Proactive AI as a Catalyst for Creativity? Balancing Human Agency and AI Contribution in Collaborative Story Writing

Paper Title

Proactive AI as a Catalyst for Creativity? Balancing Human Agency and AI Contribution in Collaborative Story Writing

Publication Info

• Topic area: Proactive AI in collaborative creative writing
• Keywords: Proactive AI, human-AI collaboration, creative writing, story outlining, human agency, AI contribution, Wizard-of-Oz, creativity support, AI-augmented inspiration, user autonomy

Background and Problem

• Problem / challenge: Proactive AI systems in collaborative writing face challenges in balancing human agency with AI contributions. Excessive AI involvement risks diminishing user autonomy, while insufficient support may fail to enhance creativity or productivity.
• Significance: Addressing this balance is critical to designing AI systems that effectively support human creativity without undermining the user’s sense of ownership and control.
• Motivation and related work: Previous studies have explored AI-assisted writing and proactive AI in various contexts, but the impact of proactive AI on user creativity and agency during story outlining remains underexplored. This paper investigates how different proactive suggestion styles influence these dynamics.

Solution

• Proposed approach: A Wizard-of-Oz study simulating two proactive AI suggestion styles—intrusive suggestions (next-sentence completions) and non-intrusive suggestions (exploratory proposals)—to evaluate their effects on user creativity, agency, and writing experience.
• Novelty:
  1. Introduction of the concept and metric of AI-augmented inspiration to quantify creativity gains indirectly stimulated by AI.
  2. Empirical comparison of intrusive and non-intrusive suggestion styles in collaborative story outlining.
  3. Design guidelines for proactive AI systems to balance creativity support and user autonomy.
  4. Insights into multimodal user signals and their role in guiding AI interventions.
• Procedure and key techniques:
  • Conducted a within-subject Wizard-of-Oz study with 30 participants (15 user-wizard pairs).
  • Simulated two suggestion modes: Completion Mode (intrusive) and Proposal Mode (non-intrusive).
  • Collected data on writing speed, creativity gain, AI contribution, perceived autonomy, and satisfaction.
  • Analyzed multimodal user signals (e.g., facial expressions, writing activity) and wizard strategies for proactive intervention.

Results

• Concrete findings:
  • Proactive AI suggestions accelerated recovery from creative blocks, increasing writing speed by 1.2x during stuck phases (p = 0.006).
  • Proposal mode enhanced perceived autonomy and ownership but required higher cognitive effort, while completion mode improved fluency but reduced autonomy.
  • AI-augmented inspiration mitigated the trade-off between AI contribution and user autonomy, enabling high creativity gains without diminishing control.
  • Intermediate levels of AI contribution yielded the highest user satisfaction.
• Advantage over baselines:
  • Proposal mode encouraged active user engagement and skill development, while completion mode supported faster and more fluent writing.
  • The AI-augmented inspiration metric provided a novel way to quantify indirect creativity stimulation, addressing limitations of prior qualitative measures.
• Experiments / evaluation:
  • Writing sessions lasted 45 minutes under each mode, followed by surveys and interviews.
  • Metrics included typing speed, creativity gain (via KL divergence), AI contribution (semantic similarity), and subjective ratings (autonomy, satisfaction).
  • 486 AI suggestions analyzed across 30 writing sessions.
• Limitations and future work:
  • The study used a Wizard-of-Oz setup rather than a fully autonomous AI system, limiting real-world applicability.
  • Latency in AI suggestions disrupted user experience; future work should optimize response times and context alignment.
  • Further exploration of interaction designs and proactive suggestion strategies is needed.

Summary

This study examines the role of proactive AI in collaborative story writing, comparing intrusive and non-intrusive suggestion styles. Findings reveal that proactive AI can accelerate writing and enhance creativity but must balance AI contribution with user autonomy. The introduction of the AI-augmented inspiration metric highlights how indirect AI stimulation fosters creativity without undermining control. Proposal mode supports skill development and autonomy, while completion mode enhances fluency. Design guidelines and insights into multimodal user signals provide a foundation for future proactive AI systems that respect human agency while catalyzing creativity.