A Better AI.

[klokwerk]

I just played another game, my first with the 12/21 build. Monarch, custom continents (3 continents), standard size, aggressive AI on

It was disappointing, far too easy to my taste. I was Shaka and started near Toku and Asoka.

Unfortunately, Asoka built too few defenders, it was maybe worse than before. Most of his cities had only 1 or 2 archers to defend him, when I came with my stack he was crushed in no time. He had far better tech than me (except the military ones where it was even), but his defense was so poor I could have killed him with warriors So please, put more defenders in the builders' cities. With less than 3 defenders, it's just suicidal... Somehow Asoka should have understood I would crush him in 5 minutes if he kept 1 archer in each city and didn't research military techs... Pure suicide.

Another major trouble, I could build easily all the wonders I wanted, the great lighthouse, the colossus, the hanging gardens... Probably something wrong here too ?

Maybe a bug somewhere ? It wasn't that easy before.

I'll play again without aggressive AI since I don't seem to get better results with it. Maybe I was just lucky (again).

The good points :
- early wars between AI, and they work ! Cyrus vassalized Monty and destroyed Germans before I met him. It gives the hope to see an AI win by domination one day.

The bad points :
- still too few defenders (1 archer in a city ??), especially the builders...
- ancient wonders are too easy to build. With colossus, great lighthouse and hanging gardens, it was even easier for me...
- not enough aggression, especially on 'aggressive AI'

Couldn't test the naval invasions you added, that didn't happen in my game.

Anyway, nice work, though your changes don't seem to work exactly as advertised

I try another game after christmas and I'll report here again.

 

[Quagga]

I too have been playing a 12-21 build game.

I just paused from it because I got a crash-to-desktop when I moved a small stack of military. They had been in a city I captured, but I lost it to culture flip. I was just relocating them to one of my cities when the screen went dark. I'll try tomorrow to see if I can get a save game that demonstrates it.

EDIT: Save attached. Move the selected stack to my closest city in order to see the crash.

I dropped back to Prince, thinking that the AIs would be stronger than in the 12-12 build. I don't think so. Maybe the opposite.

As the previous poster noted, no naval activity so far in this game. Two continents.

Wonders were built very slowly by the AI. I had decided up front to resist building any, but I couldn't believe how long most of them lasted.

A stack of Ragnar cavalry should have taken out one of my cities, but didn't. He'd attack with one or two fresh cavalry, then just sit there maybe waiting for the rest to heal. They weren't doing that because I kept rolling up cannon and giving them the business.

In one of my cities, I set emphasise commerce. I was building a wonder (couldn't resist after all; it was a marginal city). It did not prefer commerce, but kept on prefering production. Other cities behaved properly, so maybe this had something to do with wonder production.

 

[WuphonsReach]

I also had a single CTD, but reloading one of the automatic saves and continuing on and it didn't happen again. IIRC, I had just tried to move units against an enemy civ.

...

The AI's choice of promotions for Infantry seem a little inflexible. Cyrus in the most recent game had a ton with CG upgrades... but I'm not sure if those were new builds or purchased updates. (I had just gifted Cyrus a bunch of techs and a few thousand gold now that he capitulated to me. He used the money to update his units from older ones.) My final war of domination was mostly done with Infantry + Cannons.

I tend to use Infantry with a wide variety of promotions (playing as Tokugawa, running w/ +2 Vassalage and +2 Theocracy when preparing my SoD). My units were coming off the line with 9/10 XP since I think I also had Pentagon. They'll be roughly split as:

- City Defense: C1 (+10% STR), CG3 (+75% city def, +10% vs melee), Drill 1

- Anti-mounted: C2 (+20% STR), CG1, Formation (+25% vs mounted), Drill 1

- Assault: C3 (+30% STR) and Pinch (+25% vs gunpowder), CG1, Drill 1

- Medics: C1, Pinch, Medic, CG1, Drill 1

Which may or may not be a good way to setup my SoD. But my units at this level are Machine Guns, Cannons, Calvary and Infantry. So my Infantry are doing the heavy lifting as "boots on the ground".

Calvary were coming off my production line as 11/17 XP w/ C2+Pinch or C1+Flank+Sentry.

Note: I'm playing down at Chieftan level for a few days. So the AI gets huge penalties which may make it impossible to properly promote their units. Once I had Pentagon + West Point + GG trainer, I had Infantry coming off the assembly line with 15/17 XP.

