I'm still steamrolling on ridiculous. I've tried everything to lose... big maps, small maps, lots of opponents, few opponents, different victory conditions, etc. I just can't lose. Any advice?
I'm having some trouble getting a decent fight out of it as well. Specific issues in addition to those that have been mentioned previously:
* The AI cannot properly recognize the effectiveness of an early mage push and therefore does nothing to counter it. I say 'push' as opposed to 'rush' because this is a strategy that sacrifices nothing in the way of long-term development. You can start kicking ass within 20 turns and only a couple of sovereigns (Procipinee, maybe Karavox) can do anything about it.
Suggested counters: You've got to be an idiot to send a sovereign out into a hostile world alone. The AI is, therefore, an idiot. Always have backup because they can win it even if you don't.
* The AI places too high a value on retaking lost settlements. You can just sit and fortify in a lost capital, knowing that its former owner is going to come hunt you down. This would be fine if he still had any sort of forces remaining, but you've just taken out his main production center. What are the realistic chances of success here?
Suggested counters: Someone should teach these petty rulers a lesson about rebuilding instead of revenge. Could go a long way towards their longevity.
*The AI is incapable of differentiating a strong board position from a weak one. Nations repeatedly pick fights with stronger opponents and make peace with foes they could be liberally crushing - without even wringing treaties out of them in return. If I'm not paying attention and have my pants down, I should get punished for it. The fact that I don't promotes laziness and poor long-term planning.
Suggested counters: Hard to say, but probably something along the lines of shifting the higher level priorities; like, if there's an evident economic disparity between an AI and one of his opponents, he should be taking advantage of his lead, or shoring up his weakness. No idea how you would write this, not being a coder.
*The AI is willing to sacrifice what is occasionally an absolutely atrocious amount of capital in order to have you fight your battles for it. This is stupid when you get paid in excess of 1000 gold to take out a nation you have clear military superiority over. To sweeten the deal they often ask for a peace treaty and you can more or less double your price at no extra cost.
Suggested counters: The value of the 'declare war' option should have some sort of coefficient based on the relative military strengths of the two powers involved. Also, if you get paid to fight a war and go back on your deal you should be looking at some serious diplomatic penalties with your contractor.
* The AI makes foolish decisions on where to place its settlements. Its selection of resources is fine. Its appraisal of existing political situations is not.
Suggested counters: Attribute some sort of negative settlement value to 'political strife', that is, conflicting fields of influence nearby. What kind of idiot ruler sends civilians off to die in a warzone?
*The AI has research priorities skewed too much in favour of adventure options. Enemies often impale themselves on creatures they spawn before they're ready for them. This makes your situation less of 'conquering' and more of a mop-up.
Suggested counters: This one seems simple to me from an outsider's perspective (which may admittedly be skewed): just make other options more attractive. Or make the value of the adventure tree relative to the nation's relative army strength: if they're prepared for anything, they should have no fear in unleashing the beasts of the underworld. They can be the ones laughing while the rest of the world gets slaughtered.
This is good stuff. I'll be posting some journals through the holidays on what I'm doing on this.
I also plan to be on IRC some during the holiday for those of you who want to hang out (in the virtual sense) and chat about this stuff. I may even be able to pass some builds from my machine back and forth.
What is the channel/password for the IRC.
The primary concern I have with the strategic AI atm is early defense. Most of the time I win the game on ridiculous, I take over 1-3 capital cities that were nearby (and relatively unguarded).
Besides this, I have noticed that the AI when it attacks, it comes in waves. The AI will send groups of 4 (I have yet to see more than stacks of 4 units). The AI will send 3-4 waves of 4 units at my city. Each unit is roughly the combat rating of the defenders of the city. And I would have lost the city had they combined into one large army.
That last problem actually leads to a suggestion I have for the game. A "logistics" technology (or leadership rating) would do wonders to prevent the "Stack of Doom/Teleport" strategy. (Maybe charisma could affect the number of units the champion could lead with a static cap for non-hero stacks.
Anyway, Thanks for the update Frogboy.
Just a general note. I don't think you can do the best job on the AI if you have separated development of the tactical combat AI and the general strategy AI. You need to know what combat strategies will be effective in various situations, and then you need to have the AI players create, arm, and organise (stack), and deploy units in a way that allows for the appropriate combat strategy for the current situation, given the current knowledge of enemy forces.
If the AI players find themselves in combat situations where they don't have an optimal stack, they won't be able to affect a challenging combat strategy.
Hey Brad, quick question while you are here: does the ai do quests? If so how does it know what options to choose in a quest with multiple options (i.e. some of the modded quests in the ACP mod by impinc). Just curious if we should be taking the AI into account when creating quests.
