The AI in 1.12 is flat out broke when it comes to waging war. Its fine fighting other AIs as they are broke too so it evens things out, but against the player it is just crippled. On my current map I setup a me vs team of 4 vs team of 4, all AIs set to Expert. I figured id sit in the corner building up my empire while the 4 vs 4 duked it out, and when one of them was winning (or had won) they would then come after me. That part worked out beautifully and eventually after many hours I had the 3 strongest nations that were teamed all declare war on me within like 3 turns. So here I thought I would get this awesome siege on my territory that I could repel (or at least try to). Instead...
After declaring war on me, none of the three AIs, all of whos territory and cities bordered my kingdom, had even a single unit at my border ready to attack. This was like 1939 poland declaring war on germany and then trying to prep some cavalry to go across the border. This is all the while my whole armies are stationed at the borders to defend my frontier. I waited and waited and eventually the first enemy stack arrived, 5 units, three of whom were spearmen and two were archers. Why on earth would the AI send an incomplete stack? All by itself? So it died a horrible death as expected, and I waited and waited and eventually another stack showed up. 3 units, two of which were spearmen and one was an archer. That AI's military strength at the time was rated 2500. So I kept pressing next turn hoping that eventually the blitzkrieg would show up and I giant wave of enemies would descend upon me. Nope. They kept slowly streaming in one incomplete stack after another, single file, from all three AIs. EVENTUALLY the AI rolled out its giant gun: a stack of 1 champion and 5 pioneers. *facepalm* . This was followed up by 2 champions without any escorts who would then after coming back to life after 10 turns without even having full health charge and attack my main stack all by themselves. The strategy of confuse the player, brilliant!
So eventually I got bored and sent my two main stacks in opposite directions to conquer some enemy cities. I wouldnt raze them, I wouldnt build anything there, and I wouldnt station any units there to defend. Just wanted to see what the AI would do to try to take the cities back. To my surprise (and in hindsight I shouldnt have been surprised) the AI did next to nothing. It initially tried to take back one of the cities, which I let it a few times, only to retake it. It soon after that stopped trying to take them back, even though they were undefended. I went deep into the enemy territory and captured a few cities here and there and then retreated to my homeland. Surely the AI would retake undefended cities deep in its lands, yes? Sadly the answer is no, it went about its business completely ignoring the loss of the towns it once had, even though it had them fully surrounded with other towns it controlled.
So to recap:
AI does not send full stacks to attack, or at least very rarely
AI does not escort stacks with other stacks to create a big attack
AI does not escort champions enough and uses them as just a regular unit, probably treating them as just scouts
AI does not defend its border which it should do at all times, even during peace time
AI does not change attack patterns based on previous results and thus never learning from its mistake (aka always attacks same spot, same way)
AI does not setup multi pronged attacks (attack 2+ different places at the same time)
AI does not retake cities with any priority
I played Elemental 1.0 when it first came out for a few hours. Uninstalled it and waited till FE 1.1 to play again hoping to get the complete and fixed Elemental experience. Im still waiting.
Wouldn't 'distance between cities/armies' only come into play for units deemed by the AI 'surplus beyond required city defense'?
Also, there have been enemy amies nearby (then again, perhaps they were moving towards the cities and just hadn't arrived yet, or perhaps they were affected by pathfinding limitations, etc.)
How about strategic spell casting? Affected by map size?
I do not understand this reasoning at all.
I often find myself moving defenses between cities, when they are threatened. And, in my opinion, the logic needed to emulate this behavior and generally to estimate how many defenses are sufficient would be much simpler to address on small maps than on big maps.
Logic train speculation goes like this: cities are undefended, units are roaming nearby, units had to come from somewhere, it's possible some of those units came from some of those undefended cities, AI champions may not be sufficiently caring about whether city is well enough defended before shipping off units -- which would be unaffected by map size.
Capiche?
How about AIs (not) casting mid level spells like tremor/freeze on invading armies> Map size related or no?
