Plotania: Exploring Transparency Trade-offs in AI Co-Writing Through Virtual Readers and Transparent Attribution

Paper Title

Plotania: Exploring Transparency Trade-offs in AI Co-Writing Through Virtual Readers and Transparent Attribution

Publication Info

• Topic area: Human-AI collaboration in creative writing
• Keywords: AI co-writing, creative agency, transparency, virtual readers, attribution visualization, audience awareness, parasocial relationships, narrative craft, human-AI interaction, creative tools

Background and Problem

• Problem / challenge: Current AI writing tools disrupt authorial control and audience awareness, essential elements of creative agency. Transparency mechanisms are often treated as separate concerns, failing to address their interdependence and potential trade-offs.
• Significance: Enhancing creative agency in AI-assisted writing is critical for empowering authors while maintaining authenticity and ownership in their work.
• Motivation and related work: Prior research has explored attribution tracking and audience feedback separately but has not examined their combined effects on creative agency. Existing systems lack mechanisms to balance transparency with creative empowerment, leaving gaps in understanding how these features interact.

Solution

• Proposed approach: Plotania, a co-writing system integrating virtual readers for real-time audience feedback and transparent attribution layers to track human vs. AI contributions.
• Novelty:
  1. Introduction of "agency-preserving transparency" principles that balance information provision with creative empowerment.
  2. Development of Virtual Readers calibrated to genre-specific conventions, providing structured and contextual feedback.
  3. Implementation of a transparent attribution system that tracks contributions at granular levels while addressing psychological impacts on authorship.
• Procedure and key techniques:
  • Transparent authorship tracking using a six-dimensional taxonomy of creative contributions.
  • Virtual Readers offering genre-calibrated feedback with controllable scope and intensity.
  • Flexible structural support tools like a negotiable plot canvas and project-persistent memory for long-context coherence.
  • Anti-formalization guardrails to preserve voice authenticity in AI-generated text.

Results

• Concrete findings:
  • Attribution visualization reduced AI usage and increased clarity of human vs. AI contributions (effect size d=0.87).
  • Virtual Readers enhanced audience awareness (+0.38) but modestly reduced perceived creative agency (-0.13).
  • Combined transparency features achieved complementary effects, modestly improving creative agency (+0.13).
• Advantage over baselines:
  • Full System (both attribution and virtual readers) achieved higher overall preference (80%) and writing quality ratings (70%) compared to individual features.
  • Virtual Readers fostered parasocial relationships and individualized skill development, offering emotional and cognitive support.
• Experiments / evaluation:
  • Controlled study with 20 participants using a 2×2 factorial design across four conditions: Baseline, Attribution-only, Reader-only, Full System.
  • Measures included writing productivity, subjective creative agency, audience awareness, cognitive load, and qualitative interviews.
  • Phase-dependent effects observed, with transparency mechanisms showing varied impacts across narrative stages.
• Limitations and future work:
  • Small sample size (N=20) limits statistical power for detecting small effects.
  • Laboratory setting may not fully capture long-term creative processes.
  • Attribution metrics focused on textual similarity rather than higher-order creative contributions, potentially misrepresenting authorship dynamics.
  • Future work should explore adaptive transparency systems, deeper AI partnerships, and multi-dimensional attribution frameworks.

Summary

Plotania addresses critical tensions in human-AI co-writing by integrating virtual readers and transparent attribution mechanisms. Controlled experiments reveal complex trade-offs: virtual readers enhance audience awareness but reduce perceived agency, while attribution visualization creates identity anxiety yet empowers editorial control. Combined features achieve complementary benefits, suggesting transparency requires careful orchestration rather than maximization. These findings contribute principles for agency-preserving transparency and highlight the need for adaptive systems that balance creative autonomy with collaborative support.