Books on the craft of writing say that the "main character" plays three roles at once: the character the story is about, the character we root for, and the character through whose eyes we see the story. Typically, these three distinct roles are called the protagonist, the hero, and the viewpoint character, respectively. (Interactive media add a fourth role, the controlled character or avatar, who effects agency.) Also typically, in both media all these roles are rolled into one, collectively called the protagonist, among other things.
But casting the avatar as the protagonist is not always the best choice from a gameplay perspective. Gameplay is about decisions with consequence. The protagonist's goals and opposition are already set, so the only thing left remaining is the work, the "how", to attain those goals. Such an activity shares more with puzzle design than gameplay. The player just goes through the motions to complete the pre-written story, all impactful decisions already made for him.
The interesting choices exist near, and before, a conflict. From this vantage, one has many options on how one wants to relate to the conflict, effectively creating a gameplay resource out of the conflict itself. These choices can affect the tone of the game, so allowing the player all of them may not further the design goal, but they offer the player what is arguably the only freedom that counts: the freedom to choose one's impact on the world. Here they are:
Ally. Pick a side, any side, even if it is a new side. Though the result won't be much different than being the protagonist oneself -- indeed, an ally could become the protagonist in the event of a loss of faith or life -- having the choice is generally what games are supposed to be about. But also, once we consider life outside the major conflict, shades of grey can reveal themselves to us.
Support. Much like allying, but covertly. The avatar's alliance is kept hidden from most if not all characters. This leaves open options in case the conflict tips the wrong way. As long as the avatar pays lip service to multiple sides, sympathetic to all (at least in private), the player cannot definitively lose. Also, many major plot revelations in fiction involve a hidden ally -- or betrayer -- coming exposed.
Defuse. Play the diplomat. This again plays out like a puzzle game, but the pieces are the conflicting desires of people. Provided their desires are confused or chimeric enough to warrant sorting out, the compromise needn't even be an optimal, perfect solution that so many puzzles in videogames demand. Also, diplomacy handled badly can trigger a story advancement and gameplay change, with two added bonuses. One, later repercussions on the avatar can be nearly anything, which is a handy hook for a drama manager to have. And two, the player can certainly say he experienced agency in the story! In many ways, failing the diplomacy puzzle is more interesting than succeeding at it!
Escalate. Raise the stakes. Make it worse. Purposes for doing so vary, and depending on the stakes, may even be mannerly or morally suspect. But pushing things to a climax can be fun in both the gaming and fictive senses. Game-wise, fireworks happen and the pace picks up. Fictionally, one or more characters, including some not directly involved, will be pushed toward change and growth.
Prolong. Keep the conflict going as long as possible; ensure no one will win, ever. This is useful if the avatar would accomplish things that one or both of the conflicted parties would oppose. Since the conflict consumes their attention, much good work can be done in the lurch. Of course, prolonging a conflict is a little like keeping multiple plates spinning on their stilts: one is in constant danger of it all coming down on one's head.
Ignore. The last case, whether considered an edge case or the base case, is needed for more than just contrast or completeness. Negligence has consequences, too. Outside the saving-the-world trope, an avatar's political power, social standing, parental ability, freedom, or even humanity can be called into question by non-player characters if the game's AI is capable of recognizing them. For negligence to exist requires only a duty or a relationship to one or more persons involved in the conflict. No man is an island, and abandonment by one's adventuring party or ostracism by one's people is surely as much of a reason for Game Over as death. In some cultures, exile equals death. In some games, marginalization equals loss of agency.
Interactive media specialize in a wide variety of emotions: guilt, shame, guardianship, responsibility, complicity, achievement, competence, and anything related to roles and their expectations. But to mobilize these emotions, we must give players the power to affect NPCs in ways more subtle than life-or-death. This makes a new direction for game AI, since each character must be able to calculate his or her relationship and response to the avatar based upon their shared gameplay history. But in an industry filled with mindless hordes of enemies, perhaps an AI that must justify its antagonism is a creature -- and a choice -- worth making.