New AI Tool Can Predict People’s Vaccine Attitudes

Andrew Corselli

UC College of Pharmacy students administer the COVID-19 vaccine during a drive-through clinic sponsored by UC Health in this 2020 file image. (Image: Colleen Kelley/UC Marketing + Brand)

A powerful new tool in artificial intelligence can predict whether someone is willing to be vaccinated against COVID-19. The predictive system uses a small set of data from demographics and personal judgments such as aversion to risk or loss.

A team led by researchers at the University of Cincinnati and Northwestern University has created a predictive model using an integrated system of mathematical equations describing the lawful patterns in reward and aversion judgment with machine learning.

“We used a small number of variables and minimal computational resources to make predictions,” said Lead Author Nicole Vike. “COVID-19 is unlikely to be the last pandemic we see in the next decades. Having a new form of AI for prediction in public health provides a valuable tool that could help prepare hospitals for predicting vaccination rates and consequential infection rates.”

The study was published in the Journal of Medical Internet Research Public Health and Surveillance.

Here is an exclusive Tech Briefs interview — edited for length and clarity — with Vike.

Tech Briefs: What was the biggest technical challenge you faced while developing this technology?

Vike: Determining which machine learning framework to use was probably the biggest technical hurdle. As you can imagine, there are almost limitless frameworks to choose from. At the end of the day, we wanted to ensure our approach was achievable without large computational demands, so we focused on traditional machine learning frameworks. We then compared the performance of multiple approaches to determine what would be best for the current study, and for future work.

Tech Briefs: Can you explain in simple terms how it works?

Vike: Participants complete a simple task where they are asked to rate how much they like or dislike pictures. These pictures range from aggressive animals to beautiful landscapes. We then model these ratings into graphs using a framework called ‘relative preference theory.’ This framework has been validated previously to produce graphical profiles of how people make judgments about rewarding and aversive things. From these graphs, we derived 15 features that mathematically define these behaviors. We combined these features with some basic demographic information (age, income level, etc.) in a machine learning framework to make our predictions.

Tech Briefs: The article I read quotes Co-Senior Author Aggelos Katsaggelos as saying, “It can work very simply. It doesn’t need super-computation, it’s inexpensive and can be applied with anyone who has a smartphone.” How soon could we see this applied at a commercial scale?

Vike: The goal is to build an automated algorithm that collects picture ratings, derives judgment features using relative preference theory, and then uses those features (with demographics) to make predictions — whether it be related to vaccination, mental health conditions, cognitive dysfunction, the list goes on. Once automated, really any corporation — whether it be private, government, or healthcare — could disseminate the picture rating task to their populations of interest since the only limitation is access to a smart device. Our backend would collect the rating data and automatically produce prediction metrics.

Tech Briefs: Katsaggelos goes on to say, “It is likely you will be seeing other applications regarding alterations in judgment in the very near future.” Can you elaborate on that please (how soon, what kind of such applications, etc.)?

Vike: We are actively producing papers that use this same framework to predict indicators of different mental health conditions — depression, anxiety, substance use disorders, suicidal thoughts, and behaviors — as well as indicators of cognitive decline. Model performances are rooted in the use of psychologically based features that are derived with relative preference theory — think economic variables like risk aversion and loss aversion, but, instead, they are derived from judgments that precede decisions instead of the decisions themselves. Together, we refer to this approach as Computational Cognition AI, or Comp Cog AI, and our results support the concept of a standard model of the human mind.

Tech Briefs: Going from that, what are your next steps? Do you have any plans for further research/work/etc.?

Vike: We are actively working on the prediction of various mental health disorders, the destructive behaviors associated with them, and indicators of early cognitive dysfunction. We plan to validate our findings with additional datasets which will also improve our predictive capabilities.

Further, we want to explore how judgment profiles vary across mental health disorders with the goal of better understanding the relationships between judgments and mental health. This could have broader applications to improving therapeutic strategies.

Tech Briefs: Do you have any advice for engineers/researchers aiming to bring their ideas to fruition (broadly speaking)?

Vike: Have a deep understanding of what you’re trying to accomplish and keep your approach focused and quantitative. A lot of approaches throw a bunch of data into a prediction framework but that leaves little room for interpretation and limits adoptability at a large scale. Since we were focused on predicting behaviors, we honed in on variables that may underly such behaviors — keeping in mind that any related task we used had to be simple and easy to disseminate across a large population.

Tech Briefs: Anything else you’d like to add that I didn’t touch upon?

Vike: The power of this approach goes beyond the prediction of vaccine behaviors. It supports the use of the described judgment variables to predict numerous conditions with distinct changes in behavior — depression, anxiety, cognitive decline, the list goes on. The broader goal of this work is to identify at risk persons or populations using these behavioral features that have characteristic patterns for various mental, cognitive, and behavioral health conditions.