Using Learning Theories to Evolve Human-Centered XAI: Future Perspectives and Challenges

Explanations should be designed around how humans learn, not just technical accuracy. The paper argues that learning theories should be integrated into the XAI lifecycle, suggesting that effective AI explanations need to consider cognitive science principles rather than simply making AI systems more transparent.

Current XAI approaches may not adequately address human agency and decision-making. A learner-centered approach to XAI can enhance human agency and help mitigate risks associated with AI explanations, implying that many existing explanation methods may leave users passive or confused rather than empowered.

Complexity of modern AI systems requires rethinking what we explain. The paper poses a fundamental question: "why do we explain AI and what should we explain?" This suggests that as AI models grow larger and more complex, we need to move beyond one-size-fits-all explanation strategies and consider the specific learning needs of different users.

Assessing and evaluating AI explanations needs to shift from technical metrics to learner outcomes. The paper emphasizes adopting a "learner-centered approach to assess, design and evaluate AI explanations," indicating that current evaluation methods may not capture whether explanations actually help humans understand and make better decisions with AI systems.

Bridging XAI and learning science presents both opportunities and challenges. While the paper identifies significant potential in combining these fields, it also acknowledges key challenges remain in implementing this approach, suggesting this is an emerging area requiring further research and development.