Preparing the eLuna Visual Language for Human-AI Co-Specification and Code Automation

Abstract

This paper presents the application and refinement of the eLuna Framework. The framework includes a two-phase collaborative method for designing and specifying narrative game-based learning (GBL) systems using a visual language. Researchers tested the framework through a real-world project that developed the Super Climate Model Game, a mobile game that explores learning objectives in climate science. Participatory action research and heuristic usability inspection methods were used to evaluate the eLuna Visual Language. The goal was to determine whether it could create unambiguous and expressively complete blueprints for narrative GBL development, and subsequently to rectify uncovered ambiguities and expressive shortcomings to enable future AI agent collaboration and code automation in development of narrative GBL using eLuna. Confirming previous research, the eLuna co-design phase was found to be highly usable. It effectively structured narrative GBL elements, supported collaboration between educators and developers, and ensured compliance with characteristics enforcing positive learning outcomes. However, the visual language lacked the expressive completeness and unambiguity required for AI agent collaboration and automated code generation. Seven modifications were proposed to address these issues while preserving the framework’s empowering co-design properties alongside its originally targeted characteristics that are associated with positive effects on engagement, motivation, and learning. These seven changes improve human readability and computational parsing. They also bring the visual language closer to supporting automated coding processes. The Super Climate Model Game was developed using a hybrid approach that combined eLuna and SCRUM. This approach was necessary due to real-world constraints. The project demonstrated the flexibility of the eLuna Method and its relevance in practical development scenarios. This research moves the eLuna Visual Language closer to enabling multiagent human-AI collaboration and automated code generation. Future work will focus on creating an online visual editor for collaboration between human and AI agents. A semantic parser will also be developed to translate eLuna blueprints into structured data formats for automated code generation. These advancements will support scalable and effective narrative GBL development, delivering positive learning outcomes to diverse learner demographics.