Why AI Needs a Human in the Loop for Lung Cancer Screening Decisions

As health systems explore artificial intelligence (AI) tools to identify patients eligible for lung cancer screening and support shared decision-making, questions remain about whether those tools can serve historically underserved populations without widening existing disparities.

Lisa Carter-Bawa, PhD, MPH, APRN, ANP-C, FAAN, FSBM, founding director of the Cancer Prevention Precision Control Institute at the Hackensack Meridian Center for Discovery & Innovation, spoke with CancerNetwork^® about where she sees the greatest promise, and the greatest risk, in applying AI to lung cancer screening communication.

Transcript:

CancerNetwork: Where do you see the greatest promise, and the greatest risk, in applying AI to lung cancer screening communication and shared decision-making, particularly for populations that have historically been underserved?

Carter-Bawa: I think the greatest risk of applying AI to lung cancer screening initiatives is the fact that AI is not a person, so if you don’t keep a human in the loop, you risk losing some people. I think AI is great for being able to find people who are eligible for lung cancer screening and help get them into the clinic, but the conversation still has to happen with a person. I know there’s literature out there about chatbots being developed, and that chatbots maybe can be empathetic, but I still think you need a human in the loop because at some point, the person has to talk to a person.

One of the things I do worry about with AI is that large language models are trained on large amounts of data, and people who are historically underrepresented in that data are minorities and underserved populations. If you train the AI on data that doesn’t have good representation of those populations, you’re not necessarily going to get the best benefit of it, and it could unintentionally widen disparities, even though you think you’re doing a good thing. That’s the unintended consequence.