Robust Explainability in AI Models Special White Paper Written by Ian Hardy2 Recent research points to weaknesses in some AI explanations due to oversensitivity. We run an experiment using our explainability approaches to demonstrate that with proper methods, AI explainability can be robust and reliable. This is especially important when it comes to machine learning applications within highly regulated industries, such as credit underwriting. The following white paper provides a refresher on explainability, a discussion of selection and use of references, and an analysis of an experiment run to demonstrate robustness in explainability, including distributional referencing, results, and conclusion. Ours aren’t. Some AI explanations can be unreliable3 If we are going to reap the

Join for free to read