Things the AI does that make little sense that I haven't seen mentioned:1) The AI doesn't seem to hire very many heroes at all, giving me a vastly higher proportion of heroes to grab and equip myself. They'll do this even when sitting on piles of money.
2) The AI won't push an advantage if they are of equivalent or slightly inferior, or even slightly superior power ratings. The AI is only aggressive when its power ratings vastly outstrip my own. While I think this is fine for some nations, I think some nations should, perhaps at random, get an aggression trigger or notice an advantage of technology or position and capitalize on it.
3) The AI needs to fortify their cities better. I'll often see dozens of individual units out in the field (that is, not togther in armies), coupled with lightly defended cities. While I have equivalent (or even lesser) forces, because I concentrate my units into powerful armies, I'm able to engage their units at huge advantages and defeat each individual unit (and town) without major, or any losses.4) The AI doesn't use enough magic, both in tactical AND in strategic senses. I should feel a real sense of fear when challenging an AI player that is vastly proficient magically to me.
5) The AI doesn't recognize when I've exhausted my magical resources and capitalize on it, or press percieved advantages as strongly as it could in general. One time, when Resoln had actually defended their capital with a pretty sizable force, I lost an engagement, even after pouring 600 mana's worth of Blizzards at them (This is when you could cast more than one a turn). As my forces were very dependant on casters with this particular faction of mine, I was a sitting duck in the conquered city with my tattered army next door (composed of reserves and those heroes and offspring that got away). Resoln, despite being powerful enough still to press an attack and win, sat in their capital not only long enough for me to recover in a unit sense, but long enough for me to recover eight hundred mana and renew the assault, this time successfully.
Honestly, now that i really pay attention to what the AI is doing. The problem is not really the tac battles. Ya, the tac AI is terrible, but its really the AI for the city building. The AI can't build a city for anything. I noticed that even on the higher difficultys, my citys were easily out-growing and out-teching the AI citys. And then, the AI would expand. Huh? its first city was growing slow so its going to split its resources?
Maybe give the AI a basic plan to follow for the first 30 or so turns of its city, instead of letting it try to think it through?
Some of these AI flaws are probably things that could be solved (and perhaps better solved) by mechanics.
Unguarded cities= replace city level up mechanic, which feels artificiial somewhat, with better buildings for higher level cities (as a compensation), and militia that spawn on defense for towns when attacked. Lev 3 cities early game should be fortresses. Maybe something like sqrt of city with spears and padded armor. Have AI avoid said cities if they can't win, take your border towns, then ask for peace.
Sovereign suicides- just have the sovereign-led armies be very careful in enemy territory. Maybe even have them use teleport spell if they feel threatened to nearest border?
Militia is an intriguing concept and could be something to think about for future versions.
Teleport only works if they have the mana for it. Particularly an issue for the Kingdoms, who may never even research or be able to use their version of it.
teleport isn't that expensive. Call to Arms would be a more expensive alternative if the AI really wanted to try killing off your sovereign when you enter his territorry.
Speed kills in TBS's, and elemental does tend to make speed effectively really easy.
Idea for a spell: anchor. 1 maintenace, prevents transportation magic (teleport/return/call to arms) in a 3*3 field on the strategic map. That might help the AI also.
To WhiteElk's point, there are probably some optimal build orders to help out. Here's what I'm currently using with Kingdom and I'm not stalling or waiting for turns.
1. Build hut
2. Build study
3. Build workshop
4. Build beacon of hope
5. Build hut and farm on fertile land
At this point, my sovereign and Janusk have found some materials in goodie huts and I can start building troops/pioneers not to mention the workshop has now had three turns to crank out materials.
EDIT: Just realized this build won't work. I must be finding extra materials with my sovereign so you may have to swap #2 and #3.
Glancing around it seems like people have caught most of the glaring issues, but I'll chip in:
1. AI players don't build anything beefier than single units for the most part. Only the very top performers on the map seem to get that far.
2. The AI barely understands armor. It knows what atk value is good for and how to get it. But it can't think any higher than basic tier armor.
3. The AI doesn't understand it's under threat once war is declared. It reacts to players just fine. It will attack armies it can reasonably beat. It will make sure there are people on its cities. But it doesn't alter its production scheme. It doesn't make concerted efforts to capture player territory unless it's SCREAMING to the AI to be taken. (I.e. undefended. I see AIs taking cities from each other sometimes...but rarely from me.) As of the last beta, all they did was mob it with pioneers. It doesn't mass armies once war is declared, it generally just keeps expanding its armies at the normal rate it would have. When a player has just cut down 50% of their military, and they're still at war, the AI needs to re-up it's military as a priority.
