iLLuMinaTE: An LLM-XAI Framework Leveraging Social Science Explanation Theories Towards Actionable Student Performance Feedback

Students strongly prefer theory-informed explanations: When presented with iLLuMinaTE's explanations compared to traditional explainers, students preferred them 89.52% of the time, suggesting that grounding AI explanations in social science theories significantly improves their effectiveness in educational settings.

Social science theories make AI more understandable: The framework draws from eight different social science explanation theories (including concepts like "Abnormal Conditions" and "Contrastive Explanation") to structure how AI insights are presented, demonstrating that educational psychology and communication principles can bridge the gap between complex AI systems and non-technical users like students and teachers.

Actionability matters as much as accuracy: The research included a novel "actionability simulation" with 114 real students, indicating that responsible AI in education isn't just about explaining what happened with student performance, but ensuring explanations help students understand what to do about it—a critical distinction for practical impact.

Multiple AI models and methods can work together: The framework was tested across three different large language models (GPT-4o, Gemma2-9B, Llama3-70B) and three XAI methods (LIME, Counterfactuals, MC-LIME), suggesting that responsible AI solutions don't require choosing one "perfect" approach but can combine different technologies while maintaining understandability.

Ready-to-use frameworks reduce implementation barriers: By providing a "zero-shot, chain-of-prompts" pipeline, iLLuMinaTE offers educators a practical tool they can deploy without extensive AI expertise, addressing a key challenge in responsible AI: making ethical, explainable systems accessible to real-world users.