NOVID: Personal Pandemic Network Radar

NOVID is fundamentally different approach to controlling disease spread. It is a freely downloadable application that alerts people before they have been in contact with a person infected with COVID-19, at which point they can still protect themselves as well as other people. Instead of large-scale lockdowns, this system would automatically facilitate pockets of increased caution (and reduced transmission) within several physical contacts from each positive case; only the people who have to be concerned, are concerned. This could allow countries to maintain social and economic activity while awaiting a vaccine and give healthy people the opportunity to increase caution when necessary.

NOVID recruits an entire population of individuals to help fight a pandemic, not only to protect each other, but also to protect themselves. Rather than trying to appeal to a crowd or a population, NOVID understands and works in tandem with the incentives of each person.

The current pandemic demonstrated that behavior is a major factor in pandemic control. Before vaccines and therapeutics are available, contact tracing and quarantine are workhorses for control. However, they have an incentive misalignment problem: directly involved participants (infected individuals and their contacts) are asked to protect the rest of society from themselves. Their quarantine compliance does not directly help them avoid infection (they already were in contact with infected individuals, or positive).

Our solution focuses on the incentive alignment problem, providing a new type of intervention with the distinctive property that even selfish behavior contributes to community control. This addresses the “tragedy of the commons”. We analyze interaction data in a novel way for Category 2a: “Identify and incentivize preventive interventions.”

The scale of this problem is massive. We focus on populations where most households have a smartphone. While some cultures’ citizens align with government directives (often aided by technology) and cooperate with scalable contact tracing and enforced quarantine, most of the world is not that way. For example, in a paper by Smith et al. studying the UK’s Test, Trace, and Isolate system, only 11% of quarantined individuals declared proper adherence. Our solution would directly benefit billions of people.

We invented and implemented a new smartphone-facilitated behavioral intervention which analyzes large scale interaction data in a novel way. It uses game theory and network theory to reverse incentives so that even selfish behavior contributes to societal disease control. 

The new approach is: for each infection, don't just identify and quarantine direct contacts. Instead, tell everyone how many physical relationships separate them from each infection ("3" means you were with someone who was with someone who was with someone positive). The physical relationship network is automatically collected via Bluetooth communication between nearby smartphones running the NOVID app, with each user represented by a random identifier. There is no relationship between these random identifiers and personal data such as phone numbers, GPS coordinates, or smartphone hardware identifiers.

This fundamentally differs from every other app, and from contact tracing itself. The old approach tells you after exposure, hoping you altruistically self-quarantine to protect others. Our new approach warns before exposure, when your selfish incentive is self defense: avoid optional contact, or wear better masks. It’s the first personal radar for pandemics, quantifying your proximity to infections by counting physical relationships separating you in the network.

Our solution serves all households with smartphone access, addressing their need to avoid getting infected, while still interacting with the outside world. Many households cannot isolate for the entire pandemic, and we provide them with more precise information that they can use to budget their caution. It is not sufficient simply to know how many positive cases are in a city, because when there are few cases and the city opens up, people will get infected until cases grow out of control, and the city shuts down again. In contrast, when there are few cases, our solution dynamically alerts the socially-nearby people, helping them avoid getting infected, and helping the economy remain open.

In countries with high smartphone penetration, this solution serves everyone from schoolchildren (parents’ apps can link to their children’s classrooms) to the workforce. To understand needs, our team’s core has a User Experience Design group, which drives all feature development. They are frequently on discovery calls with institutions ranging from schools to employers to cities to nursing homes, where we seek to understand their needs. We also spoke with hundreds of users in our early experimental deployments, and have also used our app itself to survey users.

Privacy is of paramount importance to the success of this intervention. To maintain privacy, users voluntarily enter positivity information, verified by confidential passcodes from authorities. This is our only point that needs altruism. Fortunately, contact tracing apps showed that a nontrivial percentage of people voluntarily enter their positivity. Yet we go another step beyond all other apps: you can also tell NOVID you were a manually contact-traced contact of a positive case. Since our approach focuses on how many relationships separate you from infections, it is still useful for you to learn that someone 2 relationships from you is a contact-traced contact of someone positive, because then someone positive is within 3 relationships of you. This boosts the opportunities for self-reporting by an order of magnitude over the status quo.

