Logic-Based Explainability: Past, Present & Future

Most current explainability methods lack scientific rigor: The paper emphasizes that despite XAI's importance for trustworthy AI, "most work on XAI lacks rigor," and using non-rigorous explanations in high-stakes domains can actually undermine trust rather than build it.

Logic-based approaches offer a more trustworthy alternative: Logic-based explainability has emerged as a rigorous alternative to other XAI methods, providing a more scientifically sound foundation for explaining AI decisions in safety-critical applications.

The "black box" problem threatens human oversight: Advanced AI/ML models operate "often beyond the grasp of human decision makers," meaning decisions affecting people may lack proper understanding and validation—a core concern that XAI aims to address.

Explainability is essential for responsible AI deployment: The paper identifies XAI as "a cornerstone of trustworthy AI," suggesting that without proper explanations, organizations cannot responsibly deploy AI systems in high-risk or safety-critical domains.

Common XAI approaches are built on misconceptions: The paper highlights "many myths that pervade non-rigorous approaches for XAI," suggesting that popular explainability techniques may be based on flawed assumptions rather than solid scientific principles.