Social Dynamics of Human-AI Collaboration in Creative Writing

Title of the Paper

Social Dynamics of AI Support in Creative Writing

Paper Information

• Subject Area: Social Dynamics of AI and Creative Writing
• Keywords: Creative Writing, Writing Support Tools, Human-AI Collaboration, Language Models, Human-Computer Interaction

Research Background and Problem

1. Problems or Challenges Identified by the Authors:

   • Large language models have made significant progress in generating coherent and creative text, but there is limited understanding of the socio-technical gap when these models collaborate with users.
   • Investigating when and why writers choose to seek support from computers rather than peers or mentors.

2. Significance:

   • Creative writing is a cornerstone of human culture, and studying its social dynamics with AI collaboration can influence broader complex tasks and technological practices.

3. Research Motivation and Related Work:

   • Drawing from cognitive psychology and tools for creative support, the study explores how large language models function as writing support tools.
   • Builds on cognitive process models of writing and existing frameworks for artist-supporter relationships.

Solution

1. Proposed Methodology:

   • Conduct interviews with 20 creative writers to analyze their attitudes toward human and computer support.
   • Delve into the decision-making basis and values behind writers seeking support.

2. Innovative Aspects:

   • Proposes a social dynamics model linking writers' needs, perceptions of supporters, and values during the writing process.
   • Extends existing models of artist-supporter collaboration types.

3. Implementation Steps:

   • Conduct semi-structured interviews with 20 writers and use qualitative analysis methods to extract key themes.
   • Organize data using multiple analytical frameworks to construct a taxonomy of support dynamics.
   • Provide design guidelines and propose directions for future research.

Research Findings

1. Specific Findings:

   • Identified three high-level categories explaining writers' decision-making criteria for seeking external help:

     • Writers' Needs: Including planning, translation (turning ideas into concrete expressions), reviewing, and motivation.
     • Perceptions of Supporters: Including availability, personalization, and trustworthiness of the supporter.
     • Writers' Values: Intent, authenticity, and perspectives on creativity.

   • Proposed an extended model describing the external and internal dynamics of support relationships.

2. Comparison with Existing Solutions and Advantages:

   • Unlike studies focusing solely on AI technical performance, this work deeply explores writers' decision-making and values within social dynamics.
   • Offers a socio-technical perspective to guide the design of writing support tools.

3. Experimental or Evaluation Results:

   • Interviews revealed that writers' attitudes toward AI assistance vary based on personal values, experiences, and creative writing goals.
   • Writers preferred deep interaction with human supporters but showed higher acceptance of AI tools for "convenience" and "initial feedback."

4. Limitations and Future Directions:

   • Limitations:

     • Many interviewed writers had limited experience with AI writing tools, so their feedback might be more hypothetical.
     • Participants were primarily from a U.S. cultural context, affecting the generalizability of results.
     • The focus on amateur writers with limited experience may bias the findings.

   • Future Directions:

     • Explore how AI writing tools can understand writers' intentions and provide customized support.
     • Study writers' mental models of AI systems and how these evolve over time.
     • Expand the analysis of social dynamics to global writer communities and examine the impact of different cultural contexts on AI collaboration.

Design Recommendations

• Personalized Writing Support: Provide clear information about system performance and limitations, such as training data and specific functionalities.
• Intent-Based Customized Support: Enable systems to understand or help writers clarify their intentions to offer more relevant suggestions.
• Focus on Authenticity: Design AI tools that align more closely with writers' styles and expressions over time, reducing cognitive and usability conflicts.

Conclusion

• This study reveals the complex social dynamics involved when writers seek support, addressing a research gap in the socio-technical aspects of AI writing tools.
• The proposed taxonomy and model not only provide valuable guidance for designing AI writing tools but also offer profound insights into the diverse dynamics of human-AI collaboration.