ETHIXAI
A Guide to Debiasing Your Product
About this Guide
This e-book provides readers with a general guide to bias in artificial intelligence algorithms and what you can do about it at various stages in the development process. You do not need to be a coder or a sociologist to gain value from this and I hope it is valuable to anyone working at a tech company. This guide is divided into three sections:
1. Understanding Bias, which defines the different kinds of bias and how they arise,
2. Accounting for Bias, which provides systematic methods of identifying bias during product development
3. Mitigating Bias, which includes an Algorithmic Audit and gives strategies to actively combat bias in our AI algorithms and products.
We also have some case studies in the appendix and further reading. Use this guide as a kicking off point to debias your product!
Why you should care about this
Protects You Legally
Federal Law prohibits discrimination based on race, sex, national origin, religion, age, or disability are protected classes. If an algorithm discriminates against any of these protected classes, the makers of the algorithm may be held liable.
Protects Reputation
Offering users a product that is based on a biased AI algorithm can result in reputational harm for your organisation. If users believe a product is biased toward them because of a protected characteristic their trust toward the company can decrease.
Builds Robust Products
As AI gets used more and more in services, debiasing these models is essential to provide clients with a robust product. Transparency about the assumptions behind any ML model is a key pillar to build credibility with clients.
