Post 0.41h AI feedback needed
- Thread starter Kael
- Start date
I originally posted this in the bug thread but it seems this might be more appropriate place for my observation.
I'm playing the Splintered Courts scenario. It's turn 150 now and my Svaltarfar allies are only building workers and warriors (which they have heaps of by now). Because of the permanent alliances I can look inside their cities and see they have no buildings except the initial ones. Luckily the Ljosalfar enemies are most likely doing the same.
I think the army building logic in the ne AI should be heavily tuned.
I'm playing the Splintered Courts scenario. It's turn 150 now and my Svaltarfar allies are only building workers and warriors (which they have heaps of by now). Because of the permanent alliances I can look inside their cities and see they have no buildings except the initial ones. Luckily the Ljosalfar enemies are most likely doing the same.
I think the army building logic in the ne AI should be heavily tuned.
NeoBasilisk
Chieftain
- Joined
- Jan 7, 2009
- Messages
- 20
Well, now that the AI builds so many more units, I get CtD's that much earlier. I guess I can only play small Erebus maps now if I still want a lot of civs. 
In this save, near the Doviello city of Heimseter, there have been 3 captured Bannor wolfpacks that Doviello will not just slide units out of their city, kill, and slide back in.
Near the Elohim city of Balinasloe, there is a huge stack, 69 axemen, and 23 archers. They have not moved for about 5 turns now, and are just watching the Bannor roll through their empire and take their cities.
Near the Elohim city of Balinasloe, there is a huge stack, 69 axemen, and 23 archers. They have not moved for about 5 turns now, and are just watching the Bannor roll through their empire and take their cities.
Onionsoilder
Reaver
- Joined
- Mar 19, 2007
- Messages
- 3,173
Well, I just won my very first deity game ever. I've mentioned it before in the thread, but the new AI is not beneficial; sure they make more units, but they use those new units less effectively then they did in pacth H. Throughout most of the game, the AI had at leats 5x as many units as I did, but they never did anything with those units so I could slowly pick them off, then take their cities once their stacks were dead.
I can assure you I would have not been able to do that in patch H. With J, the AI attacked me twice in the entire game. Both times I was able to neutralize the threat using a simple trick (I locked down one with Slow; the other one I played a game of 'Tag' with because they would keep doubling back to recapture cities I took from them after they left them undefended.)
The new spellcasting and economic AI is great, but the military techniques are much, much worse than they were before.
I can assure you I would have not been able to do that in patch H. With J, the AI attacked me twice in the entire game. Both times I was able to neutralize the threat using a simple trick (I locked down one with Slow; the other one I played a game of 'Tag' with because they would keep doubling back to recapture cities I took from them after they left them undefended.)
The new spellcasting and economic AI is great, but the military techniques are much, much worse than they were before.
Having tested the new AI over a few more games, I can provide a more extensive feedback:
major problems:
1. The AI is too defensive. It moves stacks back and fourth that would wipe you out. When stronger and at war, it should press its advantage! Currently it "waits" for something that might never happen until it is to late.
2. The AI seems to neglect higher tier units and prefers to spam basic units - but that does not work out, because it does not attempt to overrun its enemies (see above).
minor problems:
1. The AI attacks across rivers, even when there is no need to.
2. The AI sends its units in mass suicides against summend unit, that would disappear next turn anyway.
3. Somtimes units wander around aimlessly and alone.
4. Wounded units leave the stack and run further into enemy territory. They should stay or retreat.
5. It picks up Pieces of Barnaxas but does not rebuild.
6. It uses magic but is not prone to produce magic wielder.
7. The AI has become pacifistic. It seldom declares war.
8. It often demands tribute, but if the player refuses, see 7. - if you blackmail someone, mean it!
However, it is economic stable!
major problems:
1. The AI is too defensive. It moves stacks back and fourth that would wipe you out. When stronger and at war, it should press its advantage! Currently it "waits" for something that might never happen until it is to late.
2. The AI seems to neglect higher tier units and prefers to spam basic units - but that does not work out, because it does not attempt to overrun its enemies (see above).
minor problems:
1. The AI attacks across rivers, even when there is no need to.
2. The AI sends its units in mass suicides against summend unit, that would disappear next turn anyway.
3. Somtimes units wander around aimlessly and alone.
4. Wounded units leave the stack and run further into enemy territory. They should stay or retreat.
5. It picks up Pieces of Barnaxas but does not rebuild.
6. It uses magic but is not prone to produce magic wielder.
7. The AI has become pacifistic. It seldom declares war.
8. It often demands tribute, but if the player refuses, see 7. - if you blackmail someone, mean it!
However, it is economic stable!
slowcar
King
I reinstalled civ today on my new system and played a game on king (just to get a feeling).
