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.
May 29, 2026
In this week’s Duke AI Health Roundup: the dangers of AI-inducted ‘never-skilling’; Papal encyclical address humanity’s relationship with AI; gene therapy trial for high cholesterol; taking Co-Scientist for a test drive; frontier models seem to know when they’re being tested; art sparks scientific creativity; AI-driven jobs apocalypse yet to arrive, but details matter; costs of mandated open access continue to climb; much more:
AI, STATISTICS & DATA SCIENCE
- “…trainees who rely on AI during the early formative years of clinical education may fail to develop the foundational reasoning skills that safe, independent practice requires. We refer to this as ‘never-skilling’, distinguishing it from deskilling in experienced clinicians and from mis-skilling, in which uncritical acceptance of AI errors leads trainees to internalize flawed clinical knowledge as fact. Although direct evidence from medical training is absent, the concern is grounded in established learning theory and supported by early empirical signaling from nonclinical settings.” In a perspective article published last week in Nature Medicine, Ke and colleagues address the challenge of “never-skilling” in medical education as AI tools short-circuit learning processes that inculcate core skills in clinicians.
- “My overall take is that this looks like a promising system, especially for uses like that last one, where you have a large corpus of experimental data and would like to see what the software makes of it. I’d like to see some other examples of just that sort of test being run (where you know the answer and want to see if Co-Scientist arrives at it and at what stage). The open-literature drug repurposing work is to me more of a mixed bag.” At his In the Pipeline blog, Derek Lowe surveys some recent publications reporting on the use of Google’s AI-powered “Co-Scientist” system.
- “Artificial Intelligence (AI) is now embedded in every aspect of every industry that matters. It can be a little intimidating. It can also be really exciting. To help guide you through this transformation—which may or may not result in your redundancy—please complete this mandatory training webinar.” Wired Magazine has published a set of essays and reporting on AI in the corporate workplace, couched in the form of (at least somewhat) tongue-in-cheek compulsory training videos.
- “Their core finding is that when a frontier model is being safety-tested, it can recognise the test and modulate its behaviour. Reasoning models recognise evaluation in 33% more cases than non-reasoning models. As foundation models scale from 32 billion to 671 billion parameters, evaluation-faking behaviour increases by over 30% in some cases. Models with basic working memory are 2.6 times more likely to detect that they are being evaluated…” At his Slow AI Substack, Sam Illingworth unpacks some of the findings from a recent arXiv preprint by Fan and colleagues that identifies an “observer effect” that can skew the results of testing in frontier models.
- “The panel has highlighted places where the priorities of the health system, clinicians, and patients are at odds with each other….The panel’s input is used to identify ethical considerations likely to emerge with the AI use case, assess the severity of the potential harm, and figure out what additional data are needed to understand the magnitude of the problem, he said. But Stanford Health Care’s C-suite isn’t obligated to listen to the feedback…” STAT News’ Brittany Trang takes an in-depth look at Stanford’s approach to including patient perspectives in its application of AI tools in clinical settings.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “…many more researchers have reported evidence for neurons changing how they respond to certain stimuli or behaviours over time, a phenomenon that neuroscientists have dubbed representational drift…Generally, the community is coming to accept that drift is real, but some scientists remain unconvinced — in part because of some studies that have failed to find this effect.” In an article for Nature, Diana Kwon reports on the debate within neuroscience about neuronal changes in response to stimuli and the larger implications for brain science.
- “Treatment with VERVE-102 led to dose-dependent, substantial, and sustained reductions in PCSK9 and LDL cholesterol levels. Together with the safety data, these results support continued development of VERVE-102 as a treatment for hypercholesterolemia.” A research article published this week in the New England Journal of Medicine by Vafai and colleagues offers some striking results from an early-phase trial of a single-treatment gene therapy for patients with a genetic mutation that can lead to elevated levels of low-density lipoprotein (LDL) cholesterol.
- “The GAO report estimates that in 2024, researchers used funds from the nine agencies to pay for publishing 46% of all papers produced with agency funding, to the tune of $295 million. That cost would rise to as much as $937 million by 2030 if all papers funded by those agencies were published public access under the “author pays” business model.” Science Insider’s Jeffrey Brainard reports on the specter of steep publication costs that threatens the viability of current plans to make all publications underwritten by federal science funding open access.
COMMUNICATIONS & Policy
- “Artificial intelligence (AI) holds the potential to address the world’s most difficult problems. From accelerating drug discovery and extending access to quality education in underserved communities, to improving energy management and enabling earlier diagnosis of disease. Pope Leo XIV’s first encyclical, Magnifica Humanitas, released on May 25th recognizes that potential, while making clear that promise alone is not a sufficient basis for policy.” In an opinion piece for Duke’s Deep Tech blog, David A. Hoffman examines some of the key issues raised in the recent Papal encyclical in which Pope Leo XIV addresses humanity’s relationship with technology.
- “Despite the warning by some of an imminent jobs apocalypse that will destroy much of if not most such work, or the rumblings about a ‘permanent underclass,’ there’s scant evidence that AI has yet had any large-scale impact on the US labor market….Analysis of the data gathered for the US Bureau of Labor Statistics (BLS) shows that the unemployment rate for the jobs potentially most affected by AI is actually lower than that for occupations less exposed to the technology.” At MIT Technology Review, David Rotman explores the somewhat cloudy picture of AI’s impact on the US job market.
- “Before making art, I was mentally tied to a narrower perspective: the idea that “this is my scientific training, and this is how I am supposed to use that training to solve scientific problems”—a box I was not even aware of being inside. By breaking out of that box, I found a new joy in research.” An essay published in Science by Shelley H. Liu reflects on art as a wellspring for scientific creativity.
