Advancing Interactive Explainable AI via Belief Change Theory

Interactive explanations need formal rules: The paper argues that interactive AI explanations should be built on belief change theory—a formal mathematical framework—rather than ad-hoc methods. This provides "warranted behaviour and favouring transparency and accountability," meaning users can understand and verify how the system incorporates their feedback.

User feedback should be systematically integrated: The research proposes a logic-based system for how AI systems should handle new information from users. Rather than arbitrarily accepting or rejecting user corrections, the framework establishes principled rules for when to prioritize user input versus existing model knowledge, addressing real-world scenarios where these priorities conflict.

Current theoretical assumptions may need adjustment: The authors identify that existing belief change postulates (formal rules from philosophy and logic) may not perfectly fit real-world interactive AI systems. This finding suggests that responsible AI development requires adapting theoretical foundations to practical constraints, rather than forcing systems into existing frameworks.

Explainability requires human-machine communication standards: By formalizing how explanations are "shared between humans and machines," the paper emphasizes that responsible AI needs standardized, mutually understandable communication protocols—not just better explanations, but structured dialogue where both parties operate under clear rules.