BoolXLLM: LLM-Assisted Explainability for Boolean Models

LLMs can make formal AI explanations accessible to non-experts: The paper demonstrates that Large Language Models can translate complex Boolean logical rules into natural language explanations that non-technical stakeholders can understand, addressing a key barrier to responsible AI adoption in practice.

Human-centered feature selection improves model trustworthiness: By using LLMs to guide the selection of "semantically meaningful features" rather than relying solely on statistical methods, the framework helps ensure that AI models make decisions based on variables that domain experts and stakeholders recognize as relevant and legitimate.

Hybrid approaches can maintain both accuracy and interpretability: The research shows that integrating LLMs into Boolean rule learning systems achieves "competitive predictive performance" while improving interpretability, suggesting that explainability doesn't necessarily require sacrificing model effectiveness.

Multiple explanation levels serve different stakeholder needs: The framework provides both "global" explanations (how the model works overall) and "local" explanations (why a specific decision was made), recognizing that different users—from regulators to affected individuals—need different types of transparency.

Bridging formal logic and natural language is essential for responsible AI: The paper identifies a critical gap between mathematically rigorous AI explanations and what humans can actually understand, and demonstrates that combining symbolic AI methods with LLMs can help close this gap in real-world applications.