Designing Machine Learning Systems

Chapter 11. The Human Side of Machine Learning

Throughout this book, we’ve covered many technical aspects of designing an ML system. However, ML systems aren’t just technical. They involve business decision makers, users, and, of course, developers of the systems. We’ve discussed stakeholders and their objectives in Chapters 1 and 2. In this chapter, we’ll discuss how users and developers of ML systems might interact with these systems.

We’ll first consider how user experience might be altered and affected due to the probabilistic nature of ML models. We’ll continue to discuss organizational structure to allow different developers of the same ML system to work together effectively. We’ll end the chapter with how ML systems can affect the society as a whole in the section “Responsible AI”.

User Experience

We’ve discussed at length how ML systems behave differently from traditional software systems. First, ML systems are probabilistic instead of deterministic. Usually, if you run the same software on the same input twice at different times, you can expect the same result. However, if you run the same ML system twice at different times on the exact same input, you might get different results.¹ Second, due to this probabilistic nature, ML systems’ predictions are mostly correct, and the hard part is we usually don’t know for what inputs the system will be correct! Third, ML systems can also be large and might take an unexpectedly long time to produce a prediction.

These differences ...