Ann Aerts, head of Novartis Foundation, said severe shortages of doctors and nurses in some parts of the world demand innovative approaches to health care. Technology has already transformed the way medicine is practiced in some resource-poor settings. In Ghana, a telehealth pilot has evolved into a national telehealth program that keeps a center staffed 24 hours a day, seven days a week, with doctors, nurses, and health workers. The program isn’t a substitute for skilled medical care, but Aerts said that about 70 percent of all health care problems can be solved over the phone. The program highlights how innovative solutions can be found to the health care shortage that is a fact of life in many parts of the world.

“This calls for innovating the way we deliver health services, for transforming our health systems,” Aerts said. “We cannot go back to the old model. It doesn’t work. … Let’s look at how we can reimagine the way we deliver health services.”

John Halamka, chief information officer at Beth Israel Deaconess Medical Center and the International Healthcare Innovation Professor of Emergency Medicine at Harvard Medical School, said that AI has the potential not only to meet needs where care is scarce, but also to improve care and reduce errors where doctors and nurses are present. He told of a woman who visited a doctor in northern India because of abdominal pain. She was incorrectly diagnosed and paid for repeated testing and physician visits. Ultimately, she was forced her to sell her house and cow, impoverishing her family.

She was finally diagnosed correctly with abdominal tuberculosis, a condition that might rarely be seen in the U.S., but ought to have been considered much sooner in India, which has 1.3 million active cases of tuberculosis, Halamka said.

Halamka consulted on another case, in Bakersfield, California, that shows the problem is not restricted to the developing world. A young woman was arrested for running naked through a Walmart parking lot at 3 a.m. When Halamka called to talk to the emergency room physician, he was told that the young woman’s urine had tested positive for cannabis, so he had discharged her.

“And of course all of us in this room know that cannabis consumption results in young women running naked through parking lots at 3 in the morning — never,” Halamka said. “We flew her to Logan [International Airport], brought her to Beth Israel Deaconess … [and she had] the worst case of meningitis any of us had ever seen. Again thinking back on AI, if we had [it trained by the experiences of] a million young women with altered mental status, how many would include cannabis as the primary cause? Probably, again, none. But in a person who lives with a lot of other 20-year-olds who have lots of infectious diseases, what are the chances of meningitis? Actually, pretty significant.”

Halamka said that artificial intelligence and machine learning has progressed rapidly since the late 1990s, when it took six months of work to get AI to recognize a giraffe as a mammal. Today, he said, designing an AI algorithm itself is relatively easy, but algorithms have to be trained on large amounts of data from past cases, and it can be challenging to ensure that data is available, appropriate, and unbiased.

“Gathering and curating good-quality data, unbiased data, and using it to train these algorithms is still a challenge,” Halamka said. “The quality of our data is poor. We get social security numbers wrong 11 percent of the time. We even get gender wrong 3 percent of the time. Men have babies, women have prostatectomies. We need to work on curated data sets if we’re going to be successful with AI.”

Designers of AI health care applications, Aerts and other speakers said, should also guard against AI becoming a benefit solely for the rich, a factor that increases inequality instead of helping close the gap between haves and have-nots. Ashley Nunes, a fellow at Harvard Law School, said there’s a perception that AI will cut health care costs, making it more affordable for the poor, but there’s no guarantee that will happen.

Nunes pointed to self-driving cars, which are projected to make roads safer. Accident data show the poor suffer the most physical and economic harm from road accidents, so hypothetically they could see the most benefit. But the added cost of the driverless technology means it will likely disproportionately benefit wealthier drivers, highlighting how important it is to ensure that tech solutions in health care benefit everyone.

Jha said he looks forward to a time when AI is just another tool in the health care toolbox.

“The whole idea of digital health reminds me that about 20 years ago we used to talk about e-commerce,” he said. “We don’t talk about e-commerce anymore; we just talk about commerce.”