AI Health

Friday Roundup

The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.

April 10, 2026

In the week’s Duke AI Health Friday Roundup: astronauts’ wearables report health data back to Earth for study; finding ways to help the machine forget; study describes “cognitive surrender” in LLM users; rural-urban gap in resources may spell trouble for clinical AI implementation; moving the needle on nutrition is both hard & complex; what to do when clinicians evade institutional governance on AI use; much more:

AI, STATISTICS & DATA SCIENCE

A robot sculpture, made of tarnished metal parts, posed as if contemplating its own upraised hand. Image credit: Hasem Link/Unsplash
Image credit: Hasem Link/Unsplash
  • “The Hubble team didn’t find foolproof ways to prevent memorization. But they were able to verify some existing theories about why it happens. They confirmed, for example, that data fed to an LLM late in the training process were far more likely to be memorized and regurgitated….This poses a ‘double-edged sword’ according to Jia, as newer model training practices emphasize providing LLMs with richer, more useful text late in the training process in order to ensure they retain the information. Science’s Peter Hall describes attempts to counter the problem of memorization in LLMs, in which the applications can regurgitate (sometimes lengthy) verbatim passages from media used to train them.
  • “The real threat is a slow, comfortable drift toward not understanding what you’re doing. Not a dramatic collapse. Not Skynet. Just a generation of researchers who can produce results but can’t produce understanding. Who know what buttons to press but not why those buttons exist. Who can get a paper through peer review but can’t sit in a room with a colleague and explain, from the ground up, why the third term in their expansion has the sign that it does.” A lengthy, thoughtful blog post by astrophysicist Minas Karamanis lays out some of the subtle perils that may attend the use of large language models.
  • “Age-adjusted mortality rates, chronic disease prevalence, and socioeconomic challenges increase with rurality, while the health care workforce and infrastructure decline. This mismatch creates contexts where AI could be most beneficial. However, AI implementation capacity declines from metropolitan to rural areas across key indicators, including interoperability infrastructure, AI adoption, and large language model readiness. In addition, clinical AI systems trained predominantly on urban populations raise concerns about distribution shift and transportability when applied to rural populations. Without deliberate intervention, AI risks amplifying rather than addressing existing disparities.” A perspective article published in NEJM AI by Hwang and colleagues highlights the paradox that rural and under-resourced health systems that could potentially benefit most from AI tools are least able to access and deploy them.
  • “In the past, people have often used tools from calculators to GPS systems for a kind of task-specific ‘cognitive offloading,’ strategically delegating some jobs to reliable automated algorithms while using their own internal reasoning to oversee and evaluate the results. But the researchers argue that AI systems have given rise to a categorically different form of ‘cognitive surrender’ in which users provide “minimal internal engagement” and accept an AI’s reasoning wholesale without oversight or verification.” Ars Technica’s Kyle Orland reports on recent research that documents the phenomenon of “cognitive surrender” associated with uncritical use of AI technologies.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