I just bought the game a few weeks ago for $40. All I know is what I paid for.
It's a design choice explained here: https://www.galciv1.com/encyclopedia.asp?action=showpage&page=https://www.galciv1.com/docs/criticalmass.html
I think it's smart. When you are the underdog, don't you try to befriend the most powerful civ? It's what the AI is doing too, it's trying to get along with you so you don't slaughter them. This design choice will produce some interesting results, not seen in other 4X games, like the late Civilization V for example. In GalCiv for example, if you find yourself being the underdog, you can try to appease the powerful civs until you develop your diplomacy or economy, and then slowly build your way up. I won games in GalCiv without firing a shot.
I'm not denying that there are situations where it's going to be pointless for the AI to resist. But it's also possible to crunch the numbers and come to the realisation that in X turns, it's possible to catch up to the human player.
It's not just the current military might that the AI should be tracking, but the rate at which it is changing. A smart AI will take advantage of a marked decrease in military strength or a long period of no change to gradually build itself up to the point where it can make a challenge. It's not smart to send out your best army and not be able to replace it in a reasonable time if something horrible goes wrong.
Equally, if you just passively allow another player to build up their infrastructure then you're letting them win.
Let's put it another way. Any player (human or AI) can be beaten, up to a certain point. Up until the 'critical mass' is reached, the player's military might during times of peace is basically a bluff. You both know what the result might be if you fight a straight-up slugfest, but if those military assets can't be everywhere at once then it's possible that you can hurt the enemy somewhere before they can retaliate with their full strength, destroy their infrastructure and then withdraw to somewhere where you have an advantage in battle or have reinforcements waiting. It's also possible to wipe out the enemy forces before they're organised into an army, making it more difficult for them to mobilise killer stacks.
Because yes, once you have full stacks roaming around the map looking for a fight, there's not much that can stop them, except a meaner, nastier full stack. The sheer weight of numbers doesn't always decide the outcome of a battle, but what it does do is ensure that you will still have adequate combat strength even after taking losses.
And this widens the gap between human and AI player - while I have had games of Fallen Enchantress where my regular units have reached level 10 or more, the AI units are lucky if they gain any more than a single level before getting wiped out. It doesn't mix vets and rookies like I do, to try and maintain a balance of good units.
That being said, human players are generally better at associating. We associate big nasty dragon with lots of our regular units being turned into kindling. And we also remember that when the AI fields magical stave units in big groups, it's best to kill the stave units with a vengeance, and it won't be easy to replace them afterwards because of the material requirements, so it's worth whatever units we sacrifice in order to get it done.
It's not the AI's fault that we're so vindictive.
I disagree... with the whole AI is broken idea.
most games I've played 1 or 2 of the AI players generally outplay me to the mid game stage and then I've had to deal with their superior armies at some point.
I had in one game a stack of doom with a champion and 8 Juggernauts come at me... only reason I could stop it was I happened to have an early dragon in one of my stacks that I had to crash Civ techs to get.
Last game the AI had complete stacks that frankly were far superior to anything I had in my inventory at the time. I probably could have beaten them but it would have been a Pyrrhic victory at best. Only luck bit was that the AI is a little dumb still with dealing with monsters and I had a deadly Drake army between me and him that took out 8 of his stacks. (yes I set it up that way on purpose... it was fun)
the AI does pretty good at matching my level of play. it's far better than any other AI I've seen.
I've also seen at least 3 situation (one where I actually lost) where the AI recognized a hopeless military situation and goes for an alternate victory condition. That I've never seen another game AI do.
I think it is fair to say that the AI quality in fallen enchantress is better on medium sized maps and gets progressively weaker as map size gets bigger.
I don't know if Civilization v has this problem since that game becomes unplayable beyond medium.
Oh man! An AI discussion and I almost missed it. We're closed for the holidays.
I can assure you that the AI does, in fact, try to do all of the things that the OP says. Also, like Arzon above me mentions, yes, AIs get weaker the larger the map size. That's not just in FE but in every strategy game including Civilization, MOO, you name it.
