UH Law Center Spring 2019 External Speaker Series: Medical AI faces challenge of contextual bias, says Michigan Law School scholar

Nicholson Price of the University of Michigan Law School addresses University of Houston Law Center students and faculty during a presentation in the Hendricks Heritage Room.

March 15, 2019 - University of Michigan Law School assistant professor Nicholson Price said that artificial intelligence used in the medical field should come from a diverse set of data last week in the Hendricks Heritage Room at the University of Houston Law Center.

While medical AI is meant to democratize medical expertise and push the frontiers of medical knowledge, problems arise when algorithms are trained from a limited set of data, which will not translate properly from a high-resource context to a low-resource context.

"The structures that we currently have in place - legal, technical and otherwise - incentivize a process of development that's going to lead to contextual bias," Price said. "This is what happens when we take AI from one context, in particular academic medical centers, and try to apply those to other contexts, like a community health center or solo practitioner.

"Even when you have really good and non-problematic data, the process of taking those data and translating them over to different environments has the potential to create its own set of biases. That's something we need to be aware of when we're thinking about how to develop it and how to take it from one environment and put it in another."

Price noted that AI used in the wrong context decreases the quality or appropriateness of a recommendation made by the technology. He said a useful move would be to make data sets more representative and to develop algorithms that are developed using data from many contexts.

From a legal standpoint, Price said medical AI should be thoroughly tested and should be able to prove that it is effective in a variety of medical environments.

"Imagine an FDA approval process that said, if you're going to get an algorithm approved as a medical device, which many clinical algorithms will need to be, then part of that approval process requires that you demonstrate applicability across many different contexts," Price said.

The Spring 2019 Distinguished Speaker Series will resume on March 18 at noon every Monday in the Hendricks Heritage Room until April 8. The next speaker is Michael Livermore of the University of Virginia School of Law.

Back to the News Homepage