Exploring the value of prediction explanations

First off, the concept of explaining a model through its predictions is referred to by many other names, including explainable AI, trustable AI, transparent AI, interpretable machine learning, responsible AI, and ethical AI. Here, we will refer to the paradigm as prediction explanations, which is a clear and short way to refer to it.

Prediction explanations is not a technique that is adopted by most machine learning practitioners. The value of prediction explanations highly depends on the exact use case. Even though it is stated that explanations can increase transparency, accountability, trust, regulatory compliance, and improved model performance, not everybody cares about these points. Instead of understanding the benefits, let’s look at it from a different perspective and explore some of the common factors that drove practitioners to adopt prediction explanations that can be attributed to the following conditions:

The...