In 2016, a Google team announced it had used artificial intelligence to diagnose diabetic retinopathy — one of the fastest-growing causes of blindness — as well as trained eye doctors could.
In December 2018, Microsoft and the pharmaceutical giant Novartis announced a partnership to develop an AI-powered digital health tool to be deployed against one of humanity’s oldest scourges — leprosy, which still afflicts 200,000 new patients annually.
Around the world, artificial intelligence is being touted as the next big thing in health care, and a potential game-changer for billions living in regions where medical infrastructure is inadequate and doctors and nurses scarce.
Despite that potential, some fear that too-rosy views of AI’s promise will lead to disappointment and, worse, rob scarce health care dollars from desperately needed investments in existing medical infrastructure. Experts in AI and global health gathered at Harvard this week to survey the complex artificial intelligence-global health landscape, examine what’s being done today and what’s promised for tomorrow, and try to separate reality from fantasy when it comes to AI’s potential impact on global health.
“The promise of AI is enormous, but the challenges are often glossed over,” said Ashish Jha, faculty director of Harvard Global Health Institute (HGHI). “We have this sense that they’ll somehow take care of themselves. And we know they will not.”
About 50 experts in the use of artificial intelligence and other digital health care technologies gathered at Loeb House for an all-day symposium on “Hype vs. Reality: The Role of AI in Global Health,” sponsored by HGHI. It featured talks by representatives from academia, health care, and industry, including Google AI and the pharmaceutical giant Novartis. Principals also spoke for startups Aindra Systems, which uses AI to rapidly analyze pap smears for evidence of cervical cancer, and Ubenwa, which has developed an AI-based system to analyze infant cries to detect birth asphyxia, a leading killer of children under 5 that’s traditionally diagnosed through time-consuming blood tests and lab analyses.
Despite the symposium’s cautionary tone, several speakers described AI’s potentially far-reaching effects on global health. Image analysis, a key aspect of disease screening and diagnosis, is an area ripe for rapid change. Lily Peng, product manager for Google AI, described Google’s work on diabetic retinopathy. The company’s researchers used a “deep learning” algorithm that reviewed thousands of images of both normal and damaged eyes and developed a screening tool for the condition — a side effect of poorly controlled diabetes — that performed as well as trained ophthalmologists.
That tool, and similar ones for other conditions, can provide valuable screening and referral services in rural areas that have little medical infrastructure. Their promise is of a future where initial screening can be guided by community health workers — far more numerous than trained physicians — who can then refer those who test positive to clinics and hospitals staffed with better-trained medical personnel.
But the question remained: How effectively would such interventions perform in the field? During a Q&A session, one participant described a device developed by an engineer in Cameroon to diagnose cardiac problems in remote areas. The device was designed to be used by rural nurses who could send results to city-based cardiologists for review. Problems occurred, however, because the nurses entered data in the wrong fields, spotty internet connections made it hard to transmit the data, and finally, when the information made it to the city, the cardiologists were often too busy to examine it.
Adam Landman, chief information officer at Brigham and Women’s Hospital and an associate professor of emergency medicine at Harvard Medical School, said it’s important that AI not be viewed as a solution in search of a problem, but that the health care outcomes be considered first and AI considered as one tool among others to address it.
“The key with any technology — and AI is just one of them — is it needs to actually solve a health care [problem],” Landman said. “The need [should] drive the use of it and not the technology drive the use. They may not need AI to start. In fact, they may not need any technology to start. They may need more people, more health care workers. There may be other things to invest in before technology, or technology may be part of that solution.”