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.
July 24, 2026
In this week’s Duke AI Health Friday Roundup: Anthropic offers grants for rare disease research; contagious cancers in catfish; making “unlearning” a priority for clinical AI; grappling with noise in the ER; the dynamics of conspiracy-theory “rabbit holes”; assessing the effects of LLM tools on scientific workflows; concerns about image manipulation by lab supplier; how Google’s pivot to AI has cut traffic to websites; much more:
AI, STATISTICS & DATA SCIENCE
- “…we argue that high-risk clinical AI should be designed so that targeted remediation of learned data influence is feasible where legal, ethical, community-governance, or clinical-safety grounds require removal or correction. Our claim is forward-looking: systems built without any capability for targeted remediation are likely to become difficult to reconcile with emerging expectations for data rights, lifecycle governance, auditability, and clinical safety.” A perspective article published in NPJ Digital Medicine by Porter and colleagues makes a case for “unlearning” as a critical issue for the governance of clinical AI systems.
- “…representativeness must be defined with respect to variables relevant to the target of inference, rather than superficial demographic diversity. Here we present the Total Survey Error (TSE) framework as a methodological tool that systematically addresses the multifaceted sources of error—particularly those related to representation—that emerge throughout the research cycle.” A review article published in Compass Social and Personality Psychology by Alsalti and colleagues presents a framework for addressing sources of error and bias in representative sampling in survey studies.
- “Accepted applicants will receive up to $50,000 in Claude credits over six months, with the goal of building a community of researchers looking into how AI can reshape our understanding of rare disease. This program has two tracks: one for scientists doing basic research, and another for early-stage biotechs working on speeding up clinical development for rare diseases.” Anthropic announces a call for applications for grants in “Claude credits” for research projects investigating rare diseases.
- “In fields where LLMs are useful primarily as tools for discovering promising projects, researchers will become more selective about what they publish; where they facilitate the process of publishing existing data, researchers will become less selective. By allowing scientists to work more quickly, LLMs raise the opportunity cost of researcher time, creating incentives to refine papers less thoroughly before moving on.” A preprint article by Deude and colleagues, available at arXiv, examines the “unintended consequences” of incorporating LLM technologies into scientific workflows.
- “Our experiments show that, with carefully chosen defaults, mlr3mbo matches the performance of established state-of-the-art BO frameworks on numeric and mixed-hierarchical YAHPO Gym benchmarks while offering good runtime behavior. Our results underscore the importance of strong acquisition function optimizers with sufficiently large budgets and a surrogate tailored to the search space’s structure.” A paper published in Cell Patterns by Becker and colleagues debuts mlr3mbo, a set of R applications for performing Bayesian optimization.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “So-called melanomas are exceedingly rare in catfish, but more than a third of the bullheads carried them. Now, researchers think they know why: A contagious form of cancer is spreading through the lake, they reported on Wednesday. The outbreak started with a single catfish, the scientists believe. It developed a tumor, which shed cancer cells into the water. Those cells invaded other catfish, in effect beginning a plague.” New York Times science writer Carl Zimmer reports on the surprising emergence of what appears to be a contagious cancer among a population of catfish on the Vermont/Canada border.
- “These findings suggest that factors associated with affluence may contribute to increased susceptibility to pediatric cancers. Given that Ontario’s universal health care system and near-complete ascertainment of cases likely attenuated case ascertainment biases related to diagnosis, the observed patterns underscore the need to consider socioeconomic factors when examining trends in childhood cancer epidemiology.” A research article published in JAMA Network Open by Chiu and colleagues finds that children from affluent neighborhoods in Ontario had a higher incidence of cancer relative to other, less affluent neighborhoods.
- “We treat noise as an irritation, but it also affects patient and staff safety….This problem may not be bigger than the crisis of too few beds, too little space, and too many patients, but it’s one more way in which a crowded department becomes riskier. Unnecessary noise should be designed out of hospital environments, not left for staff and patients to endure.” In an opinion article published in BMJ, Iain Beardsell draws attention to the din that often accompanies emergency care environments – as well as what can be done to abate a nuisance that poses serious problems for staff and patients.
- “In font sizes that range from prominent to squint-worthy, the cans’ labels note that they contain THC, the psychoactive compound found in cannabis. But most of the 10 stores visited on a recent scouting trip through South Brooklyn, some within a couple blocks of schools, displayed no signs indicating that these were intoxicating beverages, nor did they ask for identification when the drinks were brought to the register.” STAT News’ Sarah Todd reports on the potentially problematic marketing and sale of drinks containing THC, which are often sold side-by-side with innocuous soft drinks.
COMMUNICATIONS & Policy
- “For publishers, businesses, banks and others that relied on Google to funnel its billions of users to their websites, the impact has been unmistakable as the company has increasingly incorporated A.I. into search. Google’s users are no longer leaving Google after a search and are just reading its A.I.-generated answers, they said, which means fewer people are coming to their websites and search traffic has dropped.” The New York Times’ Kate Conger reports on Google’s recent change in posture that imposes a layer of AI abstraction between users and the actual web pages Google’s AI systems are extracting summaries from – and incidentally depriving of traffic.
- “… adopting a conspiracy theory elevates the risk of sharing subsequent conspiracy theories…this elevation grows and persists longer with the number of conspiracy theories shared, and…transitions between theories concentrate among semantically proximate narratives, revealing semantic interactions that mediate social contagions. We also document what we term the settler effect: a user’s entry into a new semantic region is slower, but once entry occurs, subsequent within-region adoption accelerates.” A research paper by Tschofenig and colleagues, available as a preprint from arXiv, addresses the dynamics of how and why people succumb to conspiracy theories.
- “While some images may have been altered to remove marks that have little bearing on performance, Richardson says for others ‘it’s hard to imagine the blot was not manipulated to hide that the antibody doesn’t work or that it binds to an off-target protein’. He says until the original blots are provided, it can’t be known what exactly the manipulations cover up. This uncertainty could hold back research projects, while others may now be left concerned that concluded work is now undermined.” Chemistry World’s Frankie MacPherson reports on the groundswell of concern that bioscience lab giant Thermo Fisher has engaged in widespread manipulation of images of Western blots.