 

[JimMac99]

Just finished my first game with the 12-21 build and maybe it’s a co-incidence that my report follows those above, but it was my easiest Warlords game ever!
I guess I am a fairly casual Warlords player having started only around 30 Noble games, and finishing exactly 18; but this is my first victory! Yup: my first ever win (A score victory)

Some points:
-I got 16 wonders! I have never previously got any where near this number of wonders in one game . I wonder if whatever was supposed to be tweaked to increase the likelihood of the Ai’s building wonders was tweaked the wrong way?
-I had random map selected, which turned out to be small islands
-I didn’t have one CTD
-I drew the Spanish as my random civ. I found myself in first place in science and score early on and this rarely altered.
-My first war happened 2/3rds into the game when the 2nd placed civ declared war on me. Although he had lots of frigates bombarding my cities, the few landings did not cause any trouble. I noticed though that AI units would appear to sit and do nothing on my land if the main thrust of their attack was lost. They really should get back to pillaging if they can’t attack. Perhaps they were healing?
-Victoria declared war on me with 60+ turns remaining and stayed at war until the end of the game, yet never attacked me. What was the point?
-My governor management of specialists was the best yet of any build, no more building priests etc when I didn’t want them.

I’ve played a game with pretty much each build released so far, and it seems to me that the most challenging build was the 12-2 or 12-9. Still it was good to win for a change!

 

[WuphonsReach]

One thing I definitely notice in the latest build is the AI will whip defenders when I'm sitting at the gates. Which brings a new factor into conquest wars. When going up against a heavily defended city in the age of Slavery (and maybe the "draft" civic), you need to bring enough seige weapons to tackle the city defenses in a single round.

If it takes me 3 turns to take down the defenses, that's usually 3 additional units that I'll have to kill. Hopefully the AI is smart enough to calculate this into its war plans.

...

I do have one issue with city placement. Isthmus are not always taken into account correctly (maybe / maybe not).

This is a tough problem, because the AI may or may not have the knowledge at the time to determine whether a location is a key isthmus like a Panama location or not. Ideally, we'd do a calculation to determine whether ocean on one side is reachable from the other side within 5 turns.

I razed an Egyptian city in my current game (that I sent to BetterAI@) in order to reposition it on the isthmus location that cuts dozens of turns for transit.

...

Follow-up note: It looks like the AI may be too eager to chop forests on Tundra tiles. Since tundra tiles can't be used for anything except lumber mills (right?), maybe the AI needs to treat city location and tile improvements different for those until mills are ready.

 

[Powerslave]

I just played another game, my first with the 12/21 build. Monarch, custom continents (3 continents), standard size, aggressive AI on

Continents, Monarch, 12/12 (sorry, didn't download newest update yet), standard size, no options set.

It was disappointing, far too easy to my taste. I was Shaka and started near Toku and Asoka.

I was Augustus and started next to Kublai Kahn.

Unfortunately, Asoka built too few defenders, it was maybe worse than before. Most of his cities had only 1 or 2 archers to defend him, when I came with my stack he was crushed in no time.

Argh! I started with Iron in Rome's radius, started pumping out Praetorians with much evil glee, and -- inexplicably -- found myself losing the war. I don't know what happened! Oh, I took a couple cities in my initial blitzkrieg, but after that, it was a slaughter, and I wasn't doing the slaughtering. I lost my entire army, and he started taking back his cities. Kublai had millions of defenders in each city, and they all had quite appropriate promotions. After close to a thousand years of stalemate-ish war, I finally decided to settle. A dozen turns later, Kublai came at me with a vengence, and he wiped me off the face of the planet.

I guess I didn't build a large enough invasionary force, and I didn't press my advantage as hard and quickly as I should have. I also should have, ideally, picked on an easier opponent. However, we're talking about Rome with iron!! You can not imagine how humbled I was after this game.

Anyways, I wanted to post a counter-story, to show that the AI does occasionally do well in wars. Maybe it was an off day for me, but I got slaughtered. Once I recover my dignity, I'm going to play a duel game and nuke his ass back to the stone ages.

Good job to the Better AI team.

 

[JimMac99]

Continents, Monarch, 12/12 (sorry, didn't download newest update yet), standard size, no options set.

Anyways, I wanted to post a counter-story, to show that the AI does occasionally do well in wars. Maybe it was an off day for me, but I got slaughtered. Once I recover my dignity, I'm going to play a duel game and nuke his ass back to the stone ages.

Good job to the Better AI team.

Good to read. I think part of the reason I had such a surprisingly easy win was I got a lucky start: only the Japanese and I on quite a large island, I quckly blocked him off and took most of the island for myself... and when I got the Great Wall he had all the Barbs to deal with.

