You might be onto a hell of a game idea there!

I think without the ability to flee.  Staying out of range of the strongest opponent is just delaying the inevitable.

so? thats the smartest move anyway

since its *inevitable* like you say in the time you are there you have to make the most dmg possible... b4 you die

that is achieved stay out of the first combat range of the strongest melee, at least

I agree with Alstein, a simpler less intelligent AI for dumb animals and a more powerful, smarter AI for intelligent opponents.

Melee units won't run away though. Even if they did, they would reach the edge of the map and the opposing unit would be able to block them in with ZoC.

In chess there is "stalemate if the same playing position is seen 3 times" rule and "stalemate if no captures in 50 moves". These forbid forever lasting games. I think the last rule would be good for elemental, too. Stalemate if no damage dealt in last N moves. I don't know the right N, probably something like 3 * number of units in the battle or something...

This way the AI could play the most intelligent strategy. Playing out a draw in a situation where your other choice is losing the battle is a good strategy. Although this might be boring from playability perspective.

From the videos it seems one of the biggest problems is that the AI lets the enemy hit him first by moving into strike distance. This can easily decide the whole battle. Of course, an AI that waits for the human to attack is pretty boring from playability viewpoint. I think the best solution for this is to have, as often as possible, battles where one side benefits from moving in. This can be done by giving ranged attacks (even weak ones) or weak magical attacks to the units. The side which has more long-range potential benefits from staying back. Thus, it is completely OK for the AI to stay back if it has more long-range potential.

I think the most crucial aspect of the tactical battles is when to "mass-attack" your enemy, and how to execute that attack. By "mass-attacking" I mean a move where you try to get as many of your units to attack the enemy with as little enemy counter attacks as possible. I don't think the AI is particularly good at this, but it is hard to tell because in many of the videos the forces are so unbalanced that it doesn't seem to matter much what the AI does.

Don't spread this around, but I usually let the AI attack first. Since it will always come at me in no particular formation, I can sit back. The highest level heroes will hit me first, but be so far ahead of their army I can wreck their best units and then mop up the rest. It's more effective in terms of keeping all my units at full health. 

Also, when sitting back this is the best formation:

X is empty tile and O is my defender units. 

XXOXXOXXOXX

This way the enemy can't get past you because of ZoC. They will choose one of the O's to attack. On your turn you can swarm that unit. It is best to have your damage dealing units waiting behind this line. You usually get back behind your defenders before their regular army has caught up. 

after watching this, i don't get why the AI moves the archers away from an enemy.  is there some kind of penalty or something for moving and attacking or for archers attacking inside melee range?  if there isn't then there should be.  at least make it so that that if you don't move at all then you get an extra attack or something.

Given all parameters involved in the fights, and assuming that every move has a countermove, and assuming that the AI gets to be as clever as you'd like it to see, this game will end in a stalemate. Chess is not won because someone has an advantage. Chess* is won because someone made a mistake. Given enough time and CPU power, Chess is a stalemate only broken by white going first, and the rule that you HAVE to move. Theoretically speaking, any game with balance works like this.This is what I refer to when saying "In essence, this would turn tactical combat into a game we cannot win".

Its all theorycraft anyway, because 1. chances of this turning up are pretty darn lowand 2. chances of Brad programming a stalemate AI are pretty low and 3. you can check, beforehand, what the enemy's stack contains so there's no reason why this should occur, ever. You just rock-paper-sciccor em.

*note that Chess computers do not have a real AI, they merely compute all possible moves and pick the statistically most succesfull. This is the same process that humans use to win, but Chess computers can look further into the future... loking more than 3 moves ahead on either side is very difficult, even for 2400+ players. I used to have one as a co-worker

I don't think a stalemate would be possible.

A) non conquest victories

B ) not all troops can be in the same location at the same time

C) not all economies are equal

D) not all terrain is equal

how about after X turns of inaction, a draw is concluded and the battle ends? (inaction being no damage taken)