- while the AI built many troops (not able to rush by far) it just did nothing. i stole the first worker and fortified a single warrior on a wooded hill. they took my warrior on turn 90, with a stack of 15 units. they didn't come after me, even though i was happily settling away and had only 3-4 warriors in my capital (and some nearby, just in case).
they stayed with a single city and strangled their research with their many troops i guess, i picked them up with bronze swordsmen later on, a total of 3 units entered my territory, all single.
total game was very short and never challenging, in the end i won with 2400 points, strongest enemy before the end turns had 700 i think.
edit: oh, and the ljosalfar used their worldspell when i attacked, stacked some 20 treants (+15 warriors + silveric) in front of one city i captured with loki beforehand, didn't attack though. it was "defended" by two warriors.
the AI really needs aggression! get blake on the phone, he made some great aggressive AI patches back then
- while the AI built many troops (not able to rush by far) it just did nothing. i stole the first worker and fortified a single warrior on a wooded hill. they took my warrior on turn 90, with a stack of 15 units. they didn't come after me, even though i was happily settling away and had only 3-4 warriors in my capital (and some nearby, just in case).
they stayed with a single city and strangled their research with their many troops i guess, i picked them up with bronze swordsmen later on, a total of 3 units entered my territory, all single.
total game was very short and never challenging, in the end i won with 2400 points, strongest enemy before the end turns had 700 i think.
edit: oh, and the ljosalfar used their worldspell when i attacked, stacked some 20 treants (+15 warriors + silveric) in front of one city i captured with loki beforehand, didn't attack though. it was "defended" by two warriors.
the AI really needs aggression! get blake on the phone, he made some great aggressive AI patches back then
Onionsoilder
Reaver
- Joined
- Mar 19, 2007
- Messages
- 3,173
Having tested the new AI over a few more games, I can provide a more extensive feedback:
major problems:
1. The AI is too defensive. It moves stacks back and fourth that would wipe you out. When stronger and at war, it should press its advantage! Currently it "waits" for something that might never happen until it is to late.
2. The AI seems to neglect higher tier units and prefers to spam basic units - but that does not work out, because it does not attempt to overrun its enemies (see above).
minor problems:
1. The AI attacks across rivers, even when there is no need to.
2. The AI sends its units in mass suicides against summend unit, that would disappear next turn anyway.
3. Somtimes units wander around aimlessly and alone.
4. Wounded units leave the stack and run further into enemy territory. They should stay or retreat.
5. It picks up Pieces of Barnaxas but does not rebuild.
6. It uses magic but is not prone to produce magic wielder.
7. The AI has become pacifistic. It seldom declares war.
8. It often demands tribute, but if the player refuses, see 7. - if you blackmail someone, mean it!
However, it is economic stable!
I agree with pretty much every one of those. With the AI upgrade, they have become much better at economics and slightly better at spells, but the military quality has degraded. I would also like to add in:
1) The AI likes to build large amount of one unit, and no others. Archers seem very common, I see them much more often than axemen, which is weird. They also use these archers for offensive strikes, which is weirder.
2) Even when the AI is at war, they seldom send troops to attack. In my recent game Basium had well over 100 units and a huge power rating, but he never attacked. Ever.
3) The AI does not leave units behind in cities after it captures them. This means you can waltz back in after they AI leaves.
The new Barb-AI is too much for the AI-controlled Civs. I like to play larger Maps with lots of unexplored Space so you normally get the 1st contact not before 100-150.
With the newest patches applied, most Civs have no real chance to properly build an empire. The 2 screenshots show how nearly all non-barbarian civs suffer (and look to the upper left, the Hippus lost all cities except of their capital)
This is no fun anymore if all Rivals are that much under pressure.
And this is Deity-Difficulty, so if the AI has so many problems even with all the extra Units and Techs at the start, how can anyone play games at noble?
My suggestion would be to give the randomly spawning barbs the suicide-AI back and only apply the new one to Barbs built in barbarian cities.
With the newest patches applied, most Civs have no real chance to properly build an empire. The 2 screenshots show how nearly all non-barbarian civs suffer (and look to the upper left, the Hippus lost all cities except of their capital)
This is no fun anymore if all Rivals are that much under pressure.
And this is Deity-Difficulty, so if the AI has so many problems even with all the extra Units and Techs at the start, how can anyone play games at noble?
My suggestion would be to give the randomly spawning barbs the suicide-AI back and only apply the new one to Barbs built in barbarian cities.
Spoiler :
The new Barb-AI is too much for the AI-controlled Civs. I like to play larger Maps with lots of unexplored Space so you normally get the 1st contact not before 100-150.
With the newest patches applied, most Civs have no real chance to properly build an empire. The 2 screenshots show how nearly all non-barbarian civs suffer (and look to the upper left, the Hippus lost all cities except of their capital)
This is no fun anymore if all Rivals are that much under pressure.
And this is Deity-Difficulty, so if the AI has so many problems even with all the extra Units and Techs at the start, how can anyone play games at noble?
My suggestion would be to give the randomly spawning barbs the suicide-AI back and only apply the new one to Barbs built in barbarian cities.
Spoiler :
The barbs are the best part of the new AI. I have played quite a few games at Monarch, and no AI's have been crippled.
Emptiness
[]
- Joined
- Jan 7, 2009
- Messages
- 1,922
(My video card died in the middle of my first attempt to post this, so fingers crossed for take two.)
First, I'd just like to say that I'm very pleased to see the AI receiving some much-needed attention. These changes definitely represent a step in the right direction, and I think that once everything is working and balanced again FfH2 will be a much more enjoyable game as a result. My thanks to everyone who is working hard to make these improvements, and helping to iron out the problems.
-------------
The Good
The AI actually uses spells, and does so fairly effectively. This is a massive improvement over pre-H. The AI uses Rust to remove enemy metal weapon promotions, Fireballs enemies on attack and defense, even uses Revelation to remove HN and invisibiliity from enemy units.
AI economy and tech choices are improved across the board. This AI does not flounder in the way that the pre-H AI sometimes would (stuck with no economic techs trying to research an expensive tech that won't help them even after they never get it). Tech pace is much more competitive, and AI military is much stronger by virtue of higher tech level and ability to support more units.
