Machine Learning Design Patterns

Chapter 7. Responsible AI

Until this point, we’ve focused on patterns designed to help data and engineering teams prepare, build, train, and scale models for production use. These patterns mainly addressed teams directly involved in the ML model development process. Once a model is in production, its impact extends far beyond the teams who built it. In this chapter, we’ll discuss the other stakeholders of a model, both those within and outside of an organization. Stakeholders could include executives whose business objectives dictate a model’s goals, the end users of a model, auditors, and compliance regulators.

There are several groups of model stakeholders we’ll be referring to in this chapter:

Model builders

Data scientists and ML researchers directly involved in building ML models.

ML engineers

Members of ML Ops teams directly involved in deploying ML models.

Business decision makers

Decide whether or not to incorporate the ML model into their business processes or customer-facing applications and will need to evaluate whether the model is fit for this purpose.

End users of ML systems

Make use of predictions from an ML model. There are many different types of model end users: customers, employees, and hybrids of these. Examples include a customer getting a movie recommendation from a model, an employee on a factory floor using a visual inspection model to determine whether a product is broken, or a medical practitioner using a model to aid in patient diagnosis.

Regulatory ...