4. AIs still freeze. For example, in my last game, I went to war with Umber. I destroyed two cities, that then put two food resources in completely unclaimed territory between our cities. I left those resources alone, to see who would claim them. No one did, hundreds of turns later. Umber never sent people to rebuild on them, and none of the other 4 AIs tried to claim them either, even though there was another third player very close by with clear access to it. If nothing else, I expected Umber to reclaim them. He had enough resources to build armies, that he sent the opposite direction against another player. He even had pioneers running around. So yeah. Not sure what happened there. But in general some of the AIs in the mid-game just seem to stop doing anything other than produce. Their units stop moving, they stop claiming stuff. In that same game, the Pariden player had a few units camped off my borders for 300 turns, no kidding. Probably more. They never moved by the time I abandoned that game. I see the same behavior elsewhere, from other AIs. Some just stop moving when they've run out of easy things to do. Even if the AI isn't up to something, it's units should still be doing something.
And as a last request....monsters and AI players need to treat each other as threats. They don't. Monsters are there for the AIs to kill for experience, and that's totally unfair to the player. Monsters that could ransack AI empires ignore them. Maybe you think that wouldn't be fun, to have AI empires fall to monsters. But it's not any more fun to be the only thing they pay attention to. If a faction falls to monsters, I consider that how the world shook out this time around. Plus, if AIs are actually building militaries to defeat monsters....maybe they won't be such push overs when the players arrive. Monsters are THE reason as a player that I have strong militaries by the time I meet the other factions, because I had to kill my there to just find them.
And I guess a final note of caution. We want AIs that feel challenging....but we also want AIs that feel different. There isn't an inherent problem with AIs that don't build super strong armies. The problem is that almost no AIs do it. We want variety in how factions play, we don't want them all be l33t military planners and tacticians. But there has to be a base level of understanding every AI has about what it needs to do to survive. The resourcing end, I think that's *mostly* sorted out. Military design is still one place it needs a lot of work. Even if an AI is meant to be non-aggressive, it should still realize it has to build army-sized units if the other players are.
On cities:
What I've seen is this. In lieu of actual resources as a reason for building cities, the AI builds research centers. That's its basic, end of thought build plan. It goes "Hrm I need research and I can't put anymore in this city/don't have the population for it. Welp, I'll just start an outpost and build another research lab!" Eventually they find food and their research labs expand to become really large cities built entirely on research buildings and nothing else.
It seems like the AI is, above all things, terrified of falling behind in tech (magic secondarily.) So it will screw up its entire empire (and probably use its city upgrades) in the name of !SCIENCE! Which is funny because then the AI often can't, or just fails to, do anything with all that fabulous research they've been doing. Battle axes, Umberdoths, and the city upgrades, are about the most tangible thing I can point to that comes out of their research, and the occasional squad-sized army.
It's appetite for research seems endless too. Even if it has 3 Lost Libraries (which it sort of optimizes for since it's focusing on research anyways) it will happily build more cities and fill them with research.
That's why I say fundamentals are important. Army sizes are a fundamental because the vast power ranges they confer. Every AI should get that. But should Pariden prioritize military tech the same as Umber? They should both know that you need big armies to not get crushed. That doesn't mean they should both have decided to research full plate and 2-handed weapons.
Elemental already has the problem of factions feeling homogeneous. I hope in the push for better AI, they don't become moreso and it's called "good" because the AI is at least competitive. Just look at production. The AI is "competitive" there. But it's a mess, and they don't even really seem to capitalize on their faction advantages.
I love this thread. Lots of good ideas. All of my games are on medium maps, ridiculous level, custom sov. I have no idea how to code but I'll do my best to explain how the AI should feel. I have no idea how easy or difficult this would be to do. I get to live vicariously as a developer on the SD forums. The below idea could also be used to modify personalities of each faction too. Maybe one has an automatic warfare mode from the start or something.
Break the AI code into three modes:
Admittedly I don't know squat about coding so there's probably plenty of holes in the above post. I also got tired of looking at my wall of text so got lazy at the end. Anyway, I think players intuitively assess danger on distance and power rating (which is a product of quality and quantity). If the AI could mirror that assessment it would be fantastic. I know, easier said than done. On a side note, teleport probably throws the AI all out of whack because a non-threat could turn into a huge threat in just one season. All I know is if I had the ability to teleport my GalCiv2 fleets anywhere in space really it would probably have thrown the AI out of whack there too. Brad, glad to see you working on the AI. Can't wait to see the improvements.
