Draw2Learn: A Human-AI Collaborative Tool for Drawing-Based Science Learning

Drawing supports learning by externalizing mental models, but providing timely feedback at scale remains challenging. We present Draw2Learn, a system that explores how AI can act as a supportive teammate during drawing-based learning. The design translates learning principles into concrete interaction patterns: AI generates structured drawing quests, provides optional visual scaffolds, monitors progress, and delivers multidimensional feedback. We collected formative user feedback during system development and open-ended comments. Feedback showed positive ratings for usability, usefulness, and user experience, with themes highlighting AI scaffolding value and learner autonomy. This work contributes a design framework for teammate-oriented AI in generative learning and identifies key considerations for future research.

💡 Research Summary

Draw2Learn is a novel human‑AI collaborative system that supports science learning through drawing. The authors ground their design in multiple learning theories—goal‑setting theory, Vygotsky’s Zone of Proximal Development, cognitive‑load theory, formative‑feedback theory, growth‑mindset theory, and self‑determination theory—to address three interrelated challenges: (1) sustaining learner engagement during cognitively demanding drawing tasks, (2) providing adaptive scaffolding without undermining learner agency, and (3) delivering encouraging, multidimensional feedback rather than judgmental evaluation.

To operationalize these principles, the system implements four concrete interaction patterns. First, it generates structured “drawing quests” that decompose a learner‑provided learning goal into a sequence of tasks aligned with Bloom’s taxonomy, thereby turning abstract concepts into manageable sub‑goals. Second, it offers optional visual scaffolds—draggable SVG helper objects—that learners can request at any time, reducing extraneous visual‑motor load while preserving germane cognitive processing. Third, an AI “teammate” continuously monitors the canvas using a vision‑language model, producing feedback cards that address motivational, cognitive, metacognitive, and self‑relevant dimensions. The feedback is phrased in collaborative language (“we could try…”) to reinforce the teammate metaphor. Fourth, the interface rewards task completion with gem icons and, after the full quest, allows learners to apply artistic style transfers (e.g., oil painting, watercolor) to amplify a sense of accomplishment.

The user interface consists of three panels: a left “Quest” panel showing task progression and scaffold buttons, a central drawing canvas with tool palettes and style options, and a right “AI Tutor” panel displaying the avatar and color‑coded feedback cards. The AI’s periodic analysis of the canvas informs real‑time feedback and progress tracking.

During development, six participants completed drawing tasks on topics such as photosynthesis, the water cycle, and cell structure. Quantitative Likert‑scale results (1–7) showed high attractiveness (M = 6.17), positive affect (M = 5.83), overall experience (M = 6.15), ease of use (M = 5.67), and usefulness (M = 5.67). Social bond (M = 4.67) and reuse intent (M = 4.17) were lower, indicating room for strengthening the perceived partnership. Qualitative thematic analysis identified five strengths—effective AI scaffolding, deeper conceptual processing, engaging quest structure, clear guidance, and preserved autonomy—and five improvement areas: performance optimization, support for mouse‑free or voice input, clearer AI evaluation criteria, adaptive scaffolding for varied prior knowledge, and interface refinements (canvas size, tab smoothness).

The authors acknowledge limitations: a small, informal sample, lack of direct learning‑outcome measurement, and absence of classroom‑scale validation. They propose future work involving controlled experiments to assess knowledge gains, deployment in authentic educational settings, and extension of the teammate paradigm to other generative tasks such as writing or coding.

Key contributions include (1) introducing a teammate‑oriented interaction paradigm that reframes AI from a directive tutor to a collaborative supporter, (2) systematically translating abstract learning theories into concrete design patterns (quest generation, optional scaffolds, multidimensional feedback), and (3) providing preliminary empirical evidence that these patterns are perceived as intended by learners. This work offers a practical roadmap for designers of AI‑enhanced generative learning tools, emphasizing the balance between supportive guidance and learner agency.