Bias in machine learning is a real problem. When models don’t perform as intended, people and process are normally to blame. But it’s possible to employ a “fairness by design” strategy to machine learning, encompassing a few key facets.

To do so, companies can take the following steps: pair data scientists with a social scientist; annotate with caution; couple traditional machine learning metrics with fairness measures; when sampling, balance representativeness with critical mass constraints; and keep de-biasing in mind when building models. Machine learning is increasingly being used to predict individuals’ attitudes, behaviors, and preferences across an array of applications — from personalized marketing to precision medicine.

Unsurprisingly, given the speed of change and ever-increasing complexity, there have been several recent high-profile examples of “machine learning gone wrong.” A chatbot trained using Twitter was shut down after only a single day because of its obscene and inflammatory tweets. Machine learning models used in a popular search engine struggle to differentiate human images from those of gorillas, and show female searchers ads for lower paying jobs relative to male users.

More recently, a study compared the commonly used crime risk analysis tool COMPAS against recidivism predictions from 400 untrained workers recruited via Amazon Mechanical Turk. The results suggest that COMPAS has learned implicit racial biases, causing it to be less accurate than the novice human predictors.

In our federally-funded project (with Rick Netemeyer, David Dobolyi, and Indranil Bardhan), we are developing a patient-centric mobile/IoT platform for those at early risk of cardiovascular disease in the Stroke Belt — a region spanning the southeastern United States, where the incident rates for stroke are 25% to 40% higher than the national average. As part of the project, we built machine learning models based on various types of unstructured inputs including user-generated text and telemetric and sensor-based data. Read more from hbr.org…

thumbnail courtesy of hbr.org