Where Do I 'Add the Egg'?: Exploring Agency and Ownership in AI Creative Co-Writing Systems

Paper Title

Where Do I 'Add the Egg'?: Exploring Agency and Ownership in AI Creative Co-Writing Systems

Publication Info

• Topic area: Human-computer interaction in AI creative writing systems
• Keywords: AI co-writing, agency, ownership, interface metaphors, creative writing, human-AI collaboration, tool-like interfaces, agent-like interfaces, magic-like interfaces

Background and Problem

• Problem / challenge: Writers often struggle with perceptions of agency and ownership when using AI co-writing systems, which hinders adoption and satisfaction.
• Significance: Understanding agency and ownership is critical for designing systems that align with writers’ creative processes and foster long-term use.
• Motivation and related work: Prior research has explored AI co-writing systems and their impact on writing quality, ownership, and collaboration. However, little is known about how interface metaphors influence writers’ conceptualizations of agency and ownership.

Solution

• Proposed approach: Development of three AI co-writing prototypes based on distinct interface metaphors: agent-like, tool-like, and magic-like.
• Novelty:
  1. Identification of three archetypes of interface metaphors in commercial AI co-writing systems.
  2. Creation of a taxonomy of agency and ownership types in AI co-writing.
  3. Design recommendations for AI systems based on interface metaphor affordances.
• Procedure and key techniques:
  • Conducted a review of 12 commercial AI co-writing systems to identify prevalent interface metaphors.
  • Developed three prototype systems (Wordsworth, Autowrite, Magic Quill) with consistent functionality but distinct metaphorical framings.
  • Conducted an exploratory study with 18 participants (professional and non-professional writers) using semi-structured interviews to analyze conceptions of agency and ownership.

Results

• Concrete findings:
  • Identified three types of agency: explicit, implicit, and writing process agency.
  • Identified three types of ownership: stylistic, conceptual, and effort-based ownership.
  • Agent-like interfaces fostered conceptual ownership, while tool-like interfaces emphasized explicit control.
• Advantage over baselines: The study revealed nuanced interactions between interface metaphors and perceptions of agency and ownership, providing insights not captured by prior frameworks.
• Experiments / evaluation:
  • Participants used one of three prototype systems for 15 minutes followed by interviews.
  • Analysis involved open coding and thematic analysis of transcripts to identify recurring themes.
• Limitations and future work:
  • Limited functionality in prototypes may have attenuated metaphorical effects.
  • Magic-like interfaces did not consistently influence agency or ownership perceptions.
  • Future work should explore longitudinal studies, diverse cultural contexts, and more ecologically valid systems.

Summary

This paper investigates how interface metaphors in AI co-writing systems influence writers’ perceptions of agency and ownership. Through a review of commercial systems and an exploratory study with 18 participants, the authors identified three archetypes of interface metaphors (agent-like, tool-like, magic-like) and developed a taxonomy of agency and ownership types. Findings suggest that agent-like interfaces foster conceptual ownership, while tool-like interfaces emphasize explicit control. Design recommendations are provided to align interface metaphors with user expectations and creative processes. This work lays the foundation for designing AI systems that better support human-AI collaboration in creative writing.