Uncovering Bias in Machine Learning: A Guide to Im plementing Interpretable Models

With machine learning systems becoming more ubiquitous in automated decision making, it is crucial that we make these systems sensitive to the type of bias that results in discrimination, especially discrimination on illegal grounds. Machine learning is already being used to make or assist decisions in the following domains of Recruiting (Screening job applicants), Banking (Credit ratings/Loan approvals), Judiciary (Recidivism risk assessments), Welfare (Welfare Benefit Eligibility), Journalism (News Recommender Systems) etc. Given the scale and impact of these industries, it is crucial that we take measures to prevent unfair discrimination in them via legal as well as technical means.

This book will give data scientists and Machine learning engineers insight on how building machine learning models and algorithms can negatively impact users. The book will also provide tools and code examples to help document, identify, and mitigate different types of machine bias. The audience are Data Scientists, Machine Learning Engineers, and Researchers who implement and productionalize machine learning models. This book has been needed for decades because it not only helps the reader understand how human bias slips into models but gives them code and techniques to analyze the models they’ve already built. This book will also give engineers the tools to push back on demands from management that result in harmful models.

While this book will focus on machine learning that is used to predict data about users that can be impactful on their lives. Thousands of consumer products use machine learning and these algorithms can cause major damage if influenced by biased data. Google has already classified black people as “gorillas” in Google Photos. Some facial recognition doesn’t even pick up darker toned skin. In terms of trends, ML and AI are by far the hottest fields in computing. The problem with this high-paying, high-growth area is that few practitioners are actually skilled in reducing and mitigating harm caused to users. This book will allow Data Scientists, Machine Learning Engineers, Software Developers, and Researchers alike to apply these explainability steps to their system.