Towards a Learner-Centered Explainable AI: Lessons from the learning sciences

Explainable AI should prioritize helping people learn, not just providing information. The paper argues that current XAI approaches often focus on technical accuracy rather than whether explanations actually help users understand and learn from AI systems—a distinction that matters significantly for real-world effectiveness.

Learning sciences research offers practical design principles for XAI. By drawing on established theories about how people learn, the framework suggests that XAI designers should consider factors like cognitive load, prior knowledge, and active engagement—principles already proven effective in educational contexts.

One-size-fits-all explanations are insufficient. The learner-centered approach implies that explanations should be tailored to different users' backgrounds and learning needs, rather than providing identical explanations to everyone, which could improve how well people actually understand AI decisions.

Real-world applications like social work demonstrate the stakes of poor explanations. The case study in AI-augmented social work illustrates that when AI systems affect vulnerable populations, explanations must genuinely support human decision-making and learning, not just satisfy technical requirements.

Evaluation of XAI systems should measure learning outcomes, not just explanation quality. Rather than assessing whether explanations are technically correct, the framework suggests testing whether users actually understand the AI's reasoning and can make better-informed decisions as a result.