LADICA: A Large Shared Display Interface for Generative AI Cognitive Assistance in Co-located Team Collaboration

Research Background and Issues

• What problems or challenges did the authors identify?
  Co-located team collaboration faces challenges in cognitive and social motivation, such as difficulties in knowledge sharing, lack of mutual understanding of progress, and challenges in synchronizing discussions with external content. Additionally, collaborative tools involving generative AI often focus on direct functionalities while neglecting how to optimize traditional human practices and complex team cognitive processes. Existing tools may lead to excessive reliance on AI by team members, potentially limiting interpersonal communication and creativity, which are essential to collaboration.

• Why is this issue important?
  Co-located collaboration (e.g., brainstorming, knowledge construction, and planning) requires frequent interpersonal communication and real-time interaction. Cognitive and social barriers often undermine team effectiveness, and current AI designs for collaborative environments fail to balance the roles of humans and AI, thus not fully leveraging the advantages of team collaboration.

• Research Motivation and Related Work
  The authors aim to improve cognitive support in team collaboration through generative AI while maintaining human dominance. Related work includes leveraging large language models (LLMs) to optimize collaborative tools (e.g., Miro, Lucidspark) and provide cognitive support functionalities (e.g., semantic grouping, keyword expansion). However, these tools rarely focus on the cognitive processes underlying team dynamics, resulting in superficial support that fails to deeply engage with the core cognitive processes of teams.

Solution

• What methods or solutions did the authors propose?
  The authors proposed a system called LADICA (Large Display Interface with Cognitive Assistance), an AI-enhanced shared display interface. The system employs a three-layer structure to provide cognitive assistance for team brainstorming, grouping analysis, and discussion processes. The three layers are:

  1. Idea Repository: For storing and expanding ideas during the initial stages of team collaboration.
  2. Affinity Lens: Helps organize and analyze shared ideas from multiple perspectives, comparing and grouping information.
  3. Discussion Reference: Facilitates discussions based on existing content and generates new insights.

• What are the innovative aspects of this solution?
  LADICA links team members' personal devices with shared displays through intelligent synchronization and combines generative AI to provide dynamic cognitive support, enhancing teams' meta- and macro-cognitive processes (e.g., understanding task structures and knowledge construction). Furthermore, the system ensures AI assistance does not dominate discussions or override human collaborative dynamics, while promoting inclusivity and diversity in opinion expression.

• What are the implementation steps and key technologies used?

  1. Three-layer support architecture:

     • Idea Repository: Supports task decomposition, idea expansion based on queries and relationships, and key extraction from real-time discussion information.
     • Affinity Lens: Automatically generates multi-perspective analysis groups, tracks the evolution of different groupings, and allows users to return to previous content snapshots.
     • Discussion Reference: Generates discussion prompts for teams, reveals relationships between ideas, and retrieves relevant existing content from real-time discussions.

  2. Technological applications:

     • Utilizes LLMs (e.g., OpenAI GPT-4 Turbo) to generate and optimize cognitive prompts, supporting relationship hints and information grouping across users.
     • Integrates real-time collaboration tools (e.g., yjs library) to achieve synchronized editing between personal devices and shared displays.
     • Incorporates voice recognition technology (e.g., OpenAI Whisper) for real-time transcription and key point extraction from discussion content.

Research Outcomes

• What specific outcomes were achieved?

  • LADICA successfully addressed cognitive and organizational challenges in team collaboration, such as helping members quickly generate and expand ideas, enhancing shared mental models within teams, and effectively synchronizing external content and memory during discussions.
  • User studies revealed that participants found the system highly effective in facilitating team cognitive processes (e.g., idea expansion, grouping analysis, and mutual discussion). Participants particularly appreciated the system's diverse analytical perspectives and dynamic collaboration features.

• What advantages does it have compared to existing solutions?

  • Compared to existing AI-generated collaboration tools, LADICA focuses more on assisting rather than replacing human-led team collaboration processes. Through its three-layer structure, it not only provides deep cognitive support based on LLMs but also ensures a balance between collaborative dynamics and inclusivity, making it better suited for long-term team interaction needs.

• What were the experimental or evaluation results?

  • Experiments showed that 14 participants successfully used LADICA for co-located team tasks and provided positive feedback. Average ratings indicated that LADICA performed well in terms of perceived usability, effectiveness, and support for collaboration.
  • Interaction log analysis demonstrated that users primarily utilized system functionalities during brainstorming, information grouping, and discussion phases. The system facilitated connections and coordination among team members.

• Limitations and Future Directions

  • Technical limitations: Suggestions generated by LLMs were sometimes overly generalized or irrelevant, and response delays occurred when multiple users operated simultaneously.
  • Research scope: All participants were university students, which may not reflect different needs in other environments. Additionally, the study only involved two task types: travel planning and policy discussion.
  • Improvements and extensions: Future plans include enhancing LADICA's dynamic generation capabilities, testing across diverse scenarios, and expanding multimodal interaction features (e.g., facial expression and body movement recognition).

In summary, LADICA significantly enhances the cognitive support capabilities of shared display interfaces in team collaboration through generative AI and provides a foundation for exploring more complex collaborative scenarios.