In today’s Roundup: global review of bias in clinical AI studies; reconsidering hypertension in pregnancy; how humans build and share algorithms; science journals’ responsibilities to mend old harms; European regulators clear AI x-ray reader for use; starlings and Shakespeare; brain imaging reference spans entire lifespan; much more:
- “…he and Ms. Fugate started digging through archives and databases for any link between the Bard-lover and the bird. According to their findings, which were published in the journal Environmental Humanities in November, Schieffelin did release 40 pairs of European starlings into New York City twice in the springs of 1890 and 1891. But Ms. Fugate and Dr. Miller failed to find evidence that Schieffelin was the Shakespeare superfan he has been made out to be.” This New York Times story by Jason Bittel, nominally about starlings, has just about everything: Shakespeare, historical sleuthing, genetic sleuthing, invasive species, and oft-repeated apocryphal stories.
AI, STATISTICS & DATA SCIENCE
- “U.S. and Chinese datasets and authors were disproportionately overrepresented in clinical AI, and almost all of the top 10 databases and author nationalities were from high income countries (HICs). AI techniques were most commonly employed for image-rich specialties, and authors were predominantly male, with non-clinical backgrounds.” At PLOS Digital Health, Celi and colleagues present results from a global review of potential sources of bias in the clinical AI literature.
- “Remarkably, PaLM can even generate explicit explanations for scenarios that require a complex combination of multi-step logical inference, world knowledge, and deep language understanding. For example, it can provide high quality explanations for novel jokes not found on the web.” Google’s AI Blog reports on what it describes as recent performance improvements in its “Pathways Language Model” (PALM) AI on multiple tasks including translation, reasoning, and more.
- “An artificial intelligence tool that reads chest X-rays without oversight from a radiologist got regulatory clearance in the European Union last week — a first for a fully autonomous medical imaging AI, the company, called Oxipit, said in a statement. It’s a big milestone for AI and likely to be contentious…” The Verge’s Nicole Wetsman reports that ChestLink, an AI tool designed to read radiological images without human input, has been approved for use by European Union regulators.
- “The pilot will initially include Nature research journals and Academic Journals portfolios across the fields of neuroscience, ecology and evolution, chemistry, energy, cancer and transplantation. Working closely with researchers throughout, the pilot will explore and test out more integrated ways for data sharing.” A blog post at Digital Science by David Ellis reports on a pilot collaboration between publisher Springer Nature and FigShare to provide an easier, more seamless path to data-sharing as part of the process of scientific publication.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “In pregnant women with mild chronic hypertension, a strategy of targeting a blood pressure of less than 140/90 mm Hg was associated with better pregnancy outcomes than a strategy of reserving treatment only for severe hypertension, with no increase in the risk of small-for-gestational-age birth weight.” Results from a randomized trial published in the New England Journal of Medicine by Tita and colleagues suggests that previous approaches to managing hypertension during pregnancy may need to be revised.
- “Over the past few decades, neuroimaging has become a ubiquitous tool in basic research and clinical studies of the human brain. However, no reference standards currently exist to quantify individual differences in neuroimaging metrics over time, in contrast to growth charts for anthropometric traits such as height and weight…Here we assemble an interactive open resource to benchmark brain morphology derived from any current or future sample of MRI data…” A study published in Nature by Bethlehem and colleagues describes an ambitious project that scoured multiple smaller studies to create a comprehensive brain imaging reference library for human brain growth, development, and decline over the course of a lifetime.
- “Many human abilities rely on cognitive algorithms discovered by previous generations. Cultural accumulation of innovative algorithms is hard to explain because complex concepts are difficult to pass on. We found that selective social learning preserved rare discoveries of exceptional algorithms in a large experimental simulation of cultural evolution.” A report published in Science by Thompson and colleagues deploys an experimental approach to examine the cultural processes by which humans develop and preserve problem-solving algorithms over long spans of time.
- “Apart from their roles in human infectious diseases, we understand relatively little about RNA viruses in the wider world. Recently, the discovery curve has been spectacular and has revealed unexpected diversity…This is not just a numbers game; the authors also found a missing link in RNA virus evolution and discovered new phyla that dominate in the oceans and might infect mitochondria.” A study by Zayed and colleagues published in Science reports on the discovery of a bonanza of hitherto-unknown RNA viruses aswim in the world’s oceans.
COMMUNICATIONS & DIGITAL SOCIETY
- “…researchers are learning how to rework CVs to emphasize quality over quantity, and to include narratives about their broader impact. Meanwhile, hiring panels and grant evaluators need to rethink how best to assess these documents.” In a career feature at Nature, Chris Woolston shows researchers how to blow the dust off of the venerable academic CV and polish it into something a bit more suitable for highlighting their work.
- “In China, girls tend to outperform boys in maths during middle school despite a commonly held idea that boys are innately superior in scientific and technical subjects…By chance, some pupils were assigned to classrooms with a greater proportion of parents holding [the view that boys are innately superior at mathematics]…. Girls in these classrooms performed less well in maths exams than did girls in other classrooms, whereas the scores of boys in these classrooms increased.” A study published in Nature Human Behavior by Eble and Hu examined the link between parental expectations about student performance, gender, and actual performance.
- “Two surveys of scientists’ experiences of harassment during the pandemic, from personal insults to death threats, suggest a dispiriting, if unsurprising, trend: researchers with greater news-media prominence are more likely to be harassed.” At Nature news, Richard Van Noorden reports on a pandemic-related survey study that found that the more scientists were in the public eye, the more likely they were to be targeted for online harassment.
- “At a moment when students are reeling from two years of pandemic learning and isolation from their peers, the idea of intentionally making young people uncomfortable may seem misguided. But many educators and learning scientists say that now, as students look to rebuild academic confidence, is a crucial moment for teachers and parents to step back when learning gets hard and to be explicit that the challenge offers rewards.” The New York Times’ Jenny Anderson takes a look at the paradoxical benefits of allowing students to experience a degree of struggle as they contend with academic challenges.
- “While exploring my family history in Massachusetts, I made the sickening discovery that my late twin sister, Karen, might have been one of the test subjects in these experiments….Was she one of the 20 infants and children 1–3 years old mentioned in the Science paper? Due to poor record keeping, I’ll never know.” An editorial by David W. Christianson in Chemical and Engineering News grapples with the responsibilities of scientific journals to hold themselves accountable for their roles in perpetuating abuses of human rights in the name of scientific inquiry.
- “We are at a pivotal moment in the development of healthcare artificial intelligence (AI), a point at which enthusiasm for machine learning has not caught up with the scientific evidence to support the equity and accuracy of diagnostic and therapeutic algorithms.” A paper just published in BMJ Health Care and Informatics by Cerrato, Halamka, and Pencina advances a proposal for shared platform capable of assessing clinical algorithms for fairness and accuracy.