The AI is wiling to research religions, even when following another religion. The AI will even switch to a more preferable religion once it becomes available. (For example, in the attached Perpentach game, the Luchuirp converted to OO when it spread to one of their cities, then to FoL when it spread to more cities, but then switched to Order once they founded it.) The AI also knows when not to change religions, such as when doing so would mean the loss of a hero. (In another game, Arendel Phaedra of the Ljosalfar founded and switched to OO before researching FoL - probably because of the coastal location of her capitol. When she later founded FoL she remained in OO, and so did not lose her well-promoted Saverous.)
The AI researches secondary religion techs (such as Arete), and builds secondary religion tech heroes and wonders (including Mines of Gal-Dur), better than the pre-H AI.
The early game is much more interesting, thanks to changes in the behavior of the barbarians. Defending one's empire requires a much more active role. Several others have complained about this change, and the challenges it produces, and so perhaps some fine-tuning is in order. Overall, however, I consider this a big improvement.
-------------
The Bad
The new AI is more reluctant to risk combat. This has had a tremendous impact on the game, from barbarian behavior to combat tactics to even the power level of various techs. The AI builds massive armies, but does not seem to take numbers into account when determining whether it can win a battle. If the odds of victory for the best attacker in the stack are too low, then entire stack won't attack - even in cases where there is only a single defender and the attacker has dozens of units. Most humans (and the pre-H AI) would recognize that a stack of units might suffer heavy losses attacking a strong defender, but would wear down and kill it before all attacks were exhausted. The post-H AI builds massive armies but won't risk them until victory for the best unit is reasonably likely. The result is that when the AI can steamroll an opponent they do, dramatically. When the AI merely has numerical superiority, however, they declare war and then mill about waiting for the combat odds to somehow improve - rather than using their numbers to crush the opponent, as the pre-H AI would have done. Some sort of medium needs to be found; it is good that the AI does not throw away units, but it needs to have some understanding of the power of numbers, and how that affects the odds of its stack capturing an objective. Also, if the AI is at war and has decided that its stack cannot take an objective then it needs to move on to another objective, or at least pillage improvements while it waits. Trashing enemy improvements is a better combat strategy than just hanging out near an enemy city.
The AI does not immediately spend its promotions. This can sometimes be a useful tactic for a human, but the AI lacks the sophistication necessary to determine when it can get away with not promoting a unit and when it is vital to ensure that a unit has chosen every possible promotion (for the sake of its immediate survival). Humans are sneaky opponents, and there are a lot of ways to fool the AI so that it doesn't understand that a unit is in danger (invisibility, haste, unspent promotions suddenly spent on mobility). Unfortunately, the post-H AI's willingness to save promotions isn't even limited to "safe" units. It is willing to leave units in immediate danger without purchasing all the promotions available to the unit. This is a significant reduction of the AI's combat potential, depending on how many units on which the AI chooses to save promotions. Sometimes the AI seems to always spend promotions immediately, and in other cases the AI saves promotions on many units (examples of this can be seen in the Perpentach saved games I've attached to this post, in which the Ljosalfar have quite a few units they never seem to promote.)
The AI's promotion choices are haphazard and mismanaged. The worst example of this is casters without actual spell promotions, either because of unspent promotions or because they have purchased all non-spell promotions. A caster that can't cast any spells isn't worthless (any unit can be fodder) but is not worth the production investment made in choosing to produce it. The AI should be made to always spend at least the first promotion it buys for a caster to gain a spell. This problem isn't limited to casters, however. I've seen many instances of AI combat units with both Shock I and Formation I (for example). This is disadvantageous for the AI. Taking two levels of Combat would also provide +40% vs Melee and +40% vs Mounted, but would also provide +40% vs every other type of unit, as well as potentially unlocking other useful promotions (March, Magic Resistance, Guardsman, et c.) Only on a unit with Combat IV+ would mixing Shock and Formation potentially be viable, and even then it would probaby be better to have one unit with Shock I & II and one unit with Formation I & II than to have two units each with Shock I and Formation I. If the pre-H AI had these problems I had not witnessed them. Another, more specific, instance of this problem is Barnaxus having been promoted with City Raider I - III, instead of Combat promotions. I fail to grasp any situation in which Barnaxus should choose any promotion other than Combat promotions until he has all five. Although there is certainly room for different promotion philosophies, and I can accept that the AI won't be using my philosophy, this seems like a situation where common sense should produce the same decision in every viable philosophy.
The AI makes terrible decisions about whether to stop moving a unit to heal it, when to do it, and where. Injured units not threatened and in friendly territory continue to move around, blocking healing, but lone units in enemy territory adjacent to powerful enemies remain stationary to heal rather than attempting to withdraw toward safety.
The AI's economy may be too strong now. It is able to support massive armies and still advance technologically at a rapid pace. (For example, in one Marathon game on Monarch difficulty I popped two early techs from huts, but at turn 51 (about like turn 17 on normal speed) five of the seven AIs were still ahead of me in points.) The AI was alway one for large armies, but the post-H AI armies dwarf pre-H AI armies. The result is slowdown of game turns, happening earlier and becoming more severe than previously. The effect is exacerbated in peaceful games, because the AI just keeps building more units. If one city can support 30 units then imagine the army a 20-city AI can field. Naturally the impact of this will vary based on computer and also play preferences. A duel game with 4 AIs doesn't have slowdowns on my computer, but my preferred play setup pre-H (Marathon, Standard, 8 AIs) is much slower. The only thing that makes it possible to cope with these large armies is the AIs mismanagement of units, mishandling of wars, and tactical errors on the battlefield. If the post-H economy were joined to the pre-H warfare then I'd have to drop the difficulty setting 3 or 4 levels to compete. I don't mind losing to superior combat tactics and a well-managed army, but I don't want to be beaten by an AI that just drowns me in units. Several posters have called for a reevaluation of the production and unit support modifiers the AI receives. Right now the AI does not handle warfare well enough to warrant a change, but hopefully the military AI will be improved and then I definitely agree that some adjustments to reduce unit bloat will be necessary. This should also improve turn speeds, and hopefully allow some of the performance gain I was hoping to see from CAR's incorporation to shine through.
