I'm on the AI today actually.

But I need your help for those of you who are stream rolling.  The AI's weaknesses are more to do with strategy than anything else. 

To use a Starcraft analogy, it would be like the AI developer not realizing how important it is to block off their base making it easy for the Zerg to win.

Heavenfall aka "the broken record" says: The strength of the AI on higher difficulties would be multiplied if they focused on building groups and squads of units with high quality levels, instead of basic quality solo units. As soon as I get a group up, it can defeat a nearly infinite amount of solo enemies.

I would go as far as to say that as soon as the AI has access to quality and group upgrades, it should never build anything that is "less than max".

Pulling numbers out of my ass, a 4-man party of elite enemies would probably be 12 times as dangerous as these solo units running around.

I would also like to see the AI not make constant demands for gildar. It is way too easy to pay them off. Ask for gildar once, then invade!

My basic strategy on ridiculous:

1) Expand as much as I can. Get Tools of War / Equipment to build units with 9att 5def. After that, rush to libraries (or whatever the empire version is called).

2) Continue expanding until there is no more room to expand. Avoid war at all cost. Don't worry about AI declaring war, at the most they'll want some petty cash.

3) Research some weapons, armor, groups, quality upgrades. Train these with call to arms.

4) Invade. Game over, because AI has NO military strength.

I think the AI may be underestimating sovereigns and defense in its calculations.

1. The AI does not protect it's own town/cities, when I attack an AI town it rarely has more then 2 to 4 units in it, even after a 100 turns.

2. The AI builds way to many pioneer's, you can camp outside an AI town and just kill off pioneer's as they are produced.

3. Tactical combat, stop the suicide charge, the AI should wait and then try to overwhelm the nearest opponent, stop moving archer/casters, let them attack with ranged combat. Ambush mode, where a AI unit will hold their ground until an opponent get's within attack range, this will help the AI get the first attack.

4. Flanking, have the AI sent groups to attack towns to try to drawout town defenders.

So yeah, I think everyone here agrees that no amount of strategic brilliance on the AI's part is going to make up for the worst tactical AI I've ever heard of.

I think the Pioneer thing is actually a bug because during the beta I actually started hard coding limits on # of pioneers so there's some weird issue going on there. 

Now that I'm free to...indulge myself...we'll see what can be done.

I've never looked at the tactical AI, so I don't know how much I can do (and I don't want to step on someone else's area even if it's during the holidays).  

There's probably a handful of stupid things the AI does in tactical battles i could address though while I'm messing around.

I think the groups are a main problem of Elemental, because they make the scaling very difficult. A group has a higher damage output and higher HP. This makes the balancing of single units or spells very difficult. Other games like Age of Wonders: Shadow Magic do not have groups and they should be removed from Elemental, too.

How about actually never seeing the AI attack another player?

I had two ridiculous AIs declare war on me, but all their armies just sat around doing nothing.

Oh this brings up another common player tactic that AI should use. AI should stack its war armies on the border of the player before declaring war. It can then march in and stomp the area a bit. Though Teleport really breaks the game a lot, this AI change can really make a player feel that this "war" thing is something he should probably take notice of.

I've steamrolled quite a few myself, and I'll share my thoughts on the matter.

You are allowed to build up too much without harassment. I can literally turtle myself during what I like to call.. "The Closed Building Stage" and then commence with the "War Machine Stage" with my super-powered economy. At that point, I build up large armies of armor-clad juggernauts, and it's basically just a slaughterfest from then on.

Coding the AI to battle the human brain will be very tough, especially with a game that lets you do so much with everything. If the AI can build up groups of soldiers that would give me a problem, I can just wait a ways off with my army until that group leaves their city, and then I swoop in for the kill.

