In machine learning often a trade-off must be made between accuracy and intelligibility: the most accurate models usually are not very intelligible (e.g., deep nets and random forests), and the most intelligible models usually are less accurate (e.g., linear or logistic regression). This trade-off often limits the accuracy of models that can be safely deployed in mission-critical applications such as healthcare where being able to understand, validate, edit, and ultimately trust a learned model is important. We have been working on a learning method based on generalized additive models (GAMs) that is often as accurate as full complexity models, but even more intelligible than linear models. This not only makes it easy to understand what a model has learned, but also makes it easier to edit the model when it learns something inappropriate because of unexpected landmines in the data. Making it possible for experts to understand a model and interactively repair it is critical because most data has these landmines. In the talk I’ll present two healthcare cases studies where these high-accuracy GAMs discover surprising patterns in the data that would have made deploying a black-box model risky, and also discuss the importance of model intelligibility for dealing with fairness and trust issues such as race and gender bias. See more here.