Privateers immediately declare nationalilty. I'm not sure if this is alway true, or if it's some sort of response to the presence of stronger enemy vessels. In any event, declaring nationality on a Privateer throws away much of the purpose of building it to begin with. If the AI is in a situation where it feels the need to immediately declare nationality on Privateers it builds then it should instead build another type of ship. Privateers are a dead-end of the upgrade path, so if they are already so outdated that hidden nationality represents too great a risk, then building a ship with upgrade potential would make a better production investment.
Goblin Forts no longer accumulate defenders. Previously, barbarians were not allowed to move away from a Goblin Fort if it had less than three defenders. This made Goblin Forts challenging targets, because by the time you could build enough Warriors to kill the Archer, a Scout had spawned. Often it would be necessary to wait until three Scouts had spawned, and the Archer had moved off, before taking a Fort would be practical. Spawned Scouts now move off immediately, and because a lone defender is very vulnerable this makes it possible to tackle Goblin Forts much sooner. I consider this unfortunate, because instead of being some of the last bastions of barbarian resistance, Goblin Forts are now not much different from other lairs. This also has implications for the Clan of Embers, because easier elimination of Goblin Forts means that the opportunity to use them to raise armies near an opponent is decreased.
AI spell casting logic doesn't seem to take friendly fire into account, or else it is too liberal in its willingness to accept friendly fire. For example, the AI now uses Maelstrom but is willing to cast it in situation where many of its own units are damaged. This may be very difficult to address, unless there is a way to coordinate casting decisions with the movement timing of other units (which I suspect there is not). At least, the AI should choose not to cast AoE damage spells when the total number of friendly units that will be damaged is greater than the number of enemy units. (If possible, damage resistances should be taken into account, so that if many friendly units will be hit but will take little damage then the AI should be more willing to cast.)
Stacks of casters do not distribute their casting choices. For example, a stack of mages with access to Rust and Fireball all chose to cast Fireball against my attackers, and so my Weapon promotions remained. If one or two of the enemy casters had cast Rust instead the reduction of power of my stack would have been greater. Perhaps there needs to be a chance that a caster will opt to cast a "lesser" spell even when a "greater" one is available. Alternatively there could be an ordered list to the spells, such that each spell is checked in order and skipped only if casting it would not be helpful. Were such as system implemented, and Rust properly listed before Fireball, then the first Mage or two would have cast Rust, and then when most or all of my units had been stripped of metal weapons (and debuffed via the Rust promotion) the rest would have cast Fireball.
The AI does not defend newly conquered cities properly. Its attack force moves on immediately (without even stopping in the city to heal), and the city is almost always left undefended. The AI will often build a defender in another city and send it to the new one, but there is usually a window of several turns in which the city can be retaken by the original owner (or someone else) without any combat. Even when defenders arrive they are often inadequate to the task (ie not enough of them). It seems that it would be better for the attackers to remain defending the city until it has adequate defenders of its own before moving on to a new target. Perhaps the AI should include city defense units in its attack stacks, which can then remain in a conquered city to defend it. Alternatively, the city could just be razed, if the AI does not have the units available to defend it properly.
I encountered one situation where the AI actually abandoned a city I was assaulting. Most of the city's defenders had been killed, and the few remaining units were unlikely to survive the next round of combat. Perhaps this was a savy decision on the AIs part, to salvage the production value of the defenders rather than lose them in the hopeless defense of a doomed city of little value (it was a size 1 or 2 city in an icy area). At the same time, just handing cities to the enemy is generally a bad idea. I'm not sure this is a problem, since I've only seen it once, but it was odd (and in my previous experience unprecedented) and I'm still trying to recreate the situation.
-------------
Attached below are .7z archives of two games I've played to try to recreate problem situations I've seen occur. The games were played using Patch J.
Game 1: Duel Pangea, Monarch difficulty, low sea level, random climate, 4 AI, normal speed, allow permanent alliances, no tech brokering, all victories enabled. Randomly assigned Perpentach of the Balseraphs.
Game 2: Duel Pangea, Monarch difficulty, low sea level, random climate, 4 AI, normal speed, allow permanent alliances, no tech brokering, all victories enabled. Randomly assigned Os-Gabella of the Sheaim.
There are still some problems I've seen for which I'd like to provide examples. If I can replicate those situations I'll post additional example games.
First, I'd just like to say that I'm very pleased to see the AI receiving some much-needed attention. These changes definitely represent a step in the right direction, and I think that once everything is working and balanced again FfH2 will be a much more enjoyable game as a result. My thanks to everyone who is working hard to make these improvements, and helping to iron out the problems.
-------------
The Good
The AI actually uses spells, and does so fairly effectively. This is a massive improvement over pre-H. The AI uses Rust to remove enemy metal weapon promotions, Fireballs enemies on attack and defense, even uses Revelation to remove HN and invisibiliity from enemy units.