That's because of "The traveling salesman problem". As a practical matter, the AI has a very difficult time visualizing a large map. That means in terms of coordinating armies as well as the various guessing algorithms that try to calculate the odds that there are units in its FOW coming for it.
Sometimes, the AI doesn't guard its cities well, especially on big maps and late game when units move fast because it thinks that that city is "safe". On smaller maps, it does a better job guessing where units might be.
The AI also does cast strategic spells. A lot. Especially in v1.1. The problem is that it doesn't notify the player and this gives the illusion that it's jsut not casting. We need to give some visual indication that a unit has been cursed/slowed/etc. on the strategic map.
The AI does use the concept of army groups (this is something I never got into GalCiv). That is, it assigns various armies into a particular geographic group and will assign a group, prior to going to war, to head near the border. Unfortunately, I have not found the results of this algorithm to be particularly effective (i.e. it needs work).
All I can say is, if you're playing on bigger maps, I haven't forgotten you. The AI continues to get worked on. I do recommend that on larger maps, turn up the AI difficulty. You can also edit the coredifficultylevels.xml file and let the AI have the FOWcheat at lower difficulty levels so that it can see where your units are. It's a cheat to be sure but you might find it more satisfying from a challenge point of view.
Writing AI is a tricky business as you can imagine. A few people have mentioned ways of getting a better game experience such as designing more units for each player. I like the idea of having a library of units that people share. I'll talk to Kael about that.
One other thing I would add:
Whenever you see something you think the AI isn't consistently doing, like what Nick mentioned, go ahead and list it. At the very least, I can look into it better.
No doubt AI is seriously hard to do for a game like this. But there is a common thread in the comments and I think areas that if improved could go a long way toward improving the perceived level of the AI.
I’m sure there are some more things that can be added.
@cokke
Note that if mobility is high enough you might not need a two pronged attack.
Note that if magic is high enough you might not need an army.
I do not know if these are worth coding offensively, but Pariden and Resoln can be rather nasty, magic-wise.
Magic is weak in the early game, and the costs of using titan's breath and cloud walk frequently can be high, but high magic with high mastery trumps even dodge troops. (Procipine can eventually get enough initiative to defeat a heavily defended city with single target spells and after that enough damage multipliers to destroy most cities with flame wave, and Resoln can get enough damage multipliers to destroy a city with a dirge of ceresa. Procipine can also turn neutral territory into Pariden territory without pioneers... This is a bit extreme, since most games would be over by the time this happens, but even before these extremes are reached a high powered caster combined with a defensive army to buy time can accomplish a lot - you acknowledge this in 12.4 but if might be worth acknowledging it in other parts of your rules also?)
Another option for high mobility would be roads and stables and horses. This combination gives you a movement of 32 for a cavalry unit, which becomes 40 if accompanied by a champion with tireless march. This movement is limited to roads, but it's still impressive, and it sort of works in neutral territory (you lose the stables so half movement...). This is another late game tactic for most people, but a sufficiently wealthy opponent (like a current insane AI) should be able to crank out the stables really quickly.
This I think I have strewn my thought upon in earlier posts,I still think there should be changes to the way roads boost movement (and the UI should still reflect half movement points, and quarter movement points left),to the massive movement bonuses of mounts compared to non-mounted units,and In general I think the game would be more fun, and the AI less challenged, if all vision of units was boosted by 1 unit, and meaby zone of control of cities by 1 too,so you have more time to see these insanely quick moving bastards that fly across the land at lightning speed because they randomly happened to find a horse.
I can't expect the AI to counter what it cannot see.Also the AI should be better at making "Mounted Raider" stacks, where ALL the units in a certain stack is mounted, so they can charge at the enemy territory and wreak havoc at that amazing lightning-speed all mounts go by in the world of elemental.(I know that doubling movement is probably realistic for horses, but I don't think its a good gameplay feature to have mounted units move at twice the speed as normal units, especially when normal units move so dreadfully slowly, I still really think the bonus movement stuff around the game should be reconsidered...)
