A PRACTICAL GUIDE TO IDENTIFYING AND REDUCING BIAS AND UNFAIRNESS IN DATA SCIENCE

Rapid advancements in data science are causing increasing alarm around the world as governments, companies, other organizations, and individuals put new technologies to uses that were unimaginable just a decade ago. Medicine, finance, criminal justice, law enforcement, communication, marketing and other functions are all being transformed by the implementation of techniques and methods made possible by progressively more obscure manipulations of larger and larger data sets. Almost every day, new stories of AI gone awry appear. What can be done to avoid these issues?

Responsible Data Science is an insightful and practical exploration of the ethical issues that arise when the newest AI technologies are applied to the largest and most sensitive data sets on the planet. The book walks you through how to implement and audit cutting-edge AI models in ways that minimize the risks of unanticipated harms. It combines detailed technical analysis with perceptive social observations to offer data scientists a real-world perspective on their field.

The inability to explain how an artificial intelligence model uses inputs can jeopardize the willingness of regulators to even consider whether these technologies comply with existing and future regulatory and legal requirements. In this book you'll learn how to improve the interpretability of AI models, and audit them to reduce bias and unfairness, thereby inspiring greater confidence in the minds of customers, employees, regulators, legislators and other stakeholders.

Perfect for data science practitioners, statisticians, software engineers, and technically aware managers and solutions architects, Responsible Data Science will also earn a place in the libraries of regulators, lawyers, and policy makers whose decisions will determine how and when data solutions are implemented.

This groundbreaking book also covers:

• The various types of ethical challenges confronting modern day data scientists
• How the adoption of "black box" models can aggravate issues of model transparency, bias, and fairness
• How moral concepts like fairness translate (or fail to translate) into a modeling context
• How model-agnostic methods can be used to make models more interpretable, identify issues of bias, and mitigate the bias discovered