AI economy and tech choices are improved across the board. This AI does not flounder in the way that the pre-H AI sometimes would (stuck with no economic techs trying to research an expensive tech that won't help them even after they never get it). Tech pace is much more competitive, and AI military is much stronger by virtue of higher tech level and ability to support more units.
The AI is wiling to research religions, even when following another religion. The AI will even switch to a more preferable religion once it becomes available. (For example, in the attached Perpentach game, the Luchuirp converted to OO when it spread to one of their cities, then to FoL when it spread to more cities, but then switched to Order once they founded it.) The AI also knows when not to change religions, such as when doing so would mean the loss of a hero. (In another game, Arendel Phaedra of the Ljosalfar founded and switched to OO before researching FoL - probably because of the coastal location of her capitol. When she later founded FoL she remained in OO, and so did not lose her well-promoted Saverous.)
The AI researches secondary religion techs (such as Arete), and builds secondary religion tech heroes and wonders (including Mines of Gal-Dur), better than the pre-H AI.
The early game is much more interesting, thanks to changes in the behavior of the barbarians. Defending one's empire requires a much more active role. Several others have complained about this change, and the challenges it produces, and so perhaps some fine-tuning is in order. Overall, however, I consider this a big improvement.
-------------
The Bad
The new AI is more reluctant to risk combat. This has had a tremendous impact on the game, from barbarian behavior to combat tactics to even the power level of various techs. The AI builds massive armies, but does not seem to take numbers into account when determining whether it can win a battle. If the odds of victory for the best attacker in the stack are too low, then entire stack won't attack - even in cases where there is only a single defender and the attacker has dozens of units. Most humans (and the pre-H AI) would recognize that a stack of units might suffer heavy losses attacking a strong defender, but would wear down and kill it before all attacks were exhausted. The post-H AI builds massive armies but won't risk them until victory for the best unit is reasonably likely. The result is that when the AI can steamroll an opponent they do, dramatically. When the AI merely has numerical superiority, however, they declare war and then mill about waiting for the combat odds to somehow improve - rather than using their numbers to crush the opponent, as the pre-H AI would have done. Some sort of medium needs to be found; it is good that the AI does not throw away units, but it needs to have some understanding of the power of numbers, and how that affects the odds of its stack capturing an objective. Also, if the AI is at war and has decided that its stack cannot take an objective then it needs to move on to another objective, or at least pillage improvements while it waits. Trashing enemy improvements is a better combat strategy than just hanging out near an enemy city.
The AI does not immediately spend its promotions. This can sometimes be a useful tactic for a human, but the AI lacks the sophistication necessary to determine when it can get away with not promoting a unit and when it is vital to ensure that a unit has chosen every possible promotion (for the sake of its immediate survival). Humans are sneaky opponents, and there are a lot of ways to fool the AI so that it doesn't understand that a unit is in danger (invisibility, haste, unspent promotions suddenly spent on mobility). Unfortunately, the post-H AI's willingness to save promotions isn't even limited to "safe" units. It is willing to leave units in immediate danger without purchasing all the promotions available to the unit. This is a significant reduction of the AI's combat potential, depending on how many units on which the AI chooses to save promotions. Sometimes the AI seems to always spend promotions immediately, and in other cases the AI saves promotions on many units (examples of this can be seen in the Perpentach saved games I've attached to this post, in which the Ljosalfar have quite a few units they never seem to promote.)
The AI's promotion choices are haphazard and mismanaged. The worst example of this is casters without actual spell promotions, either because of unspent promotions or because they have purchased all non-spell promotions. A caster that can't cast any spells isn't worthless (any unit can be fodder) but is not worth the production investment made in choosing to produce it. The AI should be made to always spend at least the first promotion it buys for a caster to gain a spell. This problem isn't limited to casters, however. I've seen many instances of AI combat units with both Shock I and Formation I (for example). This is disadvantageous for the AI. Taking two levels of Combat would also provide +40% vs Melee and +40% vs Mounted, but would also provide +40% vs every other type of unit, as well as potentially unlocking other useful promotions (March, Magic Resistance, Guardsman, et c.) Only on a unit with Combat IV+ would mixing Shock and Formation potentially be viable, and even then it would probaby be better to have one unit with Shock I & II and one unit with Formation I & II than to have two units each with Shock I and Formation I. If the pre-H AI had these problems I had not witnessed them. Another, more specific, instance of this problem is Barnaxus having been promoted with City Raider I - III, instead of Combat promotions. I fail to grasp any situation in which Barnaxus should choose any promotion other than Combat promotions until he has all five. Although there is certainly room for different promotion philosophies, and I can accept that the AI won't be using my philosophy, this seems like a situation where common sense should produce the same decision in every viable philosophy.
The AI makes terrible decisions about whether to stop moving a unit to heal it, when to do it, and where. Injured units not threatened and in friendly territory continue to move around, blocking healing, but lone units in enemy territory adjacent to powerful enemies remain stationary to heal rather than attempting to withdraw toward safety.
