Bridging LLMs and Symbolic Reasoning in Educational QA Systems: Insights from the XAI Challenge at IJCNN 2025

Explainability in education requires hybrid approaches: The challenge demonstrated that combining Large Language Models with symbolic reasoning systems can produce AI systems that answer questions while providing clear, logic-based explanations—addressing a critical need as AI becomes more integrated into educational settings.

Dataset quality matters for trustworthy AI: The researchers constructed their dataset using logic-based templates validated with Z3 (a formal verification tool) and refined through expert student review, showing that rigorous, multi-step validation processes are essential for creating datasets that reflect real-world educational scenarios and support responsible AI development.

Hackathons can advance responsible AI beyond rapid prototyping: This competition represents a shift in how hackathons function—moving beyond speed-focused development to directly address explainability and trustworthiness in real-world contexts, suggesting that competition formats can be effective venues for advancing responsible AI practices.

Lightweight models support transparency goals: By requiring participants to use lightweight LLMs or hybrid systems rather than large proprietary models, the challenge promoted solutions that are more interpretable and potentially more accessible, aligning technical constraints with responsible AI principles.

Student feedback validates AI system design: The inclusion of expert student review in dataset refinement ensured the QA system addressed genuine academic needs, demonstrating that end-user input is crucial for building AI systems that are both explainable and practically useful in educational contexts.