Recommended Methods for Using the 2020 NIST Principles for AI Explainability

In August 2020, the United States National Institute of Standards and Technology (NIST) published (NISTIR 8312) which promulgated for review four principles for Artificial Intelligence (AI) explainability to assist researchers and practitioners in the AI field. These four principles, defined in more detail in the NIST document, are Explanation, Meaningfulness, Explanation Accuracy, and Knowledge Limits. The principles were tied to five types of explainability: User Benefits, Societal Acceptance, Regulatory and Compliance, System Development, and Owner Benefits. The proposed approach is to engage knowledgeable researchers and practitioners to offer ideas congruent with the four principles and five types of explainability. In addition to more detailed frameworks and approaches within the bounds of NISTIR 8312, participants will examine the NIST propositions for completeness and sufficiency.