Forge AI Health

Friday Roundup

The Forge AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.

October 22, 2021

In today’s Roundup: the pitfalls of oracular AI; climate change and its impact on almost every aspect of human health; big data, small data, and future directions for machine learning; the limits of tech whistleblowing; the effects of redacting identifying information on NIH grants; “universal animals” illuminate links between embodiment and intelligence; dataviz considered as superpowers; returning narrative to scientific publishing; much more:

Deep Breaths

“This ancient technology — scuffed and dinged, the lathe looks like something from a World War II submarine — is a key part of Joyful Noise’s strategy to survive the very surge of vinyl popularity the label has helped fuel.” As we get closer to Halloween, this New York Times story by Ben Sisario about old technologies rising from the dead seems especially timely. (Also, we just miss vinyl LPs.) (H/T @GYamey).

AI, STATISTICS & DATA SCIENCE

“We already have a tendency to frame AI systems in mystical terms — as unknowable entities that tap into higher forms of knowledge — and the presentation of Ask Delphi as a literal oracle encourages such an interpretation. From a more mechanical perspective, the system also offers all the addictive certainty of a Magic 8-Ball.” At The Verge, James Vincent explores a remarkable (and, at intervals, alarming) experiment in machine learning ethics: Delphi, a machine learning model trained on a database of human ethical judgments on a wide variety of situations and designed to provide – like its oracular namesake – a definitive response. In doing so, this real-world experiment is revealing significant shortcomings and biases, many of which can be manipulated by minor alterations in how questions are presented.
“The Commission recommendations make clear that, in our current system, data on health inequities are divorced from the history and community conditions that shape poor health outcomes, resulting in an incomplete picture of who is most impacted and why.” Earlier this week, the Robert Wood Johnson Foundation published a “roadmap” (which includes multiple funding opportunities) for creating a public health data infrastructure built around principles of equity.
“Small data approaches such as transfer learning offer numerous advantages over more data-intensive methods. By enabling the use of AI with less data, they can bolster progress in areas where little or no data exist, such as in forecasting natural hazards that occur relatively rarely or in predicting the risk of disease for a population set that does not have digital health records.” An opinion article at Scientific American by Georgetown’s Husanjot Chahal and Helen Toner explain the principles behind “fine-tuning” for AI – an approach (also called transfer learning) that eschews that massive quantities of “big data” associated with multiple other machine learning training methods.
“The practice of hiding human input in AI systems still remains an open secret among those who work in machine learning and AI. A 2019 analysis of tech startups in Europe by London-based MMC Ventures even found that 40% of purported AI startups showed no evidence of actually using artificial intelligence in their products.” Bloomberg’s Parmy Olson writes about the pervasive hype around many “artificial intelligence” startups – and the often overlooked fact that much of this “AI” is actually simply people being exploited via a phenomenon known as “microwork.”
“…there is a broader data crisis in machine learning. As machine learning datasets expand, they increasingly infringe on privacy by using images, text, or other material scraped without user consent; recycle toxic content; and are the source of other, more unpredictable biases and misjudgments.” An article at Undark by John McQuaid questions the societal and individual costs imposed by the ever-growing hunger for huge datasets to feed into machine learning applications.
“AI typically focuses only on the mind part, building machines to do tasks that can be mastered without a body, such as using language, recognizing images, and playing video games. But this limited repertoire could soon get old. Wrapping AIs in bodies that are adapted to specific tasks could make it easier for them to learn a wide range of new skills.” A fascinating article by MIT Technology Review’s Will Douglas Heaven highlights research using virtual “universal animals” that may reveal connections between embodiment and intelligence.
“The evolution of digital advertising has gone beyond accepted principles to find the right customer for the right product to outright surveillance of individuals.” An article at Forbes by Hessie Jones examines the privacy threats posed by an advertising (and data-mining) business strategy known as “real-time bidding.”
“Considering visualization through the language of superpowers can help us think about new opportunities for vis and what it means for a visualization to make us feel “empowered”. It also illustrates how the visualization community can look outward for new sources of inspiration, embracing design strategies like futuring that are increasingly common elsewhere, including in human-computer interaction and design.” A Medium post by a University of Calgary-based data visualization lab explores some unexpected (by us, anyway) resonances between dataviz design and comic-book superpowers.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

