Opening the Software Engineering Toolbox for the Assessment of Trustworthy AI

Software engineering testing practices can be adapted to assess AI trustworthiness: The paper argues that established software testing methodologies from the software engineering community should be applied to evaluate whether AI systems meet trustworthiness standards, rather than treating AI assessment as entirely separate from traditional software quality assurance.

The European Commission identifies seven key trustworthiness requirements for AI: The paper connects these requirements (as defined by the AI high-level expert group) to specific software engineering practices, suggesting that trustworthiness assessment should be systematic and measurable rather than ad-hoc.

AI systems remain fundamentally software-based despite their complexity: While AI systems are highly complex, they are still implemented as software, meaning existing software engineering tools and assessment frameworks can be meaningfully applied to evaluate their trustworthiness characteristics.

There is a gap between AI complexity and current assessment approaches: The paper identifies that trustworthiness assessment currently relies on guidelines, requirements, or informal expectations rather than rigorous testing procedures, suggesting a need to formalize how we evaluate whether AI systems meet trustworthiness standards.

Systematic assessment frameworks are needed for AI acceptance: The paper emphasizes that trustworthiness is "central" to public acceptance and successful deployment of AI systems, implying that rigorous, standardized assessment methods are essential for responsible AI development and deployment.