Discussion on Explainable AI Submitted to NIST

The Center for Data Innovation has provided comments on the National Institute of Standards and Technology (NIST)'s draft white paper, "Four Principles of Explainable Artificial Intelligence." The filing contains the Center's response to NIST's request for comment, offering insights and recommendations for the document.

The draft white paper seeks to develop principles encompassing the core concepts of explainable AI. NIST's focus is on principles that would help create AI systems that are interpretable and transparent, fostering trust among users.

However, the Center for Data Innovation emphasizes that reliability and explainability are crucial for building user trust in artificial intelligence. The Centre argues that AI systems must be both dependable and interpretable to foster confidence among users.

One of the key recommendations made by the Center is for NIST to reconsider their suggestion for AI systems to assess when they are likely to cause harm. The Center notes that developers do not have the context-specific knowledge to know what will cause harm in a given domain application.

Another recommendation is for NIST to clarify the multiple factors that affect trust, particularly accuracy. The Center explains that the accuracy and reliability of an AI system is likely to be more important to user trust than explainability.

The Center also notes the lack of empirical data quantifying the impact of explainability on user trust and AI adoption. They suggest that NIST should revise their suggestions based on these considerations.

Furthermore, the Center argues that while trust is useful, it is not the only factor influencing AI adoption. Consumers often prioritize price and quality when making purchasing decisions.

In conclusion, the Center for Data Innovation's comments on NIST's draft white paper provide valuable insights into the importance of reliability and accuracy in AI systems for building user trust. The filing encourages NIST to revise their suggestions based on these considerations, aiming to create AI systems that are both trustworthy and effective.