1.29f AI Observations

[dac]

I was playing my first game of the 1.29f patch last night when I saw something that led me to believe that the AI might have changed a bit from pre-1.29f days.

The treacherous Japanese landed a spearman and an archer on my shores and declared war. They had landed next to a city defended by a single spearman, as were most of the cities on that chunk of the contintent. I often space my cities 3 apart in the early game, so that I can shuttle defenders around in one turn as necessary. So, the nearest city 'donated' it's spearman to the threatened city. At this point, usually the AIs obsession with attacking the weakest city kicks in and it stalks off toward the undefended city. This time, however, it pressed home it's attack on the city it had landed next to, even though it was the best defended city around for at least 10 squares. Never seen that before.

Is the AI exhibiting new behavior? Or have my observations of watching AI stacks 'chase' undefended cities around my empire been anomalous?

I of course, was able to 'exploit' this behavior later, when Aztec horsies impaled themselves on my size 3 town, on a hill, with a wall, staffed by two newly upgraded veteran pikemen. Well, they weren't veterans for long.

[Catt]

I think that the AI is improved in many ways (particularly warfare). Some specific examples (although any one or all of the below could be peculiar to my games):

(1) Better sea-borne invasions. I'm seeing more dangerous and "complete" invasions consisting of multiple units - not seeing as many "lone" longbowman invasions. I am also very happy to see that knocking a single HP off of a transport vessel doesn't result in the transport heading home for repairs -- in many instances now I am seeing transports with only one or two HPs coming ashore to drop of its forces before retreating. Finally, although it is probably a fluke, in my latest game I watched as an AI transport and its escort waited 3 tiles off-shore (I had no naval units in the area). 2 turns later it was joined by a second transport and escort, and only then did both transposts come ashore and unload -- this was in the late industrial age and the AI dropped off a very impressive (for the AI) force of 6 infantry, 5 tanks, 5 cavalries!

(2) More "offensive" units. This was actually a bug fix rather than an upgrade, but the AI civs are now building a lot more "offensive units" - knights / cavalry / tanks -- may be the end of the 65-riflemen stack of doom.

(3) More swarming tactics. I have been seeing a lot less of the "stack of doom" in general. Under 1.21f, I routinely saw the AI attacking one objective with (seemingly) it's entire army. Under 1.29f, I am seeing decnt-sized smaller forces going after multiple objectives. In my most recent game (1.29f, Emperor, all random), I faced multiple ancient and early middle age wars of aggression by the Zulu. At any given time I had 3 or 4 different Zulu attack forces going after different objectives - in this case, a Zulu attack force consisting of 2 - 3 impis, often with archers and later with swordsmen (didn;t see many horsemen, though). It really spread my defenses very thin. And this wasn't simply because the attack forces were coming from different directions outside of my empire - several times I saw a concentrated force enter my territory on advantageous defensive terrain, and then split up into separate forces with differing targets.

It's too early for me to say definitively (I could be getting lucky!), but I've seen some subtle changes in AI warfare which really have made the challenege more enjoyable.

Catt

[dac]

Catt,
Hmmm, interesting. That would be more fun. I've staved off many an invasion in previous games with only artillery firing at invading transports. I use leftover ironclads as pickets, they spot for the artillery, I scratch the paint on the transport, and it retreats. It'd be wonderful if that wouldn't work anymore.

You spoke of the AI attacking multiple targets. I have seen some of that, with some of the AI units bypassing my "hardpoint" city to attack the one behind it. Unfortunately for them, they had to bypass a mountain with lots of swordsmen on it. Very few units made it to that second city. In that same war, the Aztecs convinced the French to join in the attack. Interestingly enough, the French did not attack my empire directly, they marched at least two extra turns through Aztec land to attack the same city that the Aztecs were. I was sweating for a while there. I would love it if the AI could actually coordinate attacks between two different civilizations. I would love it if the AI could coordinate attacks from me. (I eventually convinced the French to backstab the Aztecs, that was fun, sadly though, the French are very big now.)

[candidgamera]

Originally posted by Catt
I think that the AI is improved in many ways (particularly warfare). Some specific examples (although any one or all of the below could be peculiar to my games):

[
(2) More "offensive" units. This was actually a bug fix rather than an upgrade, but the AI civs are now building a lot more "offensive units" - knights / cavalry / tanks -- may be the end of the 65-riflemen stack of doom.

