A lightbox showing a mammogram xray
Advances in neural networks and other techniques promise to transform health care while raising profound questions about our bodies and society.
AI software can identify early signs of breast cancer long before the disease can be diagnosed by conventional means.
When MIT professor Regina Barzilay received her breast cancer diagnosis, she turned it into a science project. Learning that the disease could have been detected earlier if doctors had recognized the signs on previous mammograms, Barzilay, an expert in artificial intelligence, used a collection of 90,000 breast x-rays to create software for predicting a patient’s cancer risk.
Barzilay calculates the software could have flagged her own cancer two years before it was diagnosed by conventional means. “The AI was able to detect smaller details than the human eye could pick up,” she says.
Barzilay’s work is featured in the ninth installment of the Sleepwalkers podcast, which tours the effects of artificial intelligence revolution. The episode explores how AI is transforming health care—and perhaps also what it means to be mortal.
Self-driving-car projects might be more glamorous, but health care has also benefited from a rush to apply recent developments in AI. In one case, researchers at Stanford demonstrated software that could diagnose skin cancer at an accuracy rivaling dermatologists.
“People die of cancer a lot,” says Sebastian Thrun, who worked on the skin cancer project and is now CEO of flying car company Kitty Hawk. (Thrun also started Google’s self-driving car program.) “I believe many of those deaths are actually preventable using artificial Intelligence.”
Oncologist and writer Siddhartha Mukherjee cheers the potential of medical AI. But he also warns that we should be wary of some of the less desirable aspects of the technology. One is increased collection of data on patients and their conditions. All that information could be tempting to health insurers or government agencies, which might not always have a patient’s interest as their first priority,
AI systems that can predict a person’s health far into the future—even before birth—raise even trickier questions. Researchers have shown that algorithms can process genomic data to predict individual characteristics, such as a person’s height or risk of certain diseases. Indicators of intelligence or other traits could be next, tempting parents, armed with techniques like IVF and gene editing, to shape—and not just learn—their offspring’s destiny.
Mukherjee says such powerful knowledge will require physicians to think carefully about their role in patients’ lives and in society. “In medicine, because we’re intervening on bodies, we’re intervening on culture,” he says.