If you get a chance though, I'd like to know how you get on with the 12-21 build.

 

[Quagga]

Last night I forgot to mention the one thing I really wanted to, which is about the AIs gold trading behavior. I read some posts that may have been about the same thing, but about the 12-12 build, so maybe this isn't new with 12-21. If that's true, I don't know how I failed to notice, since I played many games with 12-12.

The issue is that the AIs are dramatically less forthcoming about offering gold in trades. Here's how it looks: I offer the AI a tech; it has gold showing that it can trade. What would you offer, I say. It responds how do you like this? And this is nothing, nada, zilch. Not much of a trade. I can volunteer all his cash and then say, how about this deal? The AI accepts.

Similarly, where it might have offered Mysticism plus 120 gold for Alphabet, now it only offers the tech. Manually add the gold to the equation and the AI accepts.

Now this isn't a big deal, once you know it is happening. But CivIV took away most of that bargaining stuff familiar from CivIII. Somehow, it seems to have crept back in. I don't think it's an improvement. To me it looks like a bug.

But to end on a positive note: I've really enjoyed trying out the Better AI builds. Besides giving some variety to the game, it's interesting to see how the AI behaviors can be changed, which in turn alter the flavor of the game. I think it's really cool and I thank the Better AI team for their efforts.

 

[cymru_man]

I've also noticed that while the AI is trouncing me in military and actually kicks my ass when it does get round to invading me, its wonder building tendencies seem reduced.. well, in my experience Temple of Artemis/Great Wall seems to come a lot earlier, but Stonehenge, Oracle later, and Parthenon about the same. GL sometimes comes very early. In general though, they seem to be a lot less eager to build wonders than in earlier builds - I typically play large pangea, 12 players.

 

[klokwerk]

From the reports above it seems the 12/21 build is strangely easier than the 12/12, there could be a bug. In my 12/21 game, Asoka had only 1 archer or 2 per city and it was something like 100 AD... I built most of the ancient wonders of the world without even being industrious, there's definitely something wrong there too.

 

[Quagga]

The AI is still getting Feudalism surprisingly early. I was just boxed in by Mao who had Longbows in 350 BC -- that's when I noticed them, anyway. (I had no way to get iron, so decided to start over.)

EDIT: Monarch/Continents/Epic.

 

[Blake]

One thing I definitely notice in the latest build is the AI will whip defenders when I'm sitting at the gates. Which brings a new factor into conquest wars. When going up against a heavily defended city in the age of Slavery (and maybe the "draft" civic), you need to bring enough seige weapons to tackle the city defenses in a single round.

If it takes me 3 turns to take down the defenses, that's usually 3 additional units that I'll have to kill. Hopefully the AI is smart enough to calculate this into its war plans.

...

I do have one issue with city placement. Isthmus are not always taken into account correctly (maybe / maybe not).

Hmmm I do know of a way or two to account for Isthmus, I might implement one of them. This method would give a bonus when a city borders on two separate (and large) bodies of water, except that wouldn't work when both sides are the same body of water - for that it'd be necessary to calculate the pathing distance from each water tile to each other water tile, and that's too expensive. And to be honest I'm not sure having those canals is so important, it's just "cool".

Follow-up note: It looks like the AI may be too eager to chop forests on Tundra tiles. Since tundra tiles can't be used for anything except lumber mills (right?), maybe the AI needs to treat city location and tile improvements different for those until mills are ready.

*shrug*. I chop tundra forests and I make the AI in my image (more or less), the chance of getting a tundra city which doesn't have enough hills to soak up food is very, very low. A tundra city which does have a surplus of food can use it more productively on specialists. And a tundra city will almost never grow as large as grassland cities so the health wont be a factor. The influx of hammers on the other hand is invaluable to the cities development, accelerating it by 30 or even 60 turns.



Okay on the WAR issue. Certain leaders - mainly Americans, Indians and possibly English are really kind of pacifist by personality, they are kind of just going to be rollovers in war a lot of the time, not always mind you, if they get some strategic resources they'll be far more likely to mount a proper defense. Nearly all leaders except those ones (and even those ones sometimes) will immediately start employing the Dagger strategy when war is declared upon them, causing them to militarize to a huge extent.

Part of the problem here is that certain leaders "Gamble" on being peaceful, but because the player knows their personality it's a gamble they can be sure to lose. If it's changed to make them more militaristic then they're (at least sometimes) playing against their personality.