Catt

Any impression that in general the ai is now not building as many units in total?

[Catt]

Originally posted by candidgamera


Any impression that in general the ai is now not building as many units in total?

I don't have that impression, but I also don't have enough data to compare -- haven't really tried to keep track of total units in prior games.

[DeepO]

Hmmm.. I wish I could say the AI has improved drastically, but so far I haven't seen many improvements in 1.29f. Minor tweaks, yes (like fixing the 'only defense units, used on offense' rule, which led to reports of 'paras more favored then tanks'), but other... There is no way of telling that 'the spearman+ archer next to one city' example was an actual improvement, I've seen it happen in 1.21f games as well. And my experience with Alamo suggests that it is simply a matter of having those 2 units available in the harbor, so that when a passing galley loads one, it also loads the other.

No, there are very few improvements to the naval AI (the 1 HP down on transport --> return home still happens, BTW). There could be a lot more, as I posted a whole list of them in the Alamo bug thread . I hope these get taken into aco**** when PtW comes around.

Catt, more swarming is indeed a new tactic, but the stupid thing was that in Alamo I was hoping to see a SoD, the situation called for it. As a human in the AI position, you'd build a SoD and would be unstopable... what is in now surely is not the best (as in winning, not as in nice eyecandy) tactic around.

But don't let these negative notes fool you, I love the AI in Civ III. It's just that there are a few improvements possible... but aren't there always be a few improvements possible in AI

DeepO

[Catt]

Originally posted by DeepO
No, there are very few improvements to the naval AI (the 1 HP down on transport --> return home still happens, BTW). There could be a lot more, as I posted a whole list of them in the Alamo bug thread . I hope these get taken into aco**** when PtW comes around.

Catt, more swarming is indeed a new tactic, but the stupid thing was that in Alamo I was hoping to see a SoD, the situation called for it. As a human in the AI position, you'd build a SoD and would be unstopable... what is in now surely is not the best (as in winning, not as in nice eyecandy) tactic around.

I saw the Alamo Bugs thread with the transports turning home - I'm just seeing in my games that the transports now tend to drop of their units before turning tail - don't know why but it is a much better state of affairs - under 1.21f, bombarding a transport to 3HPs from 4HPs ("scratching the paint," as dac said )was always sufficient to send it home in my games.

My laptop is simply too slow to handle playing Alamo - I'm surprised the AI didn't send SoDs (since I understood from the posts that their were two very small chokepoints that enabled a land invasion).

Catt

[Jon Shafer]

I asked Soren directly about AI changes, and all he told me was that he changed the AI offensive/defensive 'bug'. He told me that he didn't really change anything else... Take that for what it's worth, but who knows.

[Theseus]

Vel set it up so they couldn't... he only gave them roads, and distributed their forces such that they couldn;t all show up at once. They obviously could have waited, but they each had so many troops that the algorithm clearly instructed attacks in waves (I think the testers took a while to get that right).

I can;t recall having really seen any SoDs lately, but I haven;t recently played a game where they would have arisen anyway.

[Catt]

Originally posted by Trip
I asked Soren directly about AI changes, and all he told me was that he changed the AI offensive/defensive 'bug'. He told me that he didn't really change anything else... Take that for what it's worth, but who knows.

Yeah - but I think Soren doesn't go into all the details in the chatroom environment (assuming you're talking about the chat ). In the chat, he started by saying he made a number of changes to the AI - although everyone at Firaxis clearly acknowledged that the focus of the patch was the editor - and in listing some of the changes, he immediately got dog-piled on the offensive/defensive question and the tech valauation changes discussion. He also mentioned that he tweaks the AI from time to time and can't always remember which tweaks make it into which patch. Again, if you're referring to the 1.29f chat, he didn't answer you by saying he "only" dealt with the offensive / defensive bug, he just pointed out the most prominent change.

In any event, any little tweak in the AI (offensive / defensive change) could affect gameplay in a number of ways. Any tweak which improves the AI (even if accidentlally ), gets a from me.

Catt

EDIT - Soren also said that he actually made a "bunch of other improvements too" although never got around to listing any combat improvements.

