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.
February 27, 2025
In this week’s Duke AI Health Friday Roundup: agentic AI and laboratory automation; revealing the values that underlie AI models; convolutional neural net classifies animal behavior; pilot program for autonomous AI clinical care; role of individual states in supporting research; NBER report finds evidence for AI productivity gains yet to be realized; Elsevier unveils literature-searching AI tool; much more:
AI, STATISTICS & DATA SCIENCE
- “Agentic AI systems have been developed for various applications, including drug discovery, data analysis and biomarker identification; however, several distinct challenges remain for making these systems broadly deployable in biomedical research. Here we discuss three key algorithms and seven foundational building-block characteristics that contribute to the development of agentic AI systems.” A review article published in Nature by Li and colleagues surveys the potential and challenges of utilizing agentic AI in laboratory and bench science.
- “Although model cards and regulatory guidelines describe technical specifications and broad principles, they fail to disclose the clinical values embedded in actual recommendations. Drawing on clinical scenarios where both human experts and large language models demonstrate divergent value-based decisions, we propose a ‘Values In the Model’ (VIM) framework. The VIM is a transparent labeling system that documents how AI systems navigate value-laden clinical trade-offs.” A perspective article by Goldberg and colleagues, published in NEJM AI, introduces a framework for surfacing the tacit values that underlie the functioning of health AI applications.
- “Artificial intelligence (AI) tools for scouring scientific literature have proliferated in recent years, promising researchers a better way to summarize findings and generate hypotheses. Now, Elsevier, the largest publisher of scientific articles, has entered the fray with its own AI tool—and, in a first-of-its-kind pact, has teamed up with four other publishing groups to make the full text of millions of paywalled journal articles available for its algorithm to analyze.” In an article for Science, Jeffrey Brainard reports on a new offering from scientific publishing giant Elsevier that uses an AI to sift paywalled journals, for a fee.
- “Jewett expects that autonomous labs capable of running tens of thousands of experiments in a short period of time will excel in similar efforts, in which many conditions can be tweaked to achieve a clear desired result. His protein synthesis method had around 1,000 reactions….But autonomous labs will face challenges in experiments in which there isn’t a single clear-cut assay to track progress, adds Jewett.” At Nature, Ewan Calloway reports on the debate over whether largely automated (and autonomous) research laboratories are ready for prime time.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “…we introduce BehaveAI, a biologically inspired video analysis framework that integrates static and motion information through a novel color-from-motion encoding strategy. This method translates object movement—direction, speed, and acceleration—into color gradients, enabling both human annotators and pre-trained convolutional neural networks (CNNs) to infer motion patterns while retaining high-resolution spatial detail.” A research article published in PLOS Biology by Troscianko and colleagues introduces an AI system that can rapidly identify and classify animal behaviors based on motion.
- “…we extend the storage ability to borosilicate glass, offering a lower-cost medium and reduced writing and reading complexity. Accelerated ageing tests on written voxels in borosilicate suggest data lifetimes exceeding 10,000 ” A research article published in Nature by a team from Microsoft describes the development and demonstration of an ultra-dense, ultra-long-lived optical data storage medium.
- “Individuals in this study with higher incomes and greater educational attainment were willing to travel longer durations to medical appointments. These differences suggest that individuals with higher socioeconomic status can afford higher transportation costs and have more control over their schedules. Ease of access plays a role as well.” In a research article published in JAMA Open, Burke and colleagues report findings from a survey study that assessed the capacity and willingness of older US adults to travel for access to primary and specialty medical care.
- “The pace of progress is unlike anything medicine has previously experienced. The rapid adoption of large language models illustrates how quickly new technologies can become embedded in daily life. Whether autonomous care becomes part of the solution will depend not only on what technology can do, but on how deliberately and responsibly we choose to engage with it.” In a perspective article for the Journal of the American College of Cardiology, Harlan Krumholz introduces the ADVOCATE program, a government initiative piloting the development of autonomous approaches to clinical care in cardiology.
COMMUNICATIONS & Policy
- “On average, 69 percent of businesses currently use some form of AI, and 75 percent expect to use it over the next three years….The impact of AI on employment and productivity is limited, according to those polled. More than 90 percent of managers say AI had no impact on employment at their organization over the past three years, and 89 percent saw no change in productivity.” The Register’s Dan Robinson reports on a recent working paper published by the National Bureau of Economic Research that found that hoped-for productivity gains from AI have not yet materialized for many businesses.
- “This paper presents a forensic scientometric case study of the Pharmakon Neuroscience Research Network (PNRN), a fabricated research collective that operated primarily between 2019 and 2022 while embedding itself within legitimate scholarly publishing channels….Most PNRN-linked articles remain uncorrected, demonstrating how low-visibility misconduct persists through citation leakage.” A case study by McIntosh and colleagues, available as a preprint from arXiv, details the longer-term ramifications of a bogus research network that successfully insinuated itself into scholarly publications.
- “…state funding for research will never be as broad-based as that of the federal government. States prefer to fund research that generates clear benefits to their residents by spurring economic growth—fostering start-up firms, new employment, and workforce development. That means concentrating on specific technologies and industries.” In an editorial for Science, a pair of authors from RTI International examine whether individual states are capable of offsetting declines in federal support for scientific research.
- “…the AI’s notes might refer to ‘fishfingers or flies or trees’ when in fact a child was talking about their parents fighting. Social work experts said such glitches were particularly worrying as it could cause a risky pattern of behaviour to be missed…Other social workers raised concerns about inaccuracies in transcribed conversations with people with regional accents. One described how their AI-generated transcriptions often included ‘gibberish’.” The Guardian’s Robert Booth reports on a recently published study by the Ada Lovelace Foundation that identified problematic issues arising from the use of AI transcription technologies by UK social workers.
