“Our Machines Now Have Knowledge We’ll Never Understand” by David Weinberger, Wired (2017).

The format in which machine learning algorithms represent the “knowledge” that drives their predictions is often not human-readable. What are some of the dangers of this lack of interpretability? Are there deployment situations in which we might be okay with uninterpretable algorithms which provide good results?

This book, and all H2O books, are Creative Commons licensed for sharing and re-use with the exception of certain excerpts. Any excerpts from the Restatements of the Law, Principles of the Law, and the Model Penal Code are copyright by The American Law Institute. Excerpts are reproduced with permission, not as part of a Creative Commons license.