New Truths, Old Lies ?
Managing shifting consensus in COVID-19 times
Authors: Zoe Huczok and Renee DiResta (Stanford Internet Observatory)
Adapting to evolving scientific consensus in the public view will require transparency in communication, a public forum with clear guidelines for deliberation, and policy responsiveness to scientific findings. In these respects, Wikipedia, the online encyclopedia, may offer inspiration to public health communicators.
Building consensus, a slow process
Scientific consensus takes time to build. It occurs when there is solid agreement on a proposition among an overwhelming majority of the scientific community, based on the body of available evidence. Such an agreement requires gathering facts, analyzing data, thoroughly testing hypotheses, and deliberation across the research community.
Sometimes, consensus opinions evolve slowly — or are overturned — over time. The consensus that tobacco consumption was harmful, for example, took decades to build. In the 1930s tobacco brands cited medical “evidence,” and incorporated physicians into their advertisements to argue for the harmlessness and relative merits of their products (see Figure 1). Then, a 1950 study of 684 lung cancer cases by researchers Ernest Wynder and Evarts Graham published in the Journal of the American Medical Association established a correlation between smoking and lung cancer. A larger body of research gradually came to similar conclusions. Despite these findings C. C. Little, D.Sc., chairman of the Tobacco Industry Research Committee —the Tobacco industry’s trade groups— wrote in the New York Times in 1955 that “the possible relationship between cigarette smoking and lung cancer [...] is still an unknown factor.” As late as 1961, the New England Journal of Medicine recounted a debate between Wynder and Little in an editorial, however, the journal did not give a definitive conclusion on the harmfulness of tobacco consumption. Not until 1964 did the Surgeon General start publishing reports on smoking and health, indicating strong scientific and political consensus that smoking is harmful.
Figure 1: Lucky Strike ad, 1930
The COVID-19 crisis is the latest example of shifting and evolving scientific consensus - and this time, the process is being widely debated on social media platforms with the participation of the broader public. For example, the question of whether Sars-Cov-2 could propagate through aerosolized particles was a matter of considerable debate — and one with major implications for public health policy, including social distancing and mask guidance. The airborne transmission hypothesis was largely sidelined at the start of the pandemic, but the situation later evolved :
March 28, 2020 - the World Health Organization tweets: “FACT: #COVID19 is NOT airborne”
April 3, 2020 - the WHO consults with aerosol experts – but the evidence in favor of the airborne route is deemed insufficient
July 6, 2020 - in an open letter, 239 doctors appealed to the WHO to revise its position
April 30, 2021 - the WHO indicated a change in its consensus by updating its Q&A page on virus transmission to include aerosols among modes of transmission
May 8, 2021 - after two reversals, the CDC also updates its guidance
Today, most scientists agree that COVID-19 can be transmitted through aerosolized particles, though they still disagree on how often this happens.
A very public hesitation
Nowadays, the process of coming to an understanding happens in full view of the public. The back and forth and developmental nature of building this consensus can erode the public’s confidence in scientific authority. Since the beginning of the pandemic, medical researchers tried to understand COVID-19 even as a global public was anxiously demanding definitive answers on the best protection, treatments, and the possibility of vaccines. Scientific consensus has shifted on the origin of the virus, the mechanism of transmission and the efficacy of cures (such as hydroxychloroquine). Yet when consensus is overturned as new facts come to light that invalidate prior convictions, the tacit bond of trust between the scientific community and the public is challenged.
To follow the scientific conversation, consumers of medical news needed to master scientific concepts (e.g. immunity, mRNA) and methods (e.g., vaccine trials). Lacking the tools to decipher the leading consensus among medical experts, evaluate the premises, and understand how specific updates fit among a wealth of changing information, the public naturally became confused. Confusion fed fear. To make matters worse, policies related to COVID-19 evolved with the latest scientific findings, occasionally requiring citizens to change their lifestyle and habits — sometimes with a few days’ notice. For example, since February 2020, the UK closed and reopened schools twice, based partly on shifts in the understanding of indoor transmission risks.
The Virality Project’s research shows that this critical process of changes, updates, and reversals in scientific consensus can be misused to fuel vaccine hesitancy and engender distrust in medical experts. It offers a window for anti-vaccine activists and influencers to claim that experts and policy-makers are incompetent (see first image in Figure 2), malicious (see center image in Figure 2), or complicit in a suspected collusion between government, “Big Tech” and “Big Pharma” (see third image in Figure 2). In a recent blog post, analysts wrote about the common narratives and tropes used to undermine experts. However, allegations of conspiracy and malice are not limited to the science itself, but also to the moderation of narratives on social media. For example, after Facebook announced it will no longer remove posts claiming that COVID-19 is “man-made or manufactured”, Sen. Ted Cruz responded that it was “now clear” that Facebook had been “utilizing their monopoly position to censor on behalf of the government.”
Figure 2. Instagram post claiming incompetence from perceived inconsistencies in policy (left). Instagram post (now removed) claiming “they” lied about a variety of topics on which scientific consensus has historically shifted (center). Tweet by Ted Cruz claiming that the evolution in scientific consensus over the origins of COVID-19 amounts to a “cover-up” (right).
The Wikipedia model: a way forward for public health communicators
Evolving consensus in full view of the public is here to stay, along with contradicting claims by influential detractors who are equally, if not more, visible on social media. Framing shifts in consensus as incompetence, malevolence, and conspiracy is destructive. Platforms are being asked to reckon with these evolving dynamics as they attempt to moderate misinformation. How, then, can we engender trust in an environment of evolving consensus?