The AI's economy may be too strong now. It is able to support massive armies and still advance technologically at a rapid pace. (For example, in one Marathon game on Monarch difficulty I popped two early techs from huts, but at turn 51 (about like turn 17 on normal speed) five of the seven AIs were still ahead of me in points.) The AI was alway one for large armies, but the post-H AI armies dwarf pre-H AI armies. The result is slowdown of game turns, happening earlier and becoming more severe than previously. The effect is exacerbated in peaceful games, because the AI just keeps building more units. If one city can support 30 units then imagine the army a 20-city AI can field. Naturally the impact of this will vary based on computer and also play preferences. A duel game with 4 AIs doesn't have slowdowns on my computer, but my preferred play setup pre-H (Marathon, Standard, 8 AIs) is much slower. The only thing that makes it possible to cope with these large armies is the AIs mismanagement of units, mishandling of wars, and tactical errors on the battlefield. If the post-H economy were joined to the pre-H warfare then I'd have to drop the difficulty setting 3 or 4 levels to compete. I don't mind losing to superior combat tactics and a well-managed army, but I don't want to be beaten by an AI that just drowns me in units. Several posters have called for a reevaluation of the production and unit support modifiers the AI receives. Right now the AI does not handle warfare well enough to warrant a change, but hopefully the military AI will be improved and then I definitely agree that some adjustments to reduce unit bloat will be necessary. This should also improve turn speeds, and hopefully allow some of the performance gain I was hoping to see from CAR's incorporation to shine through.
Privateers immediately declare nationalilty. I'm not sure if this is alway true, or if it's some sort of response to the presence of stronger enemy vessels. In any event, declaring nationality on a Privateer throws away much of the purpose of building it to begin with. If the AI is in a situation where it feels the need to immediately declare nationality on Privateers it builds then it should instead build another type of ship. Privateers are a dead-end of the upgrade path, so if they are already so outdated that hidden nationality represents too great a risk, then building a ship with upgrade potential would make a better production investment.
Goblin Forts no longer accumulate defenders. Previously, barbarians were not allowed to move away from a Goblin Fort if it had less than three defenders. This made Goblin Forts challenging targets, because by the time you could build enough Warriors to kill the Archer, a Scout had spawned. Often it would be necessary to wait until three Scouts had spawned, and the Archer had moved off, before taking a Fort would be practical. Spawned Scouts now move off immediately, and because a lone defender is very vulnerable this makes it possible to tackle Goblin Forts much sooner. I consider this unfortunate, because instead of being some of the last bastions of barbarian resistance, Goblin Forts are now not much different from other lairs. This also has implications for the Clan of Embers, because easier elimination of Goblin Forts means that the opportunity to use them to raise armies near an opponent is decreased.
AI spell casting logic doesn't seem to take friendly fire into account, or else it is too liberal in its willingness to accept friendly fire. For example, the AI now uses Maelstrom but is willing to cast it in situation where many of its own units are damaged. This may be very difficult to address, unless there is a way to coordinate casting decisions with the movement timing of other units (which I suspect there is not). At least, the AI should choose not to cast AoE damage spells when the total number of friendly units that will be damaged is greater than the number of enemy units. (If possible, damage resistances should be taken into account, so that if many friendly units will be hit but will take little damage then the AI should be more willing to cast.)
Stacks of casters do not distribute their casting choices. For example, a stack of mages with access to Rust and Fireball all chose to cast Fireball against my attackers, and so my Weapon promotions remained. If one or two of the enemy casters had cast Rust instead the reduction of power of my stack would have been greater. Perhaps there needs to be a chance that a caster will opt to cast a "lesser" spell even when a "greater" one is available. Alternatively there could be an ordered list to the spells, such that each spell is checked in order and skipped only if casting it would not be helpful. Were such as system implemented, and Rust properly listed before Fireball, then the first Mage or two would have cast Rust, and then when most or all of my units had been stripped of metal weapons (and debuffed via the Rust promotion) the rest would have cast Fireball.
The AI does not defend newly conquered cities properly. Its attack force moves on immediately (without even stopping in the city to heal), and the city is almost always left undefended. The AI will often build a defender in another city and send it to the new one, but there is usually a window of several turns in which the city can be retaken by the original owner (or someone else) without any combat. Even when defenders arrive they are often inadequate to the task (ie not enough of them). It seems that it would be better for the attackers to remain defending the city until it has adequate defenders of its own before moving on to a new target. Perhaps the AI should include city defense units in its attack stacks, which can then remain in a conquered city to defend it. Alternatively, the city could just be razed, if the AI does not have the units available to defend it properly.
I encountered one situation where the AI actually abandoned a city I was assaulting. Most of the city's defenders had been killed, and the few remaining units were unlikely to survive the next round of combat. Perhaps this was a savy decision on the AIs part, to salvage the production value of the defenders rather than lose them in the hopeless defense of a doomed city of little value (it was a size 1 or 2 city in an icy area). At the same time, just handing cities to the enemy is generally a bad idea. I'm not sure this is a problem, since I've only seen it once, but it was odd (and in my previous experience unprecedented) and I'm still trying to recreate the situation.
-------------
Attached below are .7z archives of two games I've played to try to recreate problem situations I've seen occur. The games were played using Patch J.
Game 1: Duel Pangea, Monarch difficulty, low sea level, random climate, 4 AI, normal speed, allow permanent alliances, no tech brokering, all victories enabled. Randomly assigned Perpentach of the Balseraphs.
Spoiler Explanation of saves in Perpentach.7z :
Game 2: Duel Pangea, Monarch difficulty, low sea level, random climate, 4 AI, normal speed, allow permanent alliances, no tech brokering, all victories enabled. Randomly assigned Os-Gabella of the Sheaim.
Spoiler Explanation of saves in Os-Gabella.7z :
There are still some problems I've seen for which I'd like to provide examples. If I can replicate those situations I'll post additional example games.
As i said, it depends on which kind of map you are playing. Any kind of Erebus/Creation cannot be compared, because there are so many tiles which cannot spawn barbs (because these tiles are Peaks).
