Thinking in Graphs with CoMAP: A Shared Visual Workspace for Designing Project--Based Learning

Paper Title

Thinking in Graphs with CoMAP: A Shared Visual Workspace for Designing Project-Based Learning

Publication Info

• Topic area: Human-AI collaboration in educational design
• Keywords: Project-based learning, instructional design, distributed cognition, graph-based collaboration, human-AI interaction, visual workspace, dual-modality AI, cognitive scaffolding, iterative design, teacher support

Background and Problem

• Problem / challenge: Existing tools for project-based learning (PBL) design are limited by linear workflows and conversational AI systems that lack persistent shared context, making it difficult for educators to manage the non-linear and iterative nature of PBL.
• Significance: Effective PBL design is crucial for fostering high-order skills like problem-solving and collaboration in students, but current tools fail to adequately support teachers in navigating the complexity of iterative instructional design.
• Motivation and related work: Prior research has explored theoretical frameworks, workflow tools, and digital environments for PBL design, but these approaches often impose rigid linearity, require precise natural language articulation, or focus more on learner support than teacher scaffolding. A gap remains in providing tools that enable shared workspaces and iterative localization for teachers.

Solution

• Proposed approach: CoMAP, a graph-based human-AI collaborative tool for PBL design, leveraging distributed cognition principles and dual-modality AI interaction.
• Novelty:
  1. Introduction of a graph-based paradigm for human-AI co-design, externalizing non-linear cognitive processes into a shared visual workspace.
  2. Development of the CoMAP system, which uses the ASSURE instructional design model to structure PBL components into an interactive graph.
  3. Empirical validation through a mixed-methods study (n=30), demonstrating improved design expression, reduced cognitive load, and enhanced trust in human-AI collaboration.
• Procedure and key techniques:
  • Structured graph canvas as a shared cognitive artifact for externalizing ideas and tracking interdependencies.
  • Dual-modality AI interaction: a global conversational agent for high-level ideation and local GUI-integrated agents for fine-grained refinement.
  • Fluid interaction design supporting non-linear exploration and iterative alignment.
  • Export options for bridging digital and traditional workflows.

Results

• Concrete findings:
  • CoMAP significantly improved perceived design expression (M = 6.13 vs. 4.15, p < .001, d = 1.11) and understanding (M = 5.77 vs. 4.29, p < .001, d = 1.16) compared to a dialogue-based AI baseline.
  • Reduced cognitive load (M = 5.67 vs. 4.74, p = .006, d = 0.62), increased collaboration (M = 5.72 vs. 4.49, p = .007, d = 0.72), and enhanced trust (M = 5.55 vs. 4.57, p = .001, d = 0.78).
• Advantage over baselines:
  • Higher controllability and transparency, fewer negative keywords, and reduced conversational overhead (e.g., fewer chat turns: M = 11.82 vs. 18.64, p < .001, d = 1.11).
• Experiments / evaluation:
  • Mixed-methods user study with 30 participants using a within-subjects crossover design.
  • Quantitative measures included perceived expression, understanding, human-AI interaction experience, and behavioral metrics.
  • Qualitative thematic analysis of interviews and interaction logs.
• Limitations and future work:
  • Limited generalizability due to participant selection (moderately experienced educators).
  • Multi-factor design prevents attribution of effects to specific features.
  • Dependency on GPT-4.1 API introduces scalability and latency challenges.
  • Future directions include exploring alternative instructional frameworks, extending the paradigm to other domains, supporting multi-user collaboration, and integrating teacher learning support.

Summary

CoMAP introduces a graph-based paradigm for human-AI collaboration in project-based learning design, addressing the limitations of linear tools and conversational AI systems. By externalizing non-linear cognitive processes into a shared visual workspace and leveraging dual-modality AI interaction, CoMAP enhances design expression, understanding, and iterative refinement. Empirical validation demonstrates significant improvements in user experience, cognitive load reduction, and trust. While the study highlights the potential of shared visual representations and integrated AI guidance, future work is needed to generalize findings, optimize system scalability, and explore applications in other design domains.