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.
January 24, 2025
In this week’s Duke AI Health Friday Roundup: examining AI’s prospects in drug development; mouse study suggests potential for xenon as Alzheimer’s therapeutic; advocating for a “master of digital health” degree; neuromorphic computing as next step in AI evolution; looking at how often the unexpected emerges from grant-funded research; narrowing the calibration gap between what LLMs “know” and what people think they know; much more:
AI, STATISTICS & DATA SCIENCE

- “Compounds found (wholly or partly) by such AI methods are still going to be subject to the same white-knuckle dice-rolling as all the others when they get into human trials, because we have (as yet) no computational tools that really help us predict whether we have picked the right target, the right disease, the right biochemical pathway, or the right compound to affect it without doing anything unexpected along the way. When AI systems start to help with those questions, the revolution may really be at hand. But as it stands, some of the current press releases sound like someone trying to sell a new car model by pointing out that its windows roll up and down much more quickly than the competition.” In an article for Chemistry World, Derek Lowe presents a skeptical perspective on the actual likely benefits of an ongoing “revolution” in AI-powered drug development.
- “Overall, the continuing advancements in AI technologies are substantially improving the efficiency and cost-effectiveness of drug development. However, it is essential to recognize that AI is not omnipotent. The strength of AI technologies lies in analyzing big and complex data and aiding quick decision-making to complement human functions and augment human capabilities, but AI is not designed to entirely replace human ingenuity or authority. The drugs designed and properties predicted by AI still require validation through wet-lab experiments, and human input will still be needed to determine the direction of AI research and use.” A review article published in Nature Medicine by Zhang and colleagues examines the current state of the art in the use of artificial intelligence tools in drug development.
- “Dr. Eric Poon said he used to think he did a good job with transcribing his clinical notes in real time on a computer during the appointment, until he started using the software….Poon estimated nearly all his patients agreed for the software to be used during their appointments with no hesitation. The internal medicine physician was one of the first with Duke Health to test the Abridge software and compare it to other similar technology before access was eventually rolled out to others this month.” A WRAL news story features interviews with Duke Health physicians who are using an “ambient” voice-to-text AI system to capture discussions with patients and abstract them into usable drafts for notes to be incorporated, after review and editing, into the electronic health record.
- “…scientists such as Zou are working on ways to make hallucinations less frequent and less problematic, developing a toolbox of tricks including external fact-checking, internal self-reflection or even, in Zou’s case, conducting “brain scans” of an LLM’s artificial neurons to reveal patterns of deception. Zou and other researchers say these and various emerging techniques should help to create chatbots that bullshit less, or that can, at least, be prodded to disclose when they are not confident in their answers. But some hallucinatory behaviours might get worse before they get better.” An article in Nature by Nicola Jones looks at efforts by AI researchers to mitigate the potential harm done by the propensity of generative AI to “hallucinate” responses to prompts.
- “Inspired by the neuronal systems in the human brain, neuromorphic computing has the potential to address the aforementioned bottleneck that can be found in traditional digital computing. Neuromorphic computers perform computations by mimicking the structure and function of neurons and synapses in the brain. Ultimately, this means that both information processing and memory are collocated and integrated into the artificial neural system, naturally avoiding the energy-costly memory movement step inherent in the von Neumann computing architecture.” An editorial appearing in Nature Computational Science argues for pursuing different directions in AI development, including ones that incorporate “neuromorphic computing” approaches.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

- “Xe[non] treatment induced mouse microglia to adopt an intermediate activation state that we have termed pre–neurodegenerative microglia (pre-MGnD). This microglial phenotypic transition was observed in mouse models of acute neurodegeneration and amyloidosis (APP/PS1 and 5xFAD mice) and tauopathy (P301S mice). This microglial state enhanced amyloid plaque compaction and reduced dystrophic neurites in the APP/PS1 and 5xFAD mouse models. Moreover, Xe inhalation reduced brain atrophy and neuroinflammation and improved nest-building behavior in P301S mice.” In an article published in Science Translational Medicine, Brandao and colleagues present findings from a mouse-model study that found that inhaling the noble gas xenon may have therapeutic potential in Alzheimer disease.
- “Despite the lack of consistency for individual drugs, there are some themes that emerge for drug classes and are consistent with previously published literature and biological plausibility. The association between antibiotics, antivirals, and vaccines and decreased risk of dementia is intriguing….In terms of increased risk, antipsychotic medication appeared strongly….Antidepressants and other drugs targeting the nervous system, and to a lesser extent drugs prescribed to manage blood glucose levels, were among the groups associated with both reduced and increased dementia risk in some of the reviewed studies.” In a review article published in Alzheimer’s & Dementia, Underwood and colleagues examined associations between prescribed drugs and risk of developing dementia, and found that the use of antimicrobial drugs, vaccinations, and anti-inflammatories were associated with reduced risk of dementia (H/T @EricTopol.bsky.social).
- “Project NextGen, a collaboration between the Biomedical and Advanced Research Development Authority and the National Institute of Allergy and Infectious Diseases (NIAID), recently requested project proposals from developers for oral COVID-19 vaccines to generate mucosal immunity at the infection site. Last year, the project awarded funding to 2 companies preparing for phase 2b trials of their intranasal COVID-19 vaccines.” A JAMA medical news article by Rita Rubin explores the reasons that the current generation of COVID vaccines, given as intramuscular injections, don’t generate sterilizing immunity, and examines prospects for next-generation vaccines that take advantage of “mucosal immunity.”
COMMUNICATION, Health Equity & Policy

- “The extent to which biomedical research is targetable has been at the center of longstanding debates in science policy. We aimed to provide new evidence to inform these debates. We used machine learning approaches to develop new measures of how often NIH funding results in publications with categories not in the original grant. Our results indicate that the majority of publications exhibit at least some degree of unexpectedness (70 %, and 58 % with clustered categories) and that, for the average publication, a substantial proportion of all its assigned categories are unexpected (40 %, and 33 % with clustered categories).” A research article published in the journal Research Policy by Aslan and colleagues examines the likelihood of unexpected findings emerging from grant-funded research.
- “As AI, advanced digital technologies, and multimodal data reshape health care, there is an urgent need for professionals who are proficient in digital health fundamentals and possess specialized expertise. The MDH represents a timely and strategic solution to this need, offering standardization, critical competencies, and clarity. By unifying fragmented educational programs, emphasizing the integration of all layers of human data, and aligning with industry demands, diverse contexts, and personal interests, the MDH will position professionals at the forefront of digital health.” In a viewpoint article published in JAMA, Josip Car and Eric Topol argue for the creation of a new professional specialty track – a master of digital health.
- “…we explore the calibration gap, which refers to the difference between human confidence in LLM-generated answers and the models’ actual confidence, and the discrimination gap, which reflects how well humans and models can distinguish between correct and incorrect answers. Our experiments with multiple-choice and short-answer questions reveal that users tend to overestimate the accuracy of LLM responses when provided with default explanations. Moreover, longer explanations increased user confidence, even when the extra length did not improve answer accuracy. By adjusting LLM explanations to better reflect the models’ internal confidence, both the calibration gap and the discrimination gap narrowed, significantly improving user perception of LLM accuracy.” An article published in Nature Machine Intelligence by Steyvers and colleagues examines the gap between the actual and perceived capabilities of large language models.