Black-box models, or algorithms too complex to understand without sophisticated analysis, are becoming more common in the quest for increasingly accurate predictive analytics. While these models have resulted in advances in the field of data science, there are also ethical concerns about their use. See this article in Nature magazine for an overview of the dangers of black-box models and their counterparts, secondary models used to explain them. See the book “Interpretable Machine Learning” by Christoph Molnar for a detailed explanation of model interpretation methods.

Image by author

If these interpretation methods are followed, there is less cause for concern of bias. Even still, there are proponents who believe that all black boxes are bad, and others who believe they can be useful when carefully applied. Still others prefer to avoid the term “black-box” and simply call them more or less interpretable.

#data-science #black-box #algorithms #machine-learning