Explanatory quote from the linked paper (and thanks for the link, qPM9l3XJrF!): "Rather than trying to create models that are inherently interpretable, there h

Explanatory quote from the linked paper (and thanks for the link, qPM9l3XJrF!): "Rather than trying to create models that are inherently interpretable, there has been a recent explosion of work on “Explainable ML,” where a second (posthoc) model is created to explain the first black box model. This is problematic. Explanations are often not reliable, and can be misleading, as we discuss below. If we instead use models that are inherently interpretable, they provide their own explanations, which are faithful to what the model actually computes"