On any kind of big map created by Pangea/Life etc, where every civ starts isolated and there is lots of space, the AI has too much trouble.
I have tested 20+ games so far and it was like this every time.
Sure, a further improvement of the AI would be best to deal with this new threat. But more realisticly seems to me just to soften these new hardcore-pillaging Barbs.
As someone said before, the Barbs in Civ are not meant to be a super army with plenty of free units but a threat in the beginning which gives u the opportunity to level your units before the real wars begin.
I play tectonics, rocky. That is only at most 15 - 20% more land covered by peaks. So that would only be a 15 - 20% decrease in barb spawn. And the civs are spread apart.
NO WAY!!!
SAYS WHO!!??? Barbs are just supposed to be experience punching bags? That's lame! Barbs are supposed to harass, pillage, and sack cities if they can.
Just read the 1st page. This is only one of many examples.
As someone different said, if in a game of 10 players, 3-5 players get overroled by the barbs, this is not a game i want to continue playing anymore. This is just too much.
In my pics its about turn 260 in epic game speed.3 Civs do not have a single improvement in their whole empire! Dont you think, this is too much? These Civs, if not wiped out by the barbs, could never become a threat to me, because at the time they have rebuilt their economy, i have already won the game.
Of course i would love to have such a dangerous barb-AI ingame, but i am convinced that the team will hardly ever be able to improve the defending AI in a way, that it can deal with this threat in 90% of the situations.
One crippled AI is ok, but if half of the opponents cannot play any other role than being cannonfodder, this is not a game i can seriously play (unfortunately!!)
Senethro
Overlord
Play smaller maps so theres less area for barbarians to spawn in. Thats part of the problem. While the AI is in an unusual place right now, you're exacerbating it by playing an unusual game.
That depends on the idea s.o. has about a world like Erebus.
For me, it is a big dangerous world, where every civ has to find out of the dark age by its own and then, after a period of establishing, discovers this world and finds out about other civs trying the same.
The mapscript "Life" is the most flavourful i have found to simulate such a world, much better than erebus&co in my opinion.
If the new patch is only playable on maps, where you have contact to 3 other civs after 10 turns, because everyone is starting so close to each other, this is just not what i feel to be appropriate for such a theme.
Don't get me wrong. As i said, i would love to have these intelligent barbs, but only if the AI-controlled civs have a successful strategy to deal with them.
If i am right, this is the feedback-thread, and my personal feedback after 20+games is, that the new barbarian-threat is way too overpowered at the moment.
And if you read this thread, you will see, that i am not alone with this view.
Something definetely has to be changed. What exactly is the task of the team.
For me, it is a big dangerous world, where every civ has to find out of the dark age by its own and then, after a period of establishing, discovers this world and finds out about other civs trying the same.
The mapscript "Life" is the most flavourful i have found to simulate such a world, much better than erebus&co in my opinion.
If the new patch is only playable on maps, where you have contact to 3 other civs after 10 turns, because everyone is starting so close to each other, this is just not what i feel to be appropriate for such a theme.
Don't get me wrong. As i said, i would love to have these intelligent barbs, but only if the AI-controlled civs have a successful strategy to deal with them.
If i am right, this is the feedback-thread, and my personal feedback after 20+games is, that the new barbarian-threat is way too overpowered at the moment.
And if you read this thread, you will see, that i am not alone with this view.
Something definetely has to be changed. What exactly is the task of the team.
![[to_xp]Gekko](https://civfanatics-data.community.forum/avatars/m/74/74677.jpg?1475248346){kind=avatar}
[to_xp]Gekko
QCT junkie
Life? never heard of this mapscript, where is it?
its not the barbs.... it's the AI's being too passive. They do the same thing with regular enemies. the barbs do not need to be changed, the AI's need to be changed.
For the record, I am no longer playing any games either, because I see the AI's fail miserably to use their armies to protect their empires. If the barbs don't run them over, the first AI to declare war on them will.
I would not doubt the ability of the team to come up with better counter attack strategies. It seems to me, right now, they have a very simple strategic system: a unit is either randomly assigned defense or attack. Attack goes into a stack, and waits for the go ahead on invasion, and defense sits in a city.
Onionsoilder
Reaver
- Joined
- Mar 19, 2007
- Messages
- 3,173
Yeah, I'm no longer playing games either. Before the patch I was struggling with Immortal, winning maybe 40% of the time. Now I can win 90%+ of my Deity games. The AI can make a lot more units now, but they don't do anything with those units. Using Hawks, I saw the Malakim let a single Orc pillage 10 tiles worth of improvements. They had 30 axemen wandering around outside the city, not killing the orc.
Just because the AI can tech properly and make a lot more units does not make the AI good. I thought it was bad when I realized a few simple tricks I've learned that would never effect a human player, can effectively halt an advancing AI army in it's tracks. Things like using summons as bait or using sacrificial cities. The fact that they can't even deal with barbarians proves that the AI is more helpless than it ever was.
Just because the AI can tech properly and make a lot more units does not make the AI good. I thought it was bad when I realized a few simple tricks I've learned that would never effect a human player, can effectively halt an advancing AI army in it's tracks. Things like using summons as bait or using sacrificial cities. The fact that they can't even deal with barbarians proves that the AI is more helpless than it ever was.
Medicine_Man_55
Warlord
I just adjourned from a test game. FFH2.41j.
Game Parameters: Shieam Civilization, Tebryn Arbandi, Normal time scale, Emperor difficulty, Huge Pangea, Tectonics, 10 AI civilizations, Raging Hordes, Aggressive AI, All victory conditions (ex Time), All unique features, No tech brokering.