I'll probably end up teaching the AI some human tricks, like giving them a fair chance to gank an exposed worker in the early game if relationships are frosty, although Gandhi and Roosevelt will still basically be militarily pushovers in the early game since they wont resort to such tricks anyway...


Final topic: AI gold trading. Are those having problems with it playing Vanilla or Warlords? It's apparent than there was a problem with the automated merge which is caused Vanilla to not work properly, if anyone is having problems in Warlords then that's a more serious problem.
EDIT: Nevermind what I just said about AI gold training. I found a bug (both versions) causing it to mess up. So that's fixed now.

 

[DevilJin]

One thing I definitely notice in the latest build is the AI will whip defenders when I'm sitting at the gates. Which brings a new factor into conquest wars. When going up against a heavily defended city in the age of Slavery (and maybe the "draft" civic), you need to bring enough seige weapons to tackle the city defenses in a single round.

Okay on the WAR issue. Certain leaders - mainly Americans, Indians and possibly English are really kind of pacifist by personality, they are kind of just going to be rollovers in war a lot of the time, not always mind you, if they get some strategic resources they'll be far more likely to mount a proper defense. Nearly all leaders except those ones (and even those ones sometimes) will immediately start employing the Dagger strategy when war is declared upon them, causing them to militarize to a huge extent.

Part of the problem here is that certain leaders "Gamble" on being peaceful, but because the player knows their personality it's a gamble they can be sure to lose. If it's changed to make them more militaristic then they're (at least sometimes) playing against their personality.

I'll probably end up teaching the AI some human tricks, like giving them a fair chance to gank an exposed worker in the early game if relationships are frosty, although Gandhi and Roosevelt will still basically be militarily pushovers in the early game since they wont resort to such tricks anyway...


.

The AI is building less defenders in their cities. and in the notes file of the 21/12 build you said it will build much more defenders. The city can't rely on the slavery(also when the city is being slaved for defenders only in trouble time, the defenders won't get the max fortify bonus, thats for sure). Not anyone will run slavery all the time, and if so, not every city will have enough population for it. Also I can (and I usually do)attack cities 1 or 2 turns after I declare the war so it won't help anyway.

Furthermore.. you didn't tell anything about the big slowness that the AI is building the wonders. In the monarch level people get almost all the wonders they want.

Hope it will help for the next build.

 

[Blake]

The AI is building less defenders in their cities. and in the notes file of the 21/12 build you said it will build much more defenders.

Also.. you didn't tell anything about the big slowness that the AI is building the wonders.

Geez it's Christmas morning for me (and I'm debugging code, wooohoo!).

The AI wont be building less defenders in it's cities, maybe not more under some circumstances but I seriously doubt it'll be building less. In a real hot war it should really spam out a lot more units.

The most important question is the difficulty, AI's at prince and lower play kinda like human noobs in multiplayer, they have a tendency to go too light on defense.

AI Wonder building is tremendously dependent on them having stone or marble (or being industrious), also on production impaired maps (like Archi) they don't know to build up a good wonder building city.

In any case in terms of being "viable" the AI is only really still viable on Monarch+, at that point it can put up a real defense and build wonders fairly well.... at lower difficulties the lack of "strategic direction" is it's undoing still...
As an example of this techs with an unclaimed wonder on them (ie Oracle) are giving a random-roll bonus to their value, so the AI will randomly choose to pursue a wonder tech. When cities choose what to build there is another random roll added to wonder build values. So to get the Oracle, the AI has to pass two random rolls - one to research Priesthood, the other to actually start the Oracle. How it should be of course is the AI should decide to build the Oracle and then research Priesthood and start the oracle ASAP, but the CIV AI is not "objective based" in any way whatsoever.

Reduction in wonder building, if real, is because the AI will now be less likely to engage in harmful wonder building, in all honesty in CIV wonder building is usually a bad idea - workers, settlers and military is a better investment. The solution to this is probably not increasing the wonder build rate, but increase the conquer-wonder-builders rate... the presence of expensive wonders should be a strong incentive for war.

 

[DevilJin]

Well.. The AI sould at least know about the specific importance of the wonders. For example,the pyramids- if the AI wants a specialist economy , and also in Archiplago map its still too easy to build the great lighthouse and the colossus before the AI in monarch(and in monarch the production of the AI is far better than the human).

