opportunity." Not an important point by me - just amplified in my mind by alcohol.

It is interesting this... Now we are finally getting something of a specific direction.

I have just considered something.
Even if the game solely affected by the loadgame issue from its initial positions, then it is still a tough situation for many gamers. They will often suffer the effects anyway, as they don't just kick back and watch for 10-20 turns like we do here. They play and load from the get-go.
A game that more or less forces the player to wait up to 20 turns is not working as it is supposed to. That is how I view it.

I don't know that anyone is questioning whether it is working as it supposed to. The issue right now seems to be the severity.

In general, I find that early turns go by quite quickly. It's not hard to play 10-15 turns in the amount of time it would take to play 2-3 in later stages. So it's less likely people will be in a save/reload situation early on.

Given the three choices A.S. listed, I'd have to go with "Saving and reloading often does noticable affect gameplay, but doesn't make the game unplayable."

Bh

i wonder if our use of the term "reload" can be misleading? it can cause some people to think that we're exploiting the game by "reloading" simply because something doesn't go our way. i would say simply "load" or "loadgame" or such...

Nice to see that some conclusions are coming out of this. I was so busy lately that I quite frankly didn't have time to sit in front of a comp for fun at all. But now that I do have some time to spare, I cannot bring myself to fire up RTW any more, and I don't see myself doing so in near future either - I realise I'm too disappointed in the game to do so. Sorry.

The latest update of the RTW FAQ addresses this issue.

Quote:

---

Originally Posted by The Shogun

Q. What's the latest regarding the load/save/siege issue.

A. Much I want to say this is now sorted and carved in stone, I can't. Please remember that expansion pack is in development and there is still much going on at the moment that I can't talk about for various reasons. I am prepared to say that it is certainly our intention to address the load/save issue in the expansion pack. "And what about a patch?" I hear you ask. Again I can't say this is set in stone but we hope to bring out a patch at roughly the same time as the expansion pack (in reality it will probably be few weeks later).
That's the state of play at the moment. When I know more, I'll make sure that you know more.

---

One of the fortunate Swedes (goes by Stalin on TWC and Pippi Longstocking on the .com, the combination of which kinda reminds me of the Bert and Osama photo, but I digress) was kind enough to post this on the .com "bugs after patch 1.3" thread

"from my rather short observation the siegebug is gone...
https://img189.imageshack.us/my.php?image=before8gp.jpg
https://img189.imageshack.us/my.php?image=after1es.jpg "

I will plead with him to read this thread and do a proper test, but if he's right, I can't blame him for getting on with playing the game and ignoring me.

Seems good. The screenshots don't do much for me, but for players to says so makes it much more so. The shot only shows one case, but I suppose they have tried several cases each.