Rivals: Kuriotates, Bannor, Sidar, Grigori, Ljosalfar, Khazad, Amurite, Calabim, Clan of Embers, Illians.
My impressions of the evolved AI continues to be a mixed bag.
The Good:
Contrary to what has been reported many times in this thread, I did not see the AI building gigantic stacks of archers to attack with. Most stacks were a mixture of axemen, hunters, archers, and adepts. The bulk of the troops were axemen.
Also contrary to what has been reported, I did not see any AI civilization in danger of being wiped out by barbarians. They stacked warriors early and later axemen/archers on defense. The only civ that was constantly getting swamped by barbarians was the Illians, who had a weaker starting location located next to a barbarian civilization (the Clan of Embers).
The Bad:
I'm not entirely sure how to interpret what I have observed. I think the enormous stacks of military units the AI is building is hurting it in the long run. The economic reasons for this are easy to surmise but I also think the AI is laboring to build huge stacks for counterattack (UNIT_COUNTER) too early in the game while at peace. Prior to close confrontation with other civilizations, a more distributed defense is truly needed. This is shown when nations with incredible armies are being pillaged harshly by barbarian raiders.
I suspect the over-large armies are also responsible for the lack of aggression. If it were simply an issue of all civilizations being too well defended, I think perhaps one or more of my rivals might have invaded me. Despite having 1/4 of the armies of my rivals I was hardly threatened and never attacked.
I think most of the improvement in the AI needs to be with the production AI. It needs to build less troops and smaller stacks in the early parts of the game. Large-to-Titanic stacks of troops for invasion or counter attack should be built as the AI secures its heartland.
At present, I agree with the sentiment that the AI is currently less of a challenge than in Patch H. It doesn't threaten the player with destruction or compete for wonders. The AI seems geared to thwart an early rush but overproduces in the long run. There is a lot of promise, so I hope Sephi is going to keep tweaking his AI mods.
I've attached my save game to this post.
Game Parameters: Shieam Civilization, Tebryn Arbandi, Normal time scale, Emperor difficulty, Huge Pangea, Tectonics, 10 AI civilizations, Raging Hordes, Aggressive AI, All victory conditions (ex Time), All unique features, No tech brokering.
Rivals: Kuriotates, Bannor, Sidar, Grigori, Ljosalfar, Khazad, Amurite, Calabim, Clan of Embers, Illians.
My impressions of the evolved AI continues to be a mixed bag.
The Good:
- The AI settles and develops cities briskly and intelligently.
- The AI researches intelligently. Early, cheap economic and infrastructure techs are prioritized.
- The AI assigns ample defenders for cities.
- The AI uses magic to terraform. The Grigori deployed adepts for this purpose very early in the game (palace water mana vs a partially desert starting location). Very impressive AI improvement.
- The AI chooses religions in an intelligent and thematic fashion.
Contrary to what has been reported many times in this thread, I did not see the AI building gigantic stacks of archers to attack with. Most stacks were a mixture of axemen, hunters, archers, and adepts. The bulk of the troops were axemen.
Also contrary to what has been reported, I did not see any AI civilization in danger of being wiped out by barbarians. They stacked warriors early and later axemen/archers on defense. The only civ that was constantly getting swamped by barbarians was the Illians, who had a weaker starting location located next to a barbarian civilization (the Clan of Embers).
The Bad:
- The AI doesn't place a high priority on building wonders. The heron throne was built and all other wonders were left available until turn 250. I deliberately didn't touch them myself.
- The AI doesn't defend its territory very well. Large stacks of troops were on patrol while small groups of barbarians overran improvements and pillaged. Wounded troops in large stacks using UNIT_COUNTER_AI were not given time to rest.
- The AI is very passive. Only one war was declared, even with aggressive AI enabled. The Clan of Embers (top 4 civ) declared on the Illians (#13 out of 13).
- The AI wasn't using catapults, despite having a fair number of both siege workshops built and chariots deployed. During the Clan assault on the Illians, massed axemen and lizardmen were employed against cities.
- Promotions were sometimes not being chosen for units, adepts in particular.
- The AI research rate was slowing down in the mid-game. I was able to overtake the tech leaders while remaining at 40% research rate. My army was much smaller than theirs, of course.
- Still no naval units or rituals from the AI.
I'm not entirely sure how to interpret what I have observed. I think the enormous stacks of military units the AI is building is hurting it in the long run. The economic reasons for this are easy to surmise but I also think the AI is laboring to build huge stacks for counterattack (UNIT_COUNTER) too early in the game while at peace. Prior to close confrontation with other civilizations, a more distributed defense is truly needed. This is shown when nations with incredible armies are being pillaged harshly by barbarian raiders.
I suspect the over-large armies are also responsible for the lack of aggression. If it were simply an issue of all civilizations being too well defended, I think perhaps one or more of my rivals might have invaded me. Despite having 1/4 of the armies of my rivals I was hardly threatened and never attacked.
I think most of the improvement in the AI needs to be with the production AI. It needs to build less troops and smaller stacks in the early parts of the game. Large-to-Titanic stacks of troops for invasion or counter attack should be built as the AI secures its heartland.
At present, I agree with the sentiment that the AI is currently less of a challenge than in Patch H. It doesn't threaten the player with destruction or compete for wonders. The AI seems geared to thwart an early rush but overproduces in the long run. There is a lot of promise, so I hope Sephi is going to keep tweaking his AI mods.
I've attached my save game to this post.
