Balancing Player Agency and Authorial Intent in Interactive Narratives

109 concepts58 papers1 proposal

In interactive storytelling, creators face a classic tug-of-war. On one side is player agency—the freedom of a participant to make meaningful choices and shape the world. On the other side is authorial intent—the designer's need to maintain a coherent dramatic arc, deliver emotional beats, or achieve specific educational goals. As Vadim Bulitko identifies in Interactive Narrative: A Novel Application of Artificial Intelligence for Computer Games, this inverse relationship is the defining challenge of interactive narrative systems. Maximize player freedom, and the story risks fragmenting into logical inconsistency or aimless repetition. Enforce the script too rigidly, and players feel "railroaded," destroying the illusion of interactivity.

Deconstructing Agency: Perceived Control vs. Actual Influence

To solve this problem, researchers first decouple a player's subjective feeling of control from their objective capacity to alter the plot. As Peter Mawhorter et al. demonstrate in Towards a theory of choice poetics, a player can experience high perceived agency through expressive choices even if those choices do not branch the underlying plot. Conversely, blind choices that drastically alter the story without the player's informed intent yield low perceived agency.

This insight reveals that agency is not a function of the sheer number of branching paths. In fact, Zhiyuan Yang shows in The Impact of Nonlinear Narrative Games on Player Immersion that high-frequency branching often causes decision fatigue and narrative fragmentation. Instead, as Justus Robertson argues in Foreseeing Meaningful Choices, meaningful agency relies on prospective agency—the player's ability to anticipate the situational consequences of their actions. Choices feel meaningful when they cause distinct situational shifts (changes in location, time, or character goals) rather than cosmetic variations.

The Mechanics of Experience Management

Instead of hard-coding every possible branch, modern systems employ an automated "Experience Manager" or "Drama Manager" to dynamically guide the narrative. The goal is to move away from overt interventions—like explicitly blocking an action—which David Roberts et al. note in Computational Influence for Training and Entertainment causes immediate user frustration.

Target Trajectory Distributions

One approach is to treat narrative management as a probabilistic steering problem. David Roberts et al. utilize Target Trajectory Distribution Markov Decision Processes (TTD-MDPs) in Player Autonomy versus Designer Intent: A Case Study of Interactive Tour Guides to guide users toward a distribution of ideal experiences without forcing a single path. Similarly, Hong Yu 0010 et al. use collaborative filtering in Toward personalized guidance in interactive narratives to predict player preferences and selectively filter options, aligning the story's progression with the player's internal motivations.

Dynamic Story Repair and Accommodation

When a player inevitably goes off-script, how does the system adapt without breaking the story? This process is known as accommodation.

Traditional automated story planners often suffer from a "pull-back effect," where exceptional player actions are ignored to force the player back onto a predetermined track. To solve this, Adam Amos-Binks proposes in Problem Formulation for Accommodation Support in Plan-Based Interactive Narratives modeling stories based on high-level character goals and causal dependencies rather than low-level actions.

When a deviation occurs, an automated planner "repairs" the narrative structure. As highlighted by Jeff Orkin et al. in Playable Experiences at AIIDE 2014, this dynamic repair ensures that non-player characters remain goal-oriented and logically consistent, preserving the author's high-level vision while validating the player's intervention. For instance, Boyang Li et al. implement this in multiplayer settings using specialized Petri Nets to handle simultaneous player influences in [Robust and Authorable Multiplayer Storytelling Experiences](/paper/art_2e98a138501b4792b5ea16ca31914603.

Maintaining Coherence in Generative and Multi-Agent Systems

The advent of Large Language Models (LLMs) and multi-agent simulations introduces new challenges to this tradeoff. While LLMs excel at generating fluent dialogue, they struggle with long-term commitment to a plot script. Yingpeng Ma et al. introduce NCP-Bench in Can LLM Agents Stick to the Script? Modeling Commitment in Interactive Narratives, demonstrating that when players actively subvert plot points, LLM agents frequently collapse into logical contradictions.

To counteract this, researchers decompose the narrative into discrete, manageable segments. Weiqi Wu et al. propose a "Narrative Chain" and "Sparse Instruction Tuning" in From Role-Play to Drama-Interaction: An LLM Solution, where the model acts as an invisible narrator that subtly nudges the user through sequential plot segments.

Furthermore, Hongqiu Wu et al. utilize a "Plot-based Reflection" mechanism in Towards Enhanced Immersion and Agency for LLM-based Interactive Drama, restricting how much of the story script an agent can modify at any single turn. This mirrors the "yo-yo model" explored by Ka Hei Carrie Lau et al. in Wrapped in Anansi's Web: Unweaving the Impacts of Generative-AI Personalization and VR Immersion in Oral Storytelling, which allows for localized branching dialogue but periodically pulls the interaction back to fixed core plot points to preserve structural integrity.

Key Foundational Work

Dynamic, User-Centered Resolution in Interactive Stories (1997) — contributed 5 concept elements to this direction.
A Framework for Plot Control in Interactive Story Systems (1996) — contributed 4 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 58 papers
  2. Version 1 · 2026-10-04 · 56 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_agency_coherence_tradeoff_8c230e30,
  title = {agency-coherence tradeoff},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/agency-coherence-tradeoff/rdr_8c230e30fffb40ee810eaf5786463f63},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_agency_coherence_tradeoff_8c230e30_v2,
  title = {agency-coherence tradeoff},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/agency-coherence-tradeoff/rdr_8c230e30fffb40ee810eaf5786463f63/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake