Data Science, Machine Learning and the Looming Shakeup

SMITH BRAIN TRUST – Data science and machine learning are revolutionizing organizations, businesses, even private lives – and not all of the changes are good ones. “The voices that caution about the potential pitfalls of machine learning, including bias and unequal share in the benefits, are growing louder, says Maryland Smith’s Margrét Bjarnadóttir. “And this is especially important in healthcare.”

For years, experts have been warning that machine learning has the potential to perpetuate social inequity. Even big tech companies, which are familiar with the challenge of maintaining fairness in machine learning, have struggled to develop unbiased models, says Bjarnadóttir, associate professor of management science and statistics at the University of Maryland's Robert H. Smith School of Business.

Amazon, for example, abandoned a machine learning recruitment tool because it ranked male job candidates higher than females. Google parent Alphabet drew controversy when its Photo app mislabeled darker-skinned people in images.

The healthcare system, when machine learning is applied, is vulnerable. And the potential for physical harm is serious, Bjarnadóttir adds. In 2021, she predicts “a revolution in how we approach building, evaluating and deploying machine learning models in the field of healthcare.”

With machine learning fueling many of the decisions in healthcare operations and increasingly diagnosis and treatment decisions, Bjarnadóttir says she expects the whole field of machine learning in healthcare to shift toward transparency, emphasizing equity.

“We are seeing academic papers fast changing their approach, and best practices will quickly follow,” she says. “What was acceptable just a few years ago now no longer gets published, and hopefully is not put into practice.”

In 2021, researchers, practitioners and software providers will start to directly challenge the status quo, she says. And that may result in industry players “demanding diverse representation in the data that fuels the modeling or transparency in modeling choices and model performance, with the aim of maximizing the benefit that machine learning and data science can bring to the field of healthcare.”