Sincerely~ Kongdej
We get a summary of events when a turn starts (unit completion, war declarations, cities taken, etc.). Including strategic spells cast (not just the ones cast on us, but cast amongst the AI -- they are strategic and our Sovs could notice them, plus we're notified when AIs take each others' cities) to the turn start notifications would be nice.
I'm wondering how you're doing this. I see various possibilities:
1- AI decides to warDec, makes warDec, then makes an army (I see armies comprised of melee and ranged, so I'm assuming you're not just throwing together whatever troops are nearby, but are intentionally constructing armies -- very nice!), then send it off.
2- AI decides to warDec, makes an army, makes warDec, then sends army off
3- AI decides to warDec, makes an army, sends army off, waits a bit, then warDecs
4- AI decides to warDec, makes an army, sends army off, waits until army reaches its destination just outside target city, then warDecs
Option 1 is least 'effective', taking the longest time between warDec and their army arriving and attacking, option 4 most effective. Option 4 guarantees an attacking army will be ready to go at warDec time, option 1 has the least guarantee and the greatest chance of the army getting 'lost' (increasingly so as map size increases).
If you don't want to use option 4 for all cases, then perhaps use option 1 for the easiest difficulty setting and/or smallest maps, and option 4 for the hardest difficulty settings and/or largest maps (and scaling between of course).
The fact that a random game map is created out of stamps has implications for optimising the AI.
I don't mean that the AI should cheat by looking at four tiles of an as-yet uncovered bit of map and saying 'oh this is stamp 239 I should see this and this three tiles away' although that did come to mind. But the optimal ways to move through that stamp and from one stamp onto the next can be determined while the world is being loaded, in much the same way that the light level on tiles in a Minecraft world due to a moving light source such as the sun can be pre-determined rather than calculated on the fly.
What I'm getting at is that when the AI is doing its planning it doesn't need to necessarily consider every possible route of attack (or when it is thinking about defence, each angle of attack). It can think about it in a general 'I need to get from this stamp to this one in order to achieve this' way.
Stamps are (usually) bigger than tiles and there are fewer of them. And even when a map is not constructed from stamps, it can be arbitrarily broken down into chunks of a similar size and treated as if it were. This means the AI can have a pre-calculated rule of thumb or answer for:
After exploring and exploiting those stamps it can say:
It can worry about which tile it precisely needs to be on when it actually gets to that stamp, if it even gets there.
Armed with this information, the strategic AI can find out if it can attack a particular city or army on a particular stamp in a reasonable time based on the information from each stamp.
If it can't attack because the route would take too long, it can think about how to speed things up with city/outpost (and hence road) building.
If it can't attack because the direct route is choked, it knows where the nearest enemy choke point is.
For particularly large maps, making the iteration cruder by only considering the start and destination stamp and the ones in between may not be enough, so then you need to group stamps into contiguous masses (like a superstamp, say) and have another strategic layer of crudeness i.e. I only need to know how to get from this big mass to this other big mass, which stamps are on the joining edges? Okay I need to go to that stamp, I'll work the rest out when I get there.
I've been working on the AI today to do better on larger maps. It'll make the turns take a bit longer but I think I've made substantial progress.
PSA-About some of the posts on this thread and really this entire forum
I wish people on these forums would start being truthful. Seems like anytime a SD person posts in a response everyone wilts. You're not helping the game or SD. You're making it worse because wilting makes SD think that the game is in a good place. People outside of this community don't even look at the game because of its shortcomings. If you're truly such fans of SD help them by being honest and explaining what's wrong with the game. There is a reason only 100k people bought this game. There is a reason EWOM failed. Be honest with SD and tell them the problems the game has. Wilting and making excuses doesn't help anybody. It doesn't give SD a more marketable product and it doesn't improve our game play experience.
About the game and ways to improve it
EFE is a great game, good enough to make you angry that it isn't as good as it could be.The core problem of the gameplay seems to be that everything is based on RNG. RNG is fine for some aspects of a game but all of the aspects that are based on it limit the quality of experience.