[DeepO]

Regarding the SoDs in Alamo: Vel set it up that it happened with the transports, and as a tester I did ask for that. But on the land, it was a different game. The problem was that even if there were several huge attackwaves for the Russians (One of the land-enemies), there was no way of telling the AI to regard those stacks as SoDs. So some of the troops went towards the chokepoint, some went on defense in the Russian cities, and some loaded in transports... I'm suspecting that the whole SoD thing was an artifact of better spreading of attack troops over the enemy cities, so the turn after declaration of war they all focused because of the RRs. That would be highly ironic, as many people have claimed to like these SoD (not nice, but a pain in the a**)... if these indeed were removed in 1.29f, I wouldn't like it one bit.

BTW, what was more a problem in Alamo was that even if the AI had a lot of troops, from a certain point on they were discouraged by the numbers of defenders. Instead of giving it all they had (which would have certainly broken through), they decided to wait for an army to show up. But, once the army was there, all te other troops had gotten already another use, most of the times transports got build and loaded, and consequently sunk.
I loved the Alamo scenario, with all the testing I must have been playing it for nearly a month, but the only reason it was engaging was that it simply was such a huge underdog position that the AI couldn't lose, whatever they did. That's also why I saw so many things on the AI there, not only do you replay certain games a few times, but you actively spy on them to see if the placement was good, and you have so many troops that the AI shows you most of its tricks. I mean, the sub-Aegis-stealths actions were impressing... but you never ever see those in normal games. The transports however: ouch.

DeepO

[edit: I misspelled my name ]

[candidgamera]

Originally posted by Catt


I don't have that impression, but I also don't have enough data to compare -- haven't really tried to keep track of total units in prior games.

reason asked: took Willem's suggestion to increase unit maintenance costs and have them at:

anarchy: 1
despotism: 2
monarchy&communism: 3
republic: 4
democracy: 6

plus some other free units depending on gov type

I'm strong verses all my 4 opponents with the new "power" rules
and the game is moving noticeably faster, don't see as many ai units period.

number of ai units seems to be the big factor in drag have concluded - just don't play with more than 5 total civs on big maps.

maybe am also seeing the new faster 1.29 way off handling out of sight ai units as well.

coming up on espionage so will soon find out what everybody has.

Probably killing the ai with the upped maintance costs though.

[Thrawn05]

I did find combat with the AI a lot more unpredicatable and fun with this patch.

When I was at war prior to 1.29f, I'd make a line of units about 6 tiles long. I would then just press on and kill anything in my path, since the AI tends to head right for it. All the while I had a few scattered units acting as horns and taking the cities behind the units. If you can't visualize, here's a rough diagram:

+++
\
\ O
\
\ O
+++

The \ is the line (slow defensive units such as pikeman) in this case moving east, the + are my scattered units (fast units suchs as Knights) in this case moveing towards the south and north. THe O is enem cities. For those who know their history, this is what Shaka used basicly.


That has changed now with the patch, the AI doesn't always go from my line of units, instead, goes for those scatered units (I guess because my scattered units had a low defensive value). My line units still were able to take the cities, but it's a lot bloodier.

EDIT: It's good to be the KING!

[NeoStar]

Originally posted by Catt

(1) Better sea-borne invasions.

If this is true ...

[fittstim]

Re: 1.29f AI Observations

Originally posted by dac
The treacherous Japanese landed a spearman and an archer on my shores and declared war. They had landed next to a city defended by a single spearman, as were most of the cities on that chunk of the contintent. I often space my cities 3 apart in the early game, so that I can shuttle defenders around in one turn as necessary. So, the nearest city 'donated' it's spearman to the threatened city. At this point, usually the AIs obsession with attacking the weakest city kicks in and it stalks off toward the undefended city. This time, however, it pressed home it's attack on the city it had landed next to, even though it was the best defended city around for at least 10 squares. Never seen that before.

My first impression after reading this is that it was a fine example of an "AI cheat". I mean, how could the Japanese know that there was an undefended city which had "donated" it's defensive troop to the threatened city?

Human players can only see the strongest defensive unit in a city they are attacking. The range of sight for a spearman or archer shouldn't be to identify an undefended city 3 tiles away from the city they were attacking- even if they were placed on a mountain.

Seems rather odd to me...