The Balancing Act of Policies in Developing Machine Learning Explanations

Due to the nature of opaque machine learning (ML) models, software engineers and data scientists struggle to understand how ML models make decisions [1]. Explainability research aims to provide transparency for these models [2] through two types of explanations. Global explanations describe how a model works generally and provide insight into its accuracy, biases, and fairness. Local explanations describe individual predictions made by the model in specific use cases.