Game Alteration Ideas to make conquering more difficult

• Possibly institute a "Casus Belli" style system. Sure, it's what Paradox does, but the idea behind it would work with a game as flexible as this. You need a reason to go to war, otherwise you'll get some sort of "bad boy" rating that effects everyone's view of you. In other words, if you go attacking people without a reason, you'll be viewed as a monster. Therefore if you expand too fast and crush everything in sight, then all the remaining AI's will declare war on you and you'll have to stand against their full weight. Reasons for 'justifying a war' would be.. 1. They declared war on you earlier, 2. You are married into their family (rights by blood - perhaps limited), 3. Your city is near theirs, yet you have greater influence (it's overlapping theirs by a large margin), 4. You are in an alliance and an alliance member goes to war (you enter as allied member - but they are in control of the diplomacy behind it - ie, they can stop the war by diplomacy, while you cannot), 5. etc.
• Before taking a city you must destroy their walls. In other words, the better the walls the longer you have to wait with the city in siege before you can capture it. This will allow the AI to ring their town bells, so to speak, and bring back their soldiers to defend it, or quickly raise soldiers (I still think there should be a "buy out" option for soldiers - cost x seasons to complete = price). In short, if the city is currently under siege, the AI will have their nearest troops make a B-line straight to the attacker. If you have hedge walls, it gives you 2 turns before the city will be under attack. If you have stone walls, it's 4 turns, and so forth. This gives the AI breathing room to reinforce their cities.
• Add "Town Militia" technology in the War tree. This allows you to convert population into soldiers for times of emergency, granted they will probably be pretty crappy soldiers, but at least it's better than nothing. On the bad side of it, these soldiers represent your population, the more that die off, the worse shape the city will be in after - even for the conqueror. Perhaps deeper in the tree there can be "Trained Militia" which gives them regular weapons and armor (pikes & leather). On the bright side of it, the AI will always have some sort of protection because of this - even if their city doesn't have regular infantry. On another note, this could work in conjunction with the "city walls" idea above. The more time you have to build up the militia, the bigger it will be. Therefore, if you have huge walls which will keep the attacker at bay for a while, you'll have plenty of more time to raise the militia to greater numbers (more time for converting population into militia).

Part of the AI problem is that it needs to predictive as there is no way to be reactive.  If city zones had a slowing effect on enemy units, it would allow the AI to maybe mount a defense.  It might also allow a sovereign to gather units via teleport or "return" and otherwise react to an assault.

In the modeling of AI unit approaches, I would like to see an elitist style (expensive, effective maximized units), a soviet russian style (masses of inexpensive, moderately effective units), and a mythical style (armies primarily composed of fantasy races or mythic beings).  You could toss in stuff like lots of mounted units, massive recruitment of champions with magical gear being a part of that plan, leaders that utilize terraforming to devasting effect, etc. 

Regardless if these are well liked or not, there should be an overriding approach or technique that gives our AI opponents some discernable differences in how each one conducts the business of war. 

Top

The AI needs to build more tech libraries and use more spells. It needs to 'zero out' it's resource piles through trading or building.

I was fighting Resoln the other day, and I noticed that the faction had stockpiled like 2000 Gildar. They had 50+ Power level above me. Eventually, they decided to declare war. I crushed Resoln utterly. The only small stumbling block I had was that they HAD purchased a demon. Individually it would have caused trouble, but I bombed it with Arcane Arrows and killed it utterly.

Really?

RESOURCE MANAGEMENT 

If the AI has Gildar and Materials, it should be building basic troops. War Staff + Leather

If the AI has Gildar and no Materials, it needs to invest in more workshops.

I think the AI does this already, but it should research Magical Weapons/Equipment to drain it's excess Gildar.

If the AI has too many resources, and nothing to do with it, they should start building outposts to take advantage of empty queues. That, or forging alliances and paying other people to do it's bidding.

MILITARY STRATEGY

-Maintain large army groups! The AI needs to have at least one 'main stack' that is composed of the strongest units it can muster when at war.

-if the AI is at war, and there are units at it's borders, the AI should use casters and teleport to 'gather up' the strongest stack it can, and try to crush enemy formations. 

