Combine DGBL with AI System: A Technical Guidance to Reduce Teacher's Burden in Digital Game-Based Learning

This paper proposes a technical guidance system that combines a large language model with blockchain technology to reduce the workload teachers face when designing digital game-based learning (DGBL). The core premise is that building game-based lessons demands substantial time for selecting games, estimating difficulty and time investment, and evaluating whether students have met learning goals after play, and that an integrated AI-supported system can ease these demands across the full instructional cycle, from pre-class planning to post-class learner profiling.

The proposed architecture is organized into three layers: a user layer serving teachers and students, an application layer that performs case matching, follow-up evaluation, and suggestion generation, and a technical layer combining an LLM built on transformer encoder-decoder components with blockchain storage. Teachers enter instructional plans and students create learner profiles, which are recorded on the blockchain for data integrity. The LLM then uses this information to match appropriate DGBL cases with a track record in similar scenarios, support the design of personalized learning blueprints, and generate questionnaires and automated assessments of student performance, with learner portraits updated as new data arrives. The blockchain layer is described as using private, public, and consortium chains to protect data, though the paper provides no implementation, empirical testing, or quantitative evaluation of the system's effectiveness.

The authors acknowledge several limitations. LLMs process tokens in parallel rather than modeling sequential dependencies, which can produce inconsistent responses, and they may generate plausible but incorrect content, a problem the paper links to the "cueing problem." The quality of outputs also depends heavily on training data and developer pre-training effort, which may offset the intended reduction in teacher burden. The authors conclude that future work should develop domain- and segment-specific LLM systems to better support teachers implementing DGBL.