“The 44 indicators of this report expose an unabated rise in the health impacts of climate change and the current health consequences of the delayed and inconsistent response of countries around the globe…” A major report by Romanello and colleagues out this week in Lancet takes a sweeping view of the potential health harms that may (and in some cases, already are) emanating from climate change.
“The results suggest that the average reinfection risk rises from about 5% four months after initial infection to 50% by 17 months. Overall, natural protection seems to last for less than half as long as it does for the three common-cold coronaviruses.” Nature’s Lynne Peeples reports on recent investigations that suggest that people who have recovered from a COVID infection but do not then get vaccinated or practice masking (or other mitigation strategies) have a substantial risk of reinfection over subsequent months.
STAT News’ Damian Garde and Matthew Herper report on the unexpected failure of a clinical trial of an antiviral therapy for COVID, and examine whether trial design issues may have played a role.
“Current efforts by the vaping industry, government agencies, and schools have thus far proved insufficient to stop the rapid spread of nicotine vaping among adolescents.” Recent correspondence by Miech and colleagues published in the New England Journal of Medicine reflects concern about an uptick in adolescent use of nicotine, including via vaping products (H/T @NHLBI_translate).
“The same money that can yield evidence and life-altering treatments, it turns out, can also end up eroding trust and deepening patients’ alienation. The medical system hasn’t inspired much faith in the fibro community to begin with. If you’ve spent year after year seeing doctor after doctor, trying test after test and treatment after treatment, an unfulfilled promise can seem less like a one-off and more like yet another data point in a pattern.” A deeply reported story by STAT News’ Eric Boodman wades into the murky waters surrounding a controversial genetic test and related clinical trials for a poorly understood condition – in this case, fibromyalgia.
“Nine months into this new role, Harllee Harper’s most powerful tool is the mayor’s initiative, Building Blocks. Drawing on public health strategies to contain the spread of gun violence, it’s designed to treat the immediate symptoms and root causes of community violence…Its workers operate almost as contact tracers, whose methods have become familiar during the pandemic.” An article at Kaiser Health News by Amanda Michelle Gomez profiles an attempt to apply public-health strategies to mitigating gun violence in Washington, DC.

COMMUNICATIONS & DIGITAL SOCIETY

“We argue that a return to narrative in scientific writing is not incompatible with rigour and objectivity; it can mitigate information overload and achieve the core purpose of publication: to communicate.” A paper recently published in Nature Human Behavior by Croxson and colleagues makes the case for reintegrating narrative – that is to say, storytelling – to formal scientific publications from which it has long been exiled.
“…whistleblowing can’t save us, because the issue isn’t an absence of information but an absence of will. And what builds will, and shifts norms, doesn’t look like a single, isolated figure speaking truth, but mass movements of people setting new standards and making clear there are costs to regulators and companies for not attending to them.” An essay at Wired by Os Keyes takes a pessimistic view of outcomes likely to result from recent whistleblower episodes at big tech companies, including (most recently) Facebook.
“Science watchdogs have always worked alone on the periphery of the research enterprise. The pandemic is laying bare how vulnerable — and vital — they are.” Buzzfeed News’ Stephanie M. Lee documents the latest wrinkles in a battle between bad-science sleuth Elisabeth Bik and French scientist Didier Raoult, whose studies touting the supposed benefits of hydroxychloroquine for COVID have drawn intense and critical scrutiny from Bik (among others).
“Our culture does not deal well with pain. We are privileged to live in a society where diverse worldviews and faiths can be held freely. Yet we discuss them so rarely, when all is well in our private world, that in moments of shattering loss there is a risk that the fragile illusion of unity will break.” A personal essay published in BMJ by Ruth Moore explores the connections between grief and chronic pain.
“Redaction reduced, but did not eliminate, reviewers’ ability to correctly guess features of identity.” A study published in eLife by Nakamura and colleagues examined whether obscuring data that could identify NIH grant applicants could potentially be effective in eliminating a source of bias in grant evaluation (H/T @drugmonkeyblog).
A study (published in PLOS One by Ruzi and colleagues) by authors from North Carolina State University describes some dividends with audiences when scientists discuss their own work publicly rather than relying solely on third-party communicators.
“Each of the later experiments tended to include groups that repeated the procedure involved in the first experiment, and Ward combined all of this data to produce an internal meta-analysis that included over 1,900 subjects. The tendency for Google users to behave as if they had demonstrated superior recall was very consistent.” Ars Technica’s John Timmer describes recent research that probes where our brains stop and Google search starts – and the tendency to mistake the latter for the former.

POLICY

“But for all the noise they create and attention they draw, sirens, combined with emergency lights and speeding, can be a force multiplier for more harm than good, some experts said.” At the New York Times, Christopher Mele reports on research that increasingly suggests that the benefits from using flashing lights and sirens to speed response to and transport from medical emergencies may be somewhere between marginal and nonexistent.
“Using a stylized model of grocery demand, we estimate that, by 2016, the indirect benefit had reduced grocery costs for the median household by approximately 4.5%.” A working paper by Jessie Handbury and Sarah Moshary at the National Bureau of Economic Research examines some of the follow-on effects of relatively recent expansions to the National School Lunch Program.
“…Black and Latinx communities and individuals, who already experience health and economic disparities caused by structural racism, are overrepresented among the Medicaid population and are at a greater risk of losing coverage due to churn. For states, churning increases administrative burdens and costs for Medicaid programs.” A post at the Harvard Petrie-Flom law school’s Bill of Health blog by Cathy Zhang explores lessons from the COVID pandemic in preventing harms resulting from “churn” created by changes in health insurance coverage.
“In the four studies, compared to free choice, requirements strengthened vaccination intentions across racial and ethnic groups, across studies, and across levels of trait psychological reactance. The results consistently suggest that fears of a backlash against vaccine mandates may be unfounded and that requirements will promote COVID-19 vaccine uptake in the United States.” A study published this week in Scientific Reports by Albarracin and colleagues suggests that vaccine mandates are an effective means not just of increasing vaccination rates, but of influencing intentions to get vaccinated.