I hate to say it, but I think the real problem with the AI is that the game is too complicated for an AI to manage. There are so many different ways to accomplish your goals and this is a great thing for the depth of the game, but it also means there are countless loopholes and exploits that the human brain heuristic processor is so adept at discovering. Version 1.1 has done a brilliant job of really making you feel like you have numerous paths to victory, but the sheer number of available choices on a given turn is extremely difficult for a computer to filter, even ignoring the consideration of long-term consequences. There is the potential for some truly epic conflict in this game, but the AI can't come anywhere close to a human in recognizing the potential amongst such a vast number of choices.
GalCiv II had excellent AI, but it was also a much simpler game. Magic changes everything, particular when a spell like Teleport means that any tile in your territory x number of casters are all possible actions. There are all sorts of reasons you might want to do this too, ranging from trading items, to intercepting monsters, to picking up reinforcements from a town, etc. I love having all of these choices, but, in some ways, I think the game would be more fun with slightly more limited choices, so that there are fewer ways to exploit situations.
Ignoring the tactical AI (or lack of same) for the moment and the strategic AI once an AI nation has begun expansion, a very real problem I have run into several times now is the "AI that fails to expand". It is the AI that doesn't wage war against anybody and has plenty of space in which to expand and by turn 100 still has one city, and the same by turn 150, by turn 200. Sure, the city's population may be impressive by turn 100, but it won't be impressive by turn 200 - it may even be the same size.
It appears to be a case of the AI fundamentally building itself into a corner by using up all the food from its initial food-spot for houses in the capital. Once it has done that, it cannot expand its capital population any further and it cannot construct new settlements. It will only progress in population once it researches a tech that increases housing capacity one way or the other or a tech that unveils more food resources, which can take a long, long, time even if it tries to research the appropriate techs (something I have absolutely no reason to believe it is doing based on its performance).
Now, if city-production was a linear function of workers that could be acceptable - after all, if it was it wouldn't matter much whether 25 man worked in the capital or work in a smaller settlement - but Elemental does not work that way. Production is divorced from the number of workers.
The only way to get parallel building queues is through making new settlements and a nation researches much quicker when it is churning out studies in two or three settlements than one.
The AI is a complete pushover when it actually has resources to waste, but that just doesn't compare in idicoy to the AI nation that stagnates and never (or so late that it is for all practical purposes irrelevant because the only reason it lives so long is because you wanted to see if it would ever expand) expands from its starting city
ANd great ideas & suggestions everyone. Can't think of a thing to add after reading the last 2 pages.
Well, part of the problem is the quasi-random nature of the game's design.
Factions can't depend on resources being there in every game, so the AI has to focus on things it knows it can do: build research.
Factions are supposed to have some differences, so not every faction mazimizes in each area, where the player does. So every AI has a major weak point that is sort of by design.
Add in continuing rebalancing over units, weapons, spell, cities, resources....and any one game can showcase a lot of places where the AI can't adapt to its situation.
It may not always result in a top quality game full of challenge. But I prefer the sort of "living world" design to a much simpler and more predictable one. The whole living nature of the game is what partially drew me to Elemental, and it's the only reason I can stand to play it now, after probably 20 games. If I was going up against the same AI, that behaved in the exact same ways every game, I probably would lose interest for a good year or more.
Can't say for certain how the AI currently works, but I feel like the AI would benefit (in both style and effectiveness) if it'd compare it's resources with those of its neighbors and emphasize (even over-emphasize) developing it's strengths. If it has gold mines, push hero development (at least until/unless the item shop changes - and please, please do change it). If arcane temples show up, imbue and start throwing combat spells - and make conquering elemental shards a high priority. Iron mines? Rush for tech research into a mid-game iron-based conquest push. Stuff like that - just over-emphasize the use of whatever resource edges the AI acquires. Having the AI treat pretty much every resource identically in every game, regardless of the available ratios, rapidly becomes boring, predictable, and produces poorly focused strategies.
Monsters do seem to go after the AI. They wiped out Umber in my last game and in fact I thought it might be one reason the AI is underperforming.
I second everything else in your post.
Also, city placement still seems a bit off. Altar colonized Umber's former start position, but place his city in such a way that he only got one food resource. I arrived later and took everything else (2 fire shards, lost library, food, diplo resource).
AI performs poorly in rush situation. Human player buys a few heroes, builds a few peons, heads for the nearest capital. AI typically has sovereign and wife as sole defenders, usually poorly geared. Battle ends predictably.
All AIs should beeline equipment and begin building at least 2-3 defenders as their first priority after getting the baseline buildings up. Baseline buildings are hut, (study, workshop OR workshop, study), hut, workshop, farm. If you just forced the AI to do that, it'd be brittle but functional and would definitely make rushing a risky choice rather than a ho-hum standard first expansion maneuver.
You should have seen it in the last beta test when every AI had Swords of Wrath and could seemingly spawn them on their children from start
There are many great features available to you once you register, including:
Sign in or Create Account
