Scientists led by AI scholar Yoshua Bengio urge privacy, opt-in approach for COVID-19 contact tracing

Testing appears to be still urgently needed for containing and ultimately eradicating COVID-19. Canadian AI scholar Yoshua Bengio and colleagues from multiple institutions urge privacy measures for contact tracing. ZDNet posed questions to the scholars about their recent editorial.
Written by Tiernan Ray, Senior Contributing Writer

Scientific reports continue to emphasize the importance of testing the population for COVID-19on a regular basis. But surveillance in broad terms raises numerous concerns about privacy. 

Those issues are front and center with contact tracing, the practice of trying to document the interactions between people infected with COVID-19, and those with whom they've been in touch. For anyone to scrutinize how another person moves through life, including their social interactions, strikes many as an invasion of privacy, especially when that scrutiny comes from a central authority. 

On Wednesday, a group of scientists led by artificial intelligence pioneer Yoshua Bengio of Montreal's MILA institute for AI, penned an opinion piece in the prominent medical journal The Lanceturging that privacy be given top priority in any contact tracing. 

The form of contact tracing that can be done through smartphones, which is being pursued by both Apple and Google, is "a powerful yet controversial strategy to combat the COVID-19 pandemic," Bengio and colleagues observe. 

"Most of the applications in use or under consideration have an impact on individual privacy that democratic societies would normally consider to be unacceptably high," they said. The authors cite Liza Lin and Timothy W Martin of The Wall Street Journal, who wrote on April 15 about a new age of surveillance that is eroding privacy.


The essay by Yoshua Bengio and colleagues in The Lancet of June 3rd, 2020.

Bengio and the team have three main points in their Lancet essay. One is that any information about contacts and interactions should remain stored on a person's phone and not be put in any central repository. A second main point is that some kind of fiduciary body should be set up to pursue transparency of the process, such as source-code disclosure for apps. And the third is that the process should be opt-in, rather than forcing or coercing people to participate. 

On each of these points, ZDNet emailed questions to Bengio and to corresponding author Dr. Abhinav Sharma, who is an assistant professor in the Department of Medicine, Divisions of Cardiology and Experimental Medicine at McGill University in Montreal. 

Dr. Sharma replied in writing in an email to the questions. 

On the first point, keeping contact records only on the phone, ZDNet asked about the proposal in the paper that "the tracing application itself can propagate alerts to high-risk contacts." How can this capability be reconciled with the need to keep information private and restricted to the phone, asked ZDNet?

Sharma replied that the authors "advocate for a more decentralized form of risk awareness whereby the details are stored on the phone and not transferred to central authorities," but that could nevertheless carry out notifications. 

Sharma pointed to a publication authored by Sharma and Bengio and additional colleagues, and posted last month on the arXiv pre-print server, in which they endorsed a form of contact tracing application called COVI.

COVI is a "peer-to-peer contact tracing and risk awareness mobile application" that the authors have developed that is intended to both give "personalized recommendations" to the user but also to "mitigate potential ethical and privacy risks," as they put it. 

As Sharma described the proposal to ZDNet:

Upon self-entering the results of a covid-19 test or new symptoms, this risk can be propagated to close contacts through the use of various strategies (such as mixed-nets) without central authorities learning about the details of the contacts. We can also envision a strategy whereby a verified covid-19 test result (eg. from a health laboratory) pushes out a result to an individual but the propagation of this results to contacts of the individuals happens in a decentralized manner to ensure that no private or public agencies learns of the contact history of any user.

At the same time, Bengio and the authors of COVI suggest anonymized, aggregate data gathered from the app can be used by epidemiologists and others to work on broader issues regarding the spread of the disease. They offer metrics by which they argue that various machine learning approaches, applied to data, can help flatten the curve of the virus, perhaps more than is achieved via other non-pharmaceutical measures, such as distancing.

On the second point in the Lancet essay, a trusted oversight board or fiduciary, the authors recommend that such a non-partisan body could ensure that a number of important protocols are followed, including that "the source code for the application and the privacy protocols used should be publicly available."

ZDNet asked Sharma how the authors expect such a body to be established. 


Bengio and colleagues make the case that their peer-to-peer digital contact tracing app can flatten the curve even more than some other non-pharmaceutical interventions. 

Alsdurf et al. 2020

Sharma replied that the formation can "take various forms depending on the needs of the strategy being used and the country," and can "include a mixture of elected officials, appointed officials, medical representatives, ethicists, and community representatives," and others. 

In the case of the COVI app that Bengio and the team have developed, Sharma said that a non-profit has been established called COVI Canada. It is lead by former Canadian supreme court judge Louis Arbour.

On the third point, about the opt-in nature of things, there is a bit of a tension described by the authors between making compliance voluntary on the one hand, but assuring sufficient participation on the other hand. Without insisting on the approach, the authors are inclined to urge voluntary consent rather than enforced, coerced participation. 

Also: Coronavirus app: What contact tracing is, and how it will work

As they write in the Lancet essay, "We would suggest that advocating an approach that emphasizes consent and prevents any central public or private authority from accessing identifiable data would embolden more individuals to download the application, thereby optimizing the population-level benefit."

To address this third point, ZDNet asked Sharma how the authors plan to pursue having their suggestions implemented. Sharma replied that the group is engaged with "governmental agencies at the local, provincial, and federal levels."

In addition, wrote Sharma, "we are in close communication with both large and small technology companies who are working towards developing mobile solutions," which, Sharma wrote, "is critical to ensure that privacy is kept at the forefront when developing a strategy to curb the COVID-19 pandemic." 

Do you have strong feelings about the use of contact tracing? Feel free to share your perspectives in the comments section. 

Editorial standards