One platform response we have observed recently is an attempt to signal to users that consensus has not yet stabilized on a particular topic. In June 2021, Google started testing a new notification for search results that are changing quickly – either because the topic is new, or because it is rapidly evolving. The warning also implies that more information may come later, encouraging users to repeat the search at some later time, and thus building search literacy. This policy intervention is potentially very useful in breaking news situations. However, there is still an open question of what to surface as the consensus continues to evolve past a breaking news event, or how to explain why something may have changed weeks or months after the fact.
Wikipedia may offer a model for showcasing the iterative progress of medical science, in a way that is accessible and maintains trust. Wikipedia has become one of the most comprehensive sources of authoritative knowledge today, even being cited in court. Several studies (Mothe and Sahut 2018, Shen et al. 2012) demonstrate high overall levels of trust in Wikipedia among students of various educational levels. During the COVID-19 pandemic, Wikipedia has been a major source of health information, with 5,000 editors dedicated to the COVID-19 tracking effort and 85 million edits. Its model for reaching consensus is defined by at least three features: transparency, an emphasis on deliberation, and clear rules for conflict resolution.
Figure 3: An example of a Google Search warning for rapidly evolving results shown in Google’s blog post from June 2021.
Wikipedia offers full transparency into its editing process. Under the tab, “View History” on Wikipedia all edits made to the article since its creation are listed. The “talk” page of each article also features the details of discussions between editors: the reader can thus access the process through which consensus was formed, and the arguments being made on both sides of the dispute. For example, the talk page for the article “Investigations into the origins of COVID-19” is seemingly endless, reflecting the divisiveness of the topic. However, Wikipedia’s dynamic edit culture is counterbalanced by friction built into the editorial mechanism, including limits on numbers of reverts on a single page, on article creation, and options for the locking (or “protection”) of controversial pages.
The Wikipedia community emphasizes deliberation with an end goal of building consensus. Each Wikipedia language community agrees on a list of reliable sources to be referenced in articles. Thus, preprint archives, or journalistic sources that do not adequately differentiate between news and opinions, cannot be cited on the English-language Wikipedia. Editors strive to make decisions in a consensual way. For example, in August 2020, editors chose not to mention Steve Bannon’s arrest and indictment for fraud in the “We Build the Wall” campaign in the first paragraph of Bannon’s page. After extensive discussion, it was agreed that the fact would be included further down in the article, unless Bannon was charged.
Conflict between editors is occasionally not resolved through consensus - in those instances, the decision is put to a majoritarian decision process. For example, as the disagreement on the origins of COVID-19 persisted, voluntary editor Roberto Fortich filed a “request for comment” on whether to include the lab leak theory in the Wikipedia article. As is customary in those cases, Wiki voluntary editors were called to present their arguments and vote. 13 out of 19 editors opposed the inclusion of the theory. As a result, a public note was made that the theory will not be mentioned in Wikipedia articles about COVID-19.
Ways forward for public health communicators
Public health communicators can learn from Wikipedia on version management, engagement with contradictory debate, and conflict resolution. Lessons on transparency have already been heeded by some major institutions. For example, the CDC allows partial transparency into the history of its science briefs, summaries of recent publications and their findings, by indicating the date and rationale for the latest updates at the top of the brief page. However, older versions (e.g. 6 months old) are not generally accessible. The generalization of dating versions of different reports and reasonings may help bolster public trust and pre-bunk the hesitancy narratives that build off perceived inconsistency. Including a warning that policy may change in the coming weeks or months, as new evidence comes to light- as the CDC recommends in its communication guidelines - would serve the same purpose.
Figure 4: Excerpt of the CDC’s Scientific Brief on SARS-CoV-2 Transmission
Building scientific consensus, not only in public, but under the auspices of public health institutions, would signal authorities’ willingness to submit their decisions to citizens’ oversight. A radical version of this effort could see public authorities host participatory forums for editing the most current or contested COVID-19 information pages - either on their own websites, or on an affiliated domain. The pool of editors would include a wide range of medical experts representing a plurality of medical views on the topic at hand (selection bias here would be a major threat to the credibility of the enterprise). Clear rules for engagement, similar to those established by Wikipedia, would govern the forum. A consensus-seeking discussion between editors would establish lists of credible sources and edit articles. Only when the consensual approach fails would edits be brought to a majority decision among editors.
In cases where a majoritarian choice must be made, the composition of the majority would be transparently visualized - showing, e.g., how editors’ views cluster on issues such as airborne transmission. Such a visualization would ensure that minority opinions are neither rendered invisible nor amplified out of proportion. Opinion diagrams have already been successfully implemented by Taiwanese civic discussion platform vTaiwan (see Figure 5). By hosting these processes, authorities would indicate their willingness to engage in a new form of bureaucratic oversight, where citizens can observe policy-makers’ responsiveness to scientific consensus. Policy-makers’ decisions would be auditable by the general public, which would mitigate accusations of malevolence or conspiracy - e.g., that COVID-19 restrictions obey a hidden political agenda.
Figure 5 : Example of opinion visualization on the platform vTaiwan
Thus, the Wikipedia model offers a method for public health communicators to convey the scientific consensus-building process in a transparent, accessible, and inclusive way, and to ground policies in the latest agreed-upon findings. The model helps address the specific brand of distrust caused by the evolution in scientific and political discourse - a step towards improving confidence in health policy more broadly.