Group photo of four astronauts from the NASA Artemis II Moon mission, all wearing mylar-covered paper sunglasses for viewing a solar eclipse. Image courtesy NASA/Johnson Space Center
Image courtesy NASA/Johnson Space Center
  • As NASA’s Artemis II mission makes headlines for its spectacular launch and stunning images of the Earth and Moon, it’s worth noting that it’s also incorporating a (literal and figurative) pilot study of human biology in spaceflight via wearable technologies: “This study will measure astronauts’ well-being, activity, sleep patterns, and interactions during the Artemis campaign that will return astronauts to deep space and the Moon for the first time in over 50 years. These missions will launch in a new space vehicle, the Orion capsule, and journey deep into space, resulting in longer exposures to space hazards including isolation, radiation, and more.”
  • “Contrary to previous studies, we found no significant difference between the effect of SMART compared to N-Back on any of the outcome measures. We discuss potential interpretations of these results (from the perspective of both proponents and critics of relational training claims). We note that there has been a good deal of variability in how SMART has been implemented in the literature, and call for greater specificity in how SMART should be most effectively implemented.” A preprint article by Cummins and colleagues available from PsyArXiv describes the results of a randomized trial that failed to find academic benefit for students who received relational operant conditioning.
  • “Although researchers have long recommended dietary patterns that improve both human and planetary health, large-scale transitions toward sustainable, healthy, and affordable diets remain elusive. Dietary behaviors are difficult to change because they are shaped by a complex web of influences, from direct factors—such as taste, affordability, convenience, and culture—to indirect forces rarely noticed by consumers, including food industry marketing, reformulation, product displays in groceries, and product innovation.” A review article by Yang and colleagues, published in Science, surveys the challenges of attempting to change the trajectories of diet in healthier directions.
  • “When a profession’s core competencies become reproducible, the central question is not whether it will disappear but how its social role will be redefined. Physicians’ aura—once grounded in the irreducible presence of one human before another—will diminish as caring becomes technologically reproducible and warmth can be simulated on demand. The big question for physicians today is how they will reimagine their roles when many of the skills for which physicians were uniquely valued are now reproducible at scale, available to all, and constantly improving.” In an article for JAMA, pediatrician and ethicist John Lantos explores how doctors may need to confront changes to the foundational aspects of their profession as the capabilities of AI systems continue to evolve.

COMMUNICATIONS & Policy

Photograph of dark pavement with the word SLOW painted on it in bright yellow capital letters. Image credit: Anisa Guari/Unsplash
Image credit: Anisa Guari/Unsplash
  • “With many physicians dissatisfied with their health system’s AI adoption speed, it is not surprising that most use occurs outside institutional oversight. AI governance within health systems is generally designed to assess AI tools oriented around specific use cases, which allows systems to quantify the potential harms of such tools so that they can be mitigated. This approach to governance fails for general-purpose generative AI platforms because the range of harms and benefits varies depending on how physicians use them.” A perspective article published in NEJM AI by Ötles and colleagues warns that clinician use of AI outside of narrow operational strictures may be presenting multiple risks currently unaccounted for by existing governance approaches.”
  • The closure comes after Axios Richmond and Poynter alerted CEO Josh Brandau that multiple stories on Nota’s sites included reporting and photographs lifted from local news outlets. The 11 sites — collectively called Nota News — launched in September as an effort to bring “bilingual local reporting and civic tools to underserved communities,” according to a press release. An article by Angela Fu at the Poynter website describes how an AI-powered news service apparently engaged in systematic plagiarism.
  • “…in medicine, the expectation is that we should stay through exhaustion, personal crisis and the slow erosion of self we are trained to ignore….Every part of medicine functions this way….the operating room, care depends on layers of people — nurses, anesthesiologists, surgical technicians, assistants — contributing a different piece of the whole….It is less like a hierarchy than an orchestra, each person entering at the right moment. There is little margin for error. When one note is missed, whether from fatigue, overload or distraction, the entire performance begins to unravel.” In opinion piece for the Chicago Tribune, oncologist Jennifer Obel describes the resonances she feels with television series The Pitt’s depictions of exhaustion and burnout among clinical staff in a busy hospital.
  • “The plan proposes cuts to federal agencies that fund or conduct research on health, space and the environment. Some of the steepest cuts would be made to the National Science Foundation (NSF) and the Environmental Protection Agency (EPA): the budgets of both would fall more than 50% in 2027 compared to their current levels… The budget for the US National Institutes of Health would drop 13%….A budget document says that the proposal would maintain funding for research on quantum information and artificial intelligence ‘to ensure the United States remains on the cutting edge’ in those arenas.” A Nature news article by Max Kozlov, Dan Garisto, and Edward Chen outlines substantial cuts for scientific research and publication in a proposed federal budget.

Sites@Duke Express is powered by WordPress. Read the Sites@Duke Express policies and FAQs, or request help.