When Teams Embrace AI: Human Collaboration Strategies in Generative Prompting in a Creative Design Task

Title of the Paper

When Teams Embrace AI: Human Collaboration Strategies in Generative Prompting in a Creative Design Task

Paper Information

• Domain: Human-AI collaboration, generative AI cooperation in creative design tasks
• Keywords: Human-AI collaboration, team co-prompting, GenAI engineering, collaborative creative design, prompt interaction

Research Background and Problem

• Problem: Current research primarily focuses on how individuals use generative AI to complete creative tasks, but there is limited exploration of how people from diverse backgrounds collaborate within teams to prompt AI systems. Furthermore, it remains unclear how designers overcome the challenges of generative AI prompting in team environments.
• Significance: Generative AI demonstrates significant potential in enhancing creativity and supporting team decision-making and consensus-building. However, effectively integrating this technology into team collaboration requires deeper exploration.
• Motivation and Related Work: This study draws inspiration from collaborative practices in multidisciplinary creative tasks (e.g., stage design) to investigate how team members express creativity through co-prompting generative AI. The research also examines perceptions of generative AI tool efficiency and team dynamics.

Solution

• Methodology:

  1. Define "co-prompting": Involves two or more individuals sharing generative AI system prompts and collaboratively exploring content.
  2. Research design: Conduct online workshops where participants work in pairs to design stage art using Midjourney and ChatGPT to generate stage design sketches.
  3. Data collection: Utilize semi-structured interviews and process observations, analyzing data through qualitative thematic analysis.

• Innovations:

  • Explore the dynamic relationships in team collaborative prompting and how they facilitate or hinder creative expression.
  • Provide insights into designing creative collaboration systems for multi-user environments.

• Implementation Steps and Key Techniques:

  1. Recruit participants, randomly grouping them into pairs, totaling 18 individuals.
  2. Workshops consist of three phases: familiarization with generative AI tools, creative design based on the poem "Do Not Go Gentle Into That Good Night," and post-session interviews reflecting on the work experience.
  3. Observe how teams define prompt content and iteratively refine generated results, studying the impact of co-prompting on collaboration.

Research Findings

• Specific Outcomes:

  1. Two overarching collaboration strategies:
     • Narrative-based workflow: Construct the stage's performance storyline first, then fill in details and generate images.
     • Concept-based workflow: Develop the stage's core concept or theme first, then generate images and refine the storyline.
  2. Co-prompting helps teams resolve prompt-related challenges and fosters mutual understanding among team members.

• Advantages:

  • Generative AI tools are perceived as neutral third parties, enabling teams to quickly experiment and validate creative ideas.
  • Co-prompting reduces psychological experimentation burdens and provides space for iterative creativity.

• Experimental or Evaluation Results:

  1. While co-prompting facilitates creative testing, teams face communication challenges when adjusting generated content.
  2. AI outputs often reflect literal meanings of words, potentially lacking emotional or implicit intent, leading to repeated iterations.
  3. Although participants sought inspiration from AI, they still prioritized suggestions from human team members.

• Limitations and Future Directions:

  1. Limitations:
     • Experimental results may be constrained by the online team collaboration setting.
     • Participants were mostly students lacking professional stage design backgrounds, limiting generalizability.
     • Task scope was confined to poetry-inspired creative design, not encompassing other task types.
  2. Directions:
     • Develop generative AI systems adaptable to different design stages, such as providing diverse results during brainstorming and precise drawing capabilities during concept validation.
     • Consider designing AI models for cross-language and cross-cultural contexts to address language comprehension issues.
     • Conduct studies with diverse populations to observe how different groups adapt to generative AI in team settings.