 

[weimingshi]

Continued my save game with 12-21 build. The sea battle definately increased. Hannibal declared war on catherine two continents away.And I already saw 4 large fleets from hannibal sailing toward catherine. Hannibal is just below me, good thing he's pleased with me. The thing that bugs me is catherine, she is also two continents away from me and she is cautions with me. While hannibal has 4 large fleets on way to her. She send 2 fleets over. But the thing is she didn't send those 2 fleet to hannibal, instead she declared war on me a few turns later. I have 12 frigates stationed in the water, and just recently mass produced calvary and cannon. my guess is catherine probably set sail to invade me before hannibal declared on her. But once hannibal declared on her and invasion fleet on way, the AI should stop and reconsider the situation, instead keep going with original plan. I mean catherine has nothing to gain now, my 12 frigate can total destroy her navy and hannibal can land all the troops he want on her soil now.

 

[Blake]

Well.. The AI sould at least know about the specific importance of the wonders. For example,the pyramids- if the AI wants a specialist economy , and also in Archiplago map its still too easy to build the great lighthouse and the colossus before the AI in monarch(and in monarch the production of the AI is far better than the human).

As I said the AI is not objective based, at this point the AI cannot think "Well gee I want the Great Lighthouse" - instead it has to pretty much randomly build a lighthouse followed by randomly choosing to build the Great Lighthouse (maybe you've noticed that at this point it's not possible to queue up a building you can't yet build but will be able to after the current build, it's an unfortunate interface flaw and it applies to the AI too), it's obviously going to take a while unless it gets lucky (while buildings are given a bonus if they unlock a wonder, this bonus obviously must not be too overpowering since being able to build the wonder is usually a relatively small benefit). There's no way for the AI to know it's playing on an Archi map, it has to guess this from what the neighborhood looks like, at this point the AI can not really have map specific strategies - the AI is adaptive and this has its blessings (if it looks like a duck and quacks like a duck the AI will treat it like a duck... or in other words there's no need to worry about accurately setting map types and stuff).

There's even a dilemma here when it comes to cheating. Take Terra map for instance, should the AI know that there is a second uninhabited continent out there in the black fog? It is actually pretty trivial to just "maphack" this information but the AI would be cheating, not in some cases since the human knows they selected Terra, but maybe in other cases like they chose Shuffle instead or it's a crafted scenario which has been designed as terra-ish but none of that info is given to the player - it's pretty hard (not sure if it's even possible) to tell what map script was used or if shuffle was used and in any case it wouldn't be right to hardcode behaviors to map scripts. The Human always has equal or better knowledge of the map - they either know what the map is, or they know they don't know.

The AI basically has to always use Generic Strategy, or be behind the curve on adopting the proper strategy, or cheat to get the strategic information. Cheating is sometimes ok, but sometimes it isn't.

I am of course constantly trying to come up with new ways for the AI to adapt and employ strategies, but I'm just sharing some the issues involved here.

 

[DevilJin]

I Don't think that it's a cheating if the AI is heading for knowing astronomy in a terra map for example, even if the map was chosen as Shuffle. It's not cheating , this is just better decisions for the AI.

 

[sveint]

12/21 build - I am winning a game on Emperor. Something seems too easy. The only AI to keep a sufficient army in my game was Ragnar - the rest were pushovers and almost asked to be invaded (random enemies gave me no warmongers, the only AI war was Ragnar taking a single city from Spain).

Please have even peaceful AIs put more defenders in cities with "culture wars". These cities are vulnerable to surprise assaults.

Please have ALL civs keep one central reserve (just a stack of even a couple of attack units would make a HUGE difference - they could for example keep it in their capital and deploy it when war is declared).

Please slightly reduce the portion of catapults/artillery.

I was navally invaded - and lost a city. But naval invasions are tricky and needs MASSIVE force to succeed. Furthermore you don't need that much catapults/artilley since your ships can bombard. And if you want to keep anything, you need a couple of defensive units. The AI landing in quest was 6 units - 1 mounted, 1 grenadier, 4 catapults.

 

[vilemerchant]

12/21 build - I am winning a game on Emperor. Something seems too easy. The only AI to keep a sufficient army in my game was Ragnar - the rest were pushovers and almost asked to be invaded (random enemies gave me no warmongers, the only AI war was Ragnar taking a single city from Spain).

I'm not yet using the BetterAI mod, but i've had this problem of enemies not fighting proper wars ever since I upgraded to 2.08. I'm playing a Monarch game right now and once again it's just a case of marching my stacks thru their land unopposed, bombarding and taking their cities from the 2 or 3 defenders. As it seems the BetterAI is not going to fix it I think i'll uninstall back to V2.0 and play on Emperor.