-Maintain parity with neighbors! If the AI is running caravans, it can see the cities of other players. The AI should try to maintain an army of equal (preferably better!) size and combat rating. It can be a champion type army, a mundane army, a summon army, whatever. The AI should calculate every player's total combat rating and try to maintain a force depending on it's disposition. (Of course, combat rating calculations need to change as well)

Military/Conquerer AIs want a force 50% larger than the largest neighbor by default. If a 'conqueror/Hitler' type of player has appeared and is waging aggressive war, then all AIs will want a 50% larger force than the conqueror 

Peaceful AIs want a force 25% larger

AIs that have a good relationship with the largest neighbor can maintain an equal force.

-The AI needs to identify where and when a 'climactic battle' will occur, IE when each player's largest armies will clash. Preferably the AI will try to force this battle if it is stronger. During this 'climactic battle' the AI should cast it's strongest spells with no regards to mana. Before the 'climactic battle' the AI should try to use all of it's enchantments, namely cheap stuff like Arcane weapon.

-The AI's military units should ALWAYS STAY CLOSE TO EACH OTHER! It should have one 'main force' or 'flagship stack'. If it needs to garrison remote territories, it should create separate units for that purpose.

-A clever, devious AI might have multiple strong stacks for the purpose of 'forcing choices'. It might attack two cities at once, or run through the countryside destroying improvements.

-If the AI has access to strategic combat spells, like volcano or column of fire, it should use them on the enemy's largest stack.

-If the AI is weaker, it should send fast units deep into enemy territory to attack resource nodes. At least 25% of the AI's military should have a movespeed >2, preferably as high as it can go.

-The AI should use Terrain spells to create chokepoints between it, and any neighbors. The AI won't do this if it is allied to a neighbor. It should pre-emptively close off paths to vulnerable cities.

-An AI on the attack should use Terrain spells to create a direct path to target enemy cities.

DIPLOMATIC STRATEGY

-Unless it is the strongest faction, or a homicidal maniac AIs should try to band together with 'like' factions.

-The AI should feel free to 'gang-bang' smaller factions when it has the resources to pay off other factions to attack.

HERO/SOVEREIGN STRATEGY

-The AI should specialize it's Heroes. There are three types, CASTER, COMBAT and CHARISMA. How many it has of each type depends on the sovereign and faction.

-AIs with many caster heroes should focus on getting tactical spells, and stuff like the fire books. Caster heroes should ONLY have Int raised, and nothing else.

-The primary trait of combat heroes is Strength. An AI with a lot of combat heroes wants magical equipment and lots of gildar to buy it.

-The AI should level up CHARISMA heroes in combat, and then leave them in a city when they are level 5 or so.

CITY STRATEGY

- The AI should have separate, specialized cities focusing on TECH, MAGIC TECH, and GILDAR advancement. What city receives what upgrades is based on nearby resources. 

-The AI needs to build MORE TECH buildings. Tech is the most vital thing in this game. Without tech, magical spells are very limited, and there is nothing to spend resources on. If the AI has a glut of resources or mana, it should focus on building technology fullstop.

-Build queues should never be idle! That is, unless the AI is running out of money/resources.

VICTORY STRATEGY

-Conquerer/Evil AIs should aggressively attack smaller factions. They should declare war once they've moved their forces in position, NOT before. They should make alliances of convenience and not be shy about backstabbing and absorbing weaker allies.

-Peaceful AIs should either focus on creating and maintaining alliances, and trying to win through an Alliance victory. They should strengthen 'friends' and weaken 'enemies'

-The strongest faction in an Alliance should be a 'peacekeeper'. If there are no more potential allies, they should move the alliance in the direction of isolating and destroying other factions.

-Weaker members of an Alliance should not be shy about gifting resources and units to a stronger ally if they are not immediately threatened by other neighbors.

-Neutral AIs maintain alliances to keep security and maintain a 'balance of power'. They try to win the game through magic or quest victory.

That's all I can think of for now.