In this episode of the Data Show, I spoke with Sharad Goel, assistant professor at Stanford, and his student Sam Corbett-Davies. They recently wrote a survey paper, “A Critical Review of Fair Machine Learning,” where they carefully examined the standard statistical tools used to check for fairness in machine learning models. It turns out that each of the standard approaches (anti-classification, classification parity, and calibration) has limitations, and their paper is a must-read tour through recent research in designing fair algorithms. We talked about their key findings, and, most importantly, I pressed them to list a few best practices that analysts and industrial data scientists might want to consider.

Why It's Hard to Design Fair Machine Learning Models

O'Reilly Media spreads the knowledge of innovators. At O’Reilly, a big part of our business is paying attention to what’s new and interesting in the world of technology. We have a pretty good record at having anticipated some of the big technology developments in recent history. For instance, we launched the first commercial Web site, GNN, in 1993; we organized the meeting at which the term “open source” was first adopted; we were early investors in Blogger, which helped launch the blogging revolution; and more recently, our Web 2.0 conference launched a world-wide meme. We call this predictive sense the “O’Reilly Radar.” And while we’re certainly not always right, we are, at least, good at making interesting guesses.

Our methodology is simple: we draw from the wisdom of the alpha geeks in our midst, paying attention to what’s interesting to them, amplifying these weak signals, and seeing where they fit into the innovation ecology. Add to that the original research conducted by our Research team, and you start to get a good picture of what the technology world is thinking about.

Why It's Hard to Design Fair Machine Learning Models

O'Reilly Radar