image of the covid-19 spike protein
An arrow pointing leftHome

Increasing public confidence in COVID-19 vaccines using NLP

  • Alexander Gary

Two professors at Columbia University have set out to prove just that. TL;DR: It’s all in the messaging

Two professors are looking to the AI community in an attempt to encourage vaccine skeptics to reconsider their positions. Dennis Tenen, a Professor of English and Comparative Literature, and Rishi Goyal, a Professor of Emergency Medicine, both at Columbia University in New York City, are behind Increasing COVID-19 Vaccine Confidence, a new project that searches for patterns in the rhetoric of online anti-vaccine protestors which then automatically formulates the right combination of words that just might change hearts and minds.

Their task isn’t easy. With so much misinformation and disinformation about both Covid and the various vaccines available to the public, a total of 8 in 10 Americans who hasn’t been vaccinated does not intend to get one in the future. The reasons vary widely — from the notion that individual liberty surpasses public health, or that vaccines interfere with our right to practice religious freedoms, or on a belief in outlandish conspiracy theories. Some fears are understandable; for example, the United States Public Health Service infamously did use Black men as test subjects in Tuskegee, Alabama, lying to them for decades about the true nature of the syphilis study they were performing on them. Nonetheless, breaking through to the intransigent isn’t easy, no matter their gripe. Tenen and Goyal are looking to AI platforms to help.

As laid out on the Columbia University’s World Projects page, the program goes something like this: Tenen, Goyal, and a team of scholars and professors work with scientists, medical professionals, the Maine Department of Health and Human Services, and Ulster County, New York, to collect millions of community posts from websites like Facebook, Reddit, Twitter, and YouTube. The dialogue of skeptics is fed into a computer, where it is treated as measurable data and looks for emergent patterns using an AI algorithm. The data from these posts is then rewritten to explain the shot in words and phrases that will resonate with anti-vaxxers. In essence, Goyal and Tenen are speaking the anti-vaccine protestors’ language, which is something many scientists, politicians, public health officials, and even influential celebrities haven’t been able to do in the nearly two years since the pandemic reached the United States.

Finding similarities between anti-vaccine arguments and other controversial topics that evoke reflexive responses is what the algorithm is after. “Language, syntax, and rhetoric used to express hesitancy has been borrowed from other debates, usually over divisive political issues,” Tenen and Goyal wrote in an LA Times Op-Ed in May. “Families invoke ‘natural parenting’ in worrying about [the vaccines’] long-term effects…political activists speak of ‘medical freedoms’ and personal rights…freethinkers who distrust authority are inclined to ‘do their own research’…those who have had an uneasy relationship with the medical establishment in the past…can be reluctant to trust the safety.”

For example, those who are religious (provably or otherwise) may try to claim that religious or philosophical beliefs should exempt them from the vaccine — such as parents who may have needed to use this status for their schoolchildren in a pre-Covid world. The invocation of the feminist phrase, “My body, my choice” (referring to individual reproductive choices and sexual autonomy), has since been co-opted by the anti-vaccine movement to suggest that women’s rights and the choice not to vaccinate hold similar weight.

Perhaps most appallingly, some proponents of the anti-vaccine movement have compared vaccine mandates to life in the Holocaust as Congresswoman Marjorie Taylor Greene had done in July — likening President Biden’s vaccine pushes to Nazi Germany-era styles of enforcement.

According to the Increasing COVID Vaccine Confidence project’s team, talking points such as these are taken from the rhetoric of other politically sensitive debates (like freedom of religion, or the right to a safe abortion) because those debates have demonstrated the ability to transform how sympathetic and understanding we become and how it reshapes our thinking of the experience. The team says this is where an AI program becomes entirely necessary: Experts have attempted to lean on scientific research and cold, hard facts to explain the nature of the pandemic to the public, but this method has yet to reach an audience that isn’t swayed by data. The algorithm will translate millions of messages against taking the vaccine and flip them on their head, relying on similar sentiments and phrases to appeal to the feelings of reluctant shot-takers. For example, the algorithm may tell a person who believes they have a right to reconsider getting a vaccine this: “Don’t let anybody take away your right to get the vaccine you deserve.” This reflects language that is familiar and empowering. “This approach [can] be adopted by other local, state, and federal officials, and elevate the role of data in driving public health messaging,” the professors write. “[This can] demonstrate how technology can be harnessed to address some of the problems that it has exacerbated.”

Everyone is hoping to see the end of this pandemic, and this team at Columbia might be able to use their AI project to reach loads of holdouts in our ongoing fight with Covid. The good news is that they won’t be the only group out there using AI algorithms to persuade the unvaccinated.

Since 2016, the Center for Disease Control and Prevention and the Food and Drug Administration have worked together to use natural language processing to better study data and respond to any vaccine hesitant individuals who may be waiting for vaccines (of any kind) be explained in clearer terms. The Tennessee-based behavior-change AI platform Lirio is currently collaborating with health care systems to use their data and the various forms of communication patients have to create an individualized message tailored to still-reluctant patients. The Johns Hopkins Bloomberg School of Public Health has teamed up with IBM to create a public AI chatbot that can give up to 300 various responses to 150 of the most common Covid-vaccine-related questions people may have []. Its name? The Vaccine Information Resource Assistant (or, “Vira” ). Artificial intelligence programs have the power to not only boost our knowledge, but maybe even literally save our world. Here’s hoping.