Sci-fi author Ray Bradbury’s short story "I Sing the Body Electric!” tells the tale of a widower who buys a robot to raise his young children. The electric grandmother becomes a beloved member of the family, meeting all their needs. Machines aren’t replacing grandmothers anytime soon, but researchers are turning to artificial intelligence to help the elderly live with dementia and other conditions of aging.
In principle, large language AI models could be used to integrate and act on the mountains of data collected from cellphones, smartwatches, medical devices, home electronics, and even cars. The need to diagnose and treat cognitive decline is certainly great, as adults over 65 are now America’s fastest-growing demographic, according to the U.S. Census Bureau. In addition, the health and long-term care costs for people with dementia could reach $1 trillion by 2050.
With new technologies on the horizon, Johns Hopkins Carey Business School’s Alonzo and Virginia Decker Professor Phillip Phan and colleagues from the A2 Collective, a federally funded consortium that is exploring AI and aging, convened a two-day symposium at the Johns Hopkins University Bloomberg Center in Washington, D.C., to discuss how AI and other technologies could detect conditions and improve care for aging conditions, particularly cognitive decline.
“We are at the starting line of discovering the power of AI for supporting the health span of older adults and their caregivers,” Phan said. “On this journey, we must always keep the person in view. Human beings are created to be emotionally and socially needy. Machines can never fully replicate those qualities.”
Jason Karlawish, professor of geriatric medicine at the University of Pennsylvania, and I. Glenn Cohen, deputy dean at Harvard Law School, were among the academics and policymakers exploring the technical, legal, and moral challenges of applying AI to elder care.
Despite the promise and clear need for assistive AI agents, the larger question for Karlawish and Cohen was whether AI agents should be entrusted with caregiving. Both agreed that society isn’t ready for systems collecting and acting on a wide range of data, even with the best of intentions. As examples, they cited cars that detect when you're driving too fast, phones that measure if you're moving enough, or even pacemakers that track your location.
“First question is who gets to collect the data on what terms and how is that data collected in a way that either protects you or doesn't protect you? So not only are we talking about collecting data that's relevant for the task that seems like a good task—but there's a strong incentive to go beyond that,” Cohen said. “Beneficial surveillance sounds very good, but maybe this is disfiguring the correct relationship between a child and a parent, even a parent who is in decline. You might say that this is very complicated and fraught territory.”
Karlawish noted that even if AI agents could detect health problems, they still lack the capacity to relate to humans and understand their needs.
“A caregiver is an agentive support for another person. But once machines, generalized pre-trained transformers step in and say, ‘I'm going to help you be agentive,’ it raises the question, what's the pre-training? What language model should be plugged in to help Glenn be Glenn or to help Jason be Jason?” Karlawish said.
Eventually, Karlawish sees potential for a new role of AI and data curator as part of the healthcare team.
“The problem is the sophistication of the theory of other minds and the sophistication of trying to anticipate what it is we want as our agents,” Cohen added. He explained that while machines are good at executing tasks, they only have a rudimentary understanding of how someone would want that task completed.
In summing up, Karlawish shared the story of a colleague who invested in elaborate technology to monitor his mother’s deteriorating condition and allow her to live independently for as long as possible. Ultimately, Karlawish says his friend regretted delaying his mother’s move to a nursing home because she eventually benefited from being with other humans. "’I should have placed her much earlier into that setting with these other humans, interacting with her. I was just too zealous with all my technology,’” Karlawish said.
