AI as Social Glue: Uncovering the Roles of Deep Generative AI during Social Music Composition

Title of the Paper

AI as Social Glue: Uncovering the Roles of Deep Generative AI During Social Music Composition

Paper Information

• Research Domain: Human-Computer Interaction, Generative Artificial Intelligence, and the social dynamics of collaborative music creation
• Keywords: Deep Generative AI, Human-AI Collaboration, Music Composition, Social Dynamics, Psychological Safety, Collaborative Technology

Research Background and Problem Statement

• Problems and Challenges:
  • Individuals face difficulties when collaborating with generative AI to create content (e.g., music composition), such as a lack of control over the creative process.
  • Limited research has explored the impact of generative AI on social dynamics in multi-person collaborations, which often involve interpersonal interactions.
• Significance:
  • Social creative practices (e.g., duet music composition) not only foster creativity but also enhance social connections.
  • Understanding how AI influences interpersonal collaboration dynamics is crucial for designing future AI tools.
• Motivation and Related Work:
  • Machine learning and deep generative networks have expanded the capabilities of music generation, but their impact on social collaborative creative processes remains underexplored.
  • A literature review reveals that while AI's role in supporting individual creation and production performance has been studied, its influence on social dynamics in interpersonal collaborations is insufficiently addressed.

Solution

• Methods and Research Design:
  • A qualitative experimental study analyzing the social dynamics of 15 pairs (30 participants in total) collaborating to compose short musical phrases, with and without generative AI assistance.
  • Utilization of the "Cococo" tool, a collaborative music editor based on the deep generative music model Coconet, allowing participants to use AI to complete musical fragments and adjust the quality of generated music.
• Innovations:
  • Investigation of the social roles of generative AI in multi-person creative processes, including psychological safety, progress facilitation, and conflict reduction.
  • Introduction of an AI-generated "multi-option interface" enabling users to select and adjust music styles.
• Implementation Steps and Key Techniques:
  • Participants were divided into two conditions (with AI and without AI) for music composition, with each group provided a mood prompt based on images.
  • Semantic analysis was employed to uncover AI's specific roles in advancing creative progress and facilitating social interaction.

Research Findings

• Specific Findings:
  • Five major roles of generative AI in collaboration were identified:
    1. Establishing Common Ground: AI-generated content serves as a starting point for discussion, helping participants quickly reach consensus.
    2. Psychological Safety Net: AI reduces creative risks, allowing participants to experiment boldly without fear of judgment.
    3. Progress Facilitator: AI can rapidly generate content, helping participants overcome creative blocks and save time.
    4. Social Lubricant: AI provides options that reduce interpersonal conflicts, acting as a "third-party" object of critique.
    5. Role Shift: Humans transition from creators to "supervisors," reducing direct hands-on creative involvement.
• Experimental or Evaluation Results:
  • AI played a significant role in accelerating creative processes and generating new ideas, particularly in handling complex music theory or highly repetitive musical segments.
  • Participants adapted to AI-generated content by using the options provided to supplement musical structures or solve specific challenges.
• Comparison with Existing Solutions:
  • AI not only supports individual creation but also acts as a lubricant and facilitator in multi-person collaborations.
• Limitations and Future Directions:
  • The training data for AI-generated content constrained creative styles, making it difficult to break out of specific frameworks (e.g., Bach-style compositions).
  • Some advanced music creators found that AI increased decision-making complexity, which requires further optimization.
  • Future research could explore designing AI systems that support deeper interpersonal interactions and expand AI's adaptability to diverse artistic styles.

Conclusion

This study reveals the unique roles of generative AI in collaborative music creation and proposes potential design principles for AI as a social lubricant. Future work should focus on designing AI systems to support richer and deeper collaborative interactions while enhancing the diversity and freedom of AI-generated creations.