UniversalWolf: I've abandoned that game...not that it wasn't fun, but I was going to win without question and it was just a long slog to the end. It ended up being all of Arcanus versus all of Myrror, which was interesting. Arcanus was populated by nothing but gnolls and klackons, oddly. Jafar would attack my cities with huge stacks of beetles and wolf riders. In competent hands his potential power would have been quite intimidating, but there are several simple strategies the AI simply can't handle, no matter what bonus you give it. For one thing, it has no concept of how to create artifacts to maximize the power of heroes. Meanwhile my stack of 5 decked-out heroes can raze all it's cities without breaking a sweat. It also doesn't really grasp the concept of going for the kill shot of attacking the enemy wizard tower. It makes futile attempts sometimes, but the planning required to pull it off successfully is too complex for a game AI.

I eventually got Stream of Life from a node, which was useful.

I must say, I was pretty unimpressed with the dwarven Steam Cannon and Golem units. Steam Cannons seem to have a really short effective range, although they're decent enough within a couple of squares. Definitely not worth the cost, though. Golems...they probably have their uses here and there, but they also don't strike me as worth the cost. Neither one is as good as Nightmares. I think the best Dwarf units are Engineers and Hammerhands. The Engineers are particularly impressive for the cost. Not only are they great at building roads, but with adamantium gear they're quite respectable in combat.

The hidden gem of the Dark Elves is definitely the Priest unit. Everyone lauds the Warlocks, but those require a lot more investment and they're physically feeble. Priests can be produced sooner, have a really good ranged attack, and have healing both for units in the stack and as a combat spell. With adamantium, mine were usually tough enough to resist a few hits, too. My standard garrison for a walled city was one Nightmare and one Priest, and that was plenty. I'd add more units for front-line cities, of course, but for "safe" cities that was enough. I never even bothered building Wizard Towers.

I had a couple Troll cities, and in retrospect I would go with War Trolls over Steam Cannons or Golems. They're unquestionably one of the best units in the game, even without adamantium.

Tervvo: Hammerhands are also very good.

UniversalWolf: Ever since my Landscaper build experience, I've been looking for a good dark elves build. Somewhere I saw someone mention dark elves with all life books, so this is my new experiment. I won my first game, so I know it can work despite how counter-intuitive it may be.

Dark elves have two major drawbacks: the have the slowest growth rate, and all other races hate them. Both these weaknesses are countered by the spell Stream of Life. It boosts the growth rate, and just as significantly it eliminates all unrest. Normally dwarves hate, hate, hate dark elves...but with Stream of Life they love dark elves, and they multiply quickly too. High Elves? Draconians? No problem.

The only drawback to Stream of Life is its expense -- 8 points per turn. But dark elves have the highest mana generation per population figure of any race, at +1 per turn, so their strength also works well for an all-life books build.

UniversalWolf: [...] but there are several simple strategies the AI simply can't handle, no matter what bonus you give it. For one thing, it has no concept of how to create artifacts to maximize the power of heroes. Meanwhile my stack of 5 decked-out heroes can raze all it's cities without breaking a sweat. It also doesn't really grasp the concept of going for the kill shot of attacking the enemy wizard tower. It makes futile attempts sometimes, but the planning required to pull it off successfully is too complex for a game AI.

shaddim: I guess you played without Kyrub's unoffical MoM patch?
Because Kyrub tried to optimize the AI behaviour and I think some of your mentioned issues are fixed with his patch:
http://realmsbeyond.net/forums/showthread.php?tid=3663

UniversalWolf: Nope, I use Insecticide. It's a great improvement, but the AI still doesn't know how to manage artifact creation and heroes, among other many other things.

UniversalWolf: I think it's beyond the ability of anyone to fix, unless maybe they have the source code and some significant programming skills. The AI just "thinks" in simple patterns, so giving it more abilities won't really make a huge difference. That's why the difficulty levels, as they increase, give the AI more and more and more units compared to the human player. The only way to make it hard for the player is to overwhelm him with so many enemies that even the AI's simple patterns have a chance to succeed.

rakenan: Regardless of language, good AI is hard. Kyrub has done amazing things with the MoM AI, but I don't think the framework of AI in MoM is sufficient to implement truly good AI. He's made it far less stupid, but I think it's just not possible to make it truly smart without rebuilding the entire project from the ground up. Even that is likely not enough - there have been many attempts at building a truly smart AI in complex strategy games, and I don't think a single one has succeeded. I mean, chess is quite simple compared to something like MoM, and we've barely managed in the past few years to build an AI that can compete with the best humans at that - and it is nowhere near developed enough to be packaged in a consumer product yet.

UniversalWolf: Makes me wonder whether MoM is more or less complex than chess, purely from an AI standpoint. Presumably more, but then chess is pretty deep.

EvilLoynis: ...and these are just some of the reasons.