``Control Is a Trajectory, Not a Point'': Conceptualizing Control in Human-AI Co-Creativity

Paper Title

“Control Is a Trajectory, Not a Point”: Conceptualizing Control in Human-AI Co-Creativity

Publication Info

• Topic area: Human-AI collaboration in creative processes
• Keywords: Human-AI co-creativity, control dynamics, autonomy, initiative, authority, creative friction, trust, co-creative systems, MOSAAIC framework, design strategies

Background and Problem

• Problem / challenge: Control in human-AI co-creativity is underexplored, with existing frameworks failing to capture the dynamic, context-dependent nature of control distribution between humans and AI.
• Significance: Understanding control dynamics is critical for designing AI systems that act as co-equal creative partners, balancing human agency and AI autonomy to enhance creativity while maintaining ethical and practical considerations.
• Motivation and related work: Prior research has developed frameworks for human-AI co-creativity, such as MOSAAIC, but lacks empirical validation and practical insights from domain experts. Questions remain about how control should be conceptualized and distributed in co-creative processes.

Solution

• Proposed approach: A three-level conceptualization of control (micro, meso, macro) based on expert insights, coupled with actionable design strategies for co-creative systems.
• Novelty:
  1. Empirical insights from domain experts on control preferences in human-AI co-creativity.
  2. A refined conceptualization of control across three interrelated levels: micro (capabilities), meso (interaction dynamics), and macro (contextual factors).
  3. Actionable design guidelines for operationalizing control in co-creative systems.
• Procedure and key techniques:
  • Conducted a systematic literature review to identify the MOSAAIC framework as a theoretical probe.
  • Organized a semi-structured focus group with nine experts in HCI, AI, and co-creativity.
  • Analyzed qualitative data from design tasks and discussions using thematic analysis and Annotated Visual Analysis (AVA).
  • Synthesized findings into themes and proposed a conceptual model and design strategies.

Results

• Concrete findings:
  • Experts view control as dynamic, evolving across creative phases, and influenced by trust, context, and user preferences.
  • Participants emphasized the importance of balancing autonomy, initiative, and authority, with preferences for human oversight and AI autonomy in specific contexts.
  • Creative friction and feedback loops were identified as productive mechanisms for enhancing co-creation.
• Advantage over baselines:
  • The study bridges the theory-practice gap by grounding abstract control concepts in empirical insights, offering a nuanced, actionable framework.
• Experiments / evaluation:
  • Focus group study with nine experts, involving design tasks and semi-structured discussions.
  • Analysis yielded nine themes grouped into three categories: negotiating control, building control, and envisioning control.
• Limitations and future work:
  • Limited sample size and diversity; future studies should include creative practitioners and larger, more diverse populations.
  • Findings shaped by the MOSAAIC framework; exploring alternative models may yield additional insights.
  • The conceptual model lacks granular operational definitions for transitions between control levels; future research should investigate real-time interactions.

Summary

This paper explores control dynamics in human-AI co-creativity, revealing that control is dynamic, context-dependent, and evolves across creative phases. Using the MOSAAIC framework as a probe, the authors conducted a focus group study with domain experts, leading to a three-level conceptualization of control (micro, meso, macro) and actionable design strategies. Key findings include the importance of balancing autonomy, initiative, and authority, leveraging creative friction, and embedding feedback loops for trust and evaluation. The study bridges theoretical and practical gaps in the literature, providing foundational insights for designing adaptive, human-centered co-creative AI systems.