I'm building a grid turn-based game, and I have a couple of questions about Utility AI that I can't wrap my head around for a couple of weeks. Suppose we have a unit that can move X tiles and attack in melee range. I want an AI controller to enumerate all possible targets for this melee attack and score them.
Are "move" and "attack" two separate decisions, or is it one decision that does both the movement and the attack?
If they are separate decisions, then how can a decision to move consider a melee target? The option is to copy all of the considerations from the "melee" decision to the "move", but it feels bad.
If they are a single decision, how to reuse it if, let's say, I have another unit that knows how to teleport instead of how to move?
My question boils down to this: where is the reasonable edge when I want to make some sequence of actions a single decision or a couple of decisions?
Edit 1
After some thought, I think I have some additional context for my problem.
Cons of separate decisions:
One of the crucial aspects of my turn-based game is a "time unit" budget that each unit has in its turn. There can happen a situation, when the unit decides - well, it is a great opporunity to go to another dude's melee zone. He runs to the dude, but then - oops, no "time units" left for the melee attack itself! It happened because "move for melee attack" does not have enough context, and so the AI chooses a degenerate action.
The second problem, as already mentioned in the answers, is that I, as a player, do not pick a tile to stand on, and then one of the possible targets near this tile. I evaluate both of them at the same time.
Cons of a single decision:
Reusability and combinatorial explosion. If the unit can move, can teleport, and also dig to the target tile, then I will need NxAction decision makers set up, where N is the number of movement methods. I don't know if it is really an issue, though; maybe it only seems this way.