The bottom line is, the AI really comes up short in this game. It's why EFE is something u can play 1-2 times and then there is no repeatable experience. Tactical, Diplomatic and world map AI all need help. Here are 3 things that could be better.
World- Army composition and world map task
Your typical experience in this game is the AI sending random half stacks of units without an intended goal moving in RNG around the map. Other times when you're playing on Insane it will amass random stacks of units such as 9 horse swordmen and attack with it. But not always, the units will sometimes move back and forth on your borders and not attack despite being at war and the city is not well defended.
Diplo-Factions go to war or befriend you without input from player. Sometimes they go to war without warning. Sometimes you refuse tribute to a stronger faction and they increase their rep with you. I'm sure this wasn't the intent. Unless the AI is programmed off of RNG too.
Tactical-Caster AI doesn't cast spells with any logic or reason. Seems to be entirely RNG. Maybe add some if then else statements. If Player has clump of archers, fireball them.
I already made a thread about it a couple weeks ago in the support subforum, but could you guys please look into the bug where asking for >100 Gildar from an AI in a resource-for-money trade causes them to overvalue their cash in the thousands? I have posted documented examples.
Happy Holidays Frogboy.
I am in the late stages of a wildlands map game and I really enjoy having all the different wildlands in one map.
What I discovered playing this map is once my champions got to about level 16 the tactical battles became one sided. Here are some of my thoughts why.
1) By this time one of the champions can cast fire ball. This spell is very powerful even when resisted. I would suggest the AI try to spread it units to try to limit the effect of this spell. This also helps against catapults too.
2) Most of my doom stack consisted of ranged units(mages, archers, and catapults) with two champions. Most AI champions wanted to attack my champions that are heavily armored and that had honey or pork to heal themselves. The champions absorbed the hits while the rest of the units stayed back. Two things I feel that would help the AI in tactical battles is to try to protect the ranged units better with spells or add traits that help in dodge or spell resistance and to better determined their targets. The AI needs to determined what spells they need to defend against and what is the largest immediate threat it needs to take care of first. Example is taking out splash type units first or to get the units close to the opposing units to make catapults less useful. Casting guarding wind or flame shield to limit damage. The AI needs to get its units up close and personal when going against ranged units. This means when spiders cast webs against ranged units, they need to move first then cast the web.
3) Another tip for the AI is that sometimes it's better to wound a group unit to a point where it becomes less effective. If a unit attacks at a 16 X 4 becomes considerably weaker when it is 16 x 2 or 16. I find this useful to soften the opponent from ranged units then finish them of with close range units.
4) When the AI comes into enemy territory it needs to make sure that they are not stacked on top of each other. I used freeze which froze three stacks of units. This allowed me to take all three stacks out.
5) The last tip I have to help the later game is to get the AI units better intuitive The units right now are being attacked several times before even getting to my units.
I am sure I will live to regret these tips, but I feel this would help the AI become more deadly.
Your efforts sir are most appreciated. We look forward to the changes!
- HOMM III (WoG)
- Civilization IV (Beyond the sword)
Even HOMM V (Tribes of the east) & the true magnificent successor of MOM, Age of Wonders (shadow magic) have better AI.
Thanks a ton, Frogboy.
I will agree with the CIV IV BTS but HOMM III and HOMM V? No way. Those games didn't even know how to play the game. They both are scripted and cheat at even the lowest level.
One real weakness with tactical AI is that if you have a unit adjacent to an AI unit, the AI unit will never move to attack a non-adjacent unit. It makes it very easy to pin AI with strong defensive troops while the vulnerable firepower sits safely behind. Some more work is needed to make it harder to protect back-line troops.
During the beta Frogboy made it very clear he wanted nothing to do with dynamic threat detection in the tactical AI. The AI chooses a target and sticks with it - simple as that.
There are many great features available to you once you register, including:
Sign in or Create Account
