The National Institutes of Health is integrating artificial intelligence and machine learning to tackle inequalities across the health care system and advance medical research.
AI’s benefits in driving biomedical and behavioral science research to predict and diagnose disease and personalize treatments are innumerable, noted Laura Biven, data science technical lead at NIH, during a virtual event.
“AI touches all parts of the spectrum. For example, how can we use AI/ML to match proteomic data with how we annotate genomes? Then how do we use those results of basic research and translate them into a clinical health environment? It really spans the entire health spectrum of basic research to translational sciences as well,” Biven said.
Critical to these efforts is creating a “fair and modern data ecosystem” to make data findable, accessible, interoperable and reusable, she added. Many biomedical applications rely on information from a variety of data sources, Biven said. Having better access to that data is critical.
“AI is one of the very cool and efficient tools that data scientists have in their belt, and the immense success that people are having with AI is driving a lot of things that we want to do in data science. It’s driving us to think in a more sophisticated way about data management, how we structure our data, how to make data accessible, and how we link data with computing platforms,” Biven said.
As NIH continues to advance its AI capabilities, it’s increasing its focus on ethical considerations to construct trustworthy algorithms from the get-go. This includes privacy, bias and, of course, access.
Many data sets in general lack data on race, ethnicity and social determinants of health, NIH noted. Plus, the data-generation process currently relies too much on human assumptions, inferences and biases, which contribute to inequality of the applications of AI.
“We really want to make sure that we’re thinking about how we can make sure that these technologies are really benefiting the broadest group of people,” Biven said.
To do so, it’s going to take a well-rounded approach from multiple angles.
“That requires broad engagement at the level of representation within data sets, at the level of determining what kinds of questions people can ask of those data sets, and having a really robust framework for testing those models and making sure that we’re addressing any biases that surface,” Biven said.
NIH’s Bridge2AI program, which aims to “set the stage for” widespread AI adoption, is one initiative that is tackling some of these complex challenges by bringing a variety of experts together to source ethical data sets and to set best practices for machine-learning analysis.