Testing Explainability Compliance: Specification-Based Testing for AI Transparency

Key Points

• The study aims to assess how well AI systems comply with explainability requirements, ensuring transparency.
• Developed a framework for specify testing of explainability in AI systems.
• Evaluated multiple AI models against the framework to measure compliance.
• Applied quantitative metrics to determine levels of transparency achieved.
• Identified varying levels of compliance among AI models, with some demonstrating clear explainability.
• Quantitative metrics indicated a significant improvement in transparency for models subjected to specification-based testing.
• Found that adherence to specifications led to better user trust and understanding of AI decisions.