NOVID represents an ambitious and scalable approach to preventive intervention, coordinating selfish human behavior using technology. It is powered by analysis of interaction data crowdsourced from a mass deployment incentivized by self-defense instincts. Regions which currently have low smartphone penetration are seeing rapid smartphone growth, and so by the next pandemic, the knowledge gained from our solution will have a global practical deployment reach.

We are engaging in the significant experimental work involved in understanding how to harness the full potential of this new approach, both in terms of the app itself, as well as in strategies to maximize uptake.

The NOVID app is freely downloadable. It automatically detects nearby NOVID users, and after an app user enters a signal of positivity, it automatically informs other users how many physical relationships separate them from the signal. This alone is quite nontrivial, because several national governments tried unsuccessfully to create apps which permitted iPhones to detect iPhones via Bluetooth even when the apps were not on screen.

NOVID was installed over 100,000 times, and was experimentally used at Georgia Tech and Carnegie Mellon University. No site has yet conducted a full deployment using all features of the NOVID system, in part because our iPhone breakthrough is new. Georgia Tech had 1,000 users interconnected with each other, and CMU had 400 (CMU had ~1,200 students living on campus). These partial deployments validated that the core technologies work, and signals of positivity were relayed and anecdotally influenced behavior.

Our solution discovered a new way to use the physical contact network data together with smartphone technology to align selfish incentives at scale. We did this at a time when a vast number of contact tracing apps had sprung up from teams all around the world, competing among one another with very similar solutions. Yet we alone developed this alternative approach. We recognized that there was a fundamental incentive misalignment with anonymous contact tracing apps, because anonymous and voluntary self-quarantine requires a high degree of altruism as compared to non-anonymous manual contact tracing. Since by then Apple and Google had already created their framework to only deliver the traditional digital contact tracing experience, we then needed to build a smartphone proximity sensing infrastructure from scratch. Even though national governments had been unsuccessful, we persevered, and ultimately created the only practical alternative framework, upon which we implemented our alternative “pandemic network radar” approach.

Currently, NOVID is installed on 117,000 devices globally. Our solution is scalable and our target demographic is anyone with a smartphone or Chromebook, so in the next year we predict that our solution will serve between 1-10 million individuals, globally.

Our longer-term goals (~5 years) include having our solution as a native disease prevention tool on smartphones, like a weather app. We predict this could impact billions of people.

In each community where we deploy, our app measures not only the number of raw downloads, but also the robustness of the interaction network. We specifically track the total number of users within interconnected clusters of at least 100 users, where each connection corresponds to two users spending a significant amount of time near each other. In measuring installation, we seek to optimize the percentage of people in the community who are counted by this metric. This would validate that our incentive-aligned solution is indeed compelling.

We also measure the raw number of signals of positivity voluntarily entered by app users, so that we can track what fraction of positive users actually end up reporting their status.

Finally, unlike all contact tracing apps based on the Apple-Google infrastructure, we are able to keep track of the frequency of interactions between arbitrary users. For example, once we have a fully controlled pilot, we seek to observe the strength of correlation between a user’s relationship distance to the nearest positive signal, and the user’s frequency of interactions.

Full-time: 5

Part-Time: 10

Paid full-time interns: 3

Our team seamlessly bridges between academic research and commercial-grade product development. Founder Po-Shen Loh is a mathematics researcher who has been distinguished by a USA Presidential Early Career Award for Scientists and Engineers, who has been running  a technology social enterprise for over 7 years, which previously created a math and science website that sees over 500,000 unique visitors per month. The product team is centered around user experience design, supported by strong engineering talent, much of which is from Carnegie Mellon University.

Leveraging his academic background, Loh reached out and discussed the concept of NOVID with many researchers in public health during its development. One of them, Luca Ferretti from the Fraser group at Oxford, is providing direct epidemiological guidance for a potential European pilot, as well as directly collaborating on simulation research.

This team itself already has a track record of firsts. It was the first Bluetooth COVID app in the USA Apple and Google app stores, and the only COVID app to achieve sub-meter proximity accuracy (via ultrasound). It is the only app to use our new paradigm. It is the only app to practically support arbitrary Bluetooth communication between iPhones when apps are not on screen.