Explainable AI for Practitioners

Chapter 8. Putting It All Together

In this book, we have given you, the ML practitioner, a framework for how to use Explainable AI (XAI) and where it can be best applied. We also gave you a toolbox of explainable techniques to apply in different scenarios, and guides for crafting responsible and beneficial interactions with explanations. In this chapter, we step back to focus on the bigger picture around Explainable AI. With the tools and capabilities that we covered in this book, how can you approach the entire ML workflow and build with explainability in mind? We also provide a preview of the upcoming AI regulations and standards that will require explainability.

Building with Explainability in Mind

Many times, explainability is approached as an afterthought to model development, an added bonus to developing your most recent top-performing model or a post hoc feature request required by your boss trying to adhere to some new regulatory constraint that’s been imposed on the business. However, explainability and the goal of XAI is much more than that.

Throughout this book, we’ve discussed in detail a number of explainability techniques and seen how they can be applied for tabular, image, and text data. For the most part, we’ve explored these techniques in isolation so you, the reader, can quickly get up to speed on commonly used methods, how they work, their pros, and their cons. Of course, in practice explainability doesn’t occur in isolation. You should consider these techniques ...