The Case For Letting Algorithms Run The Vaccine Rollouts
Belgium's vaccination campaign is a prime example, computer scientist Hugues Bersini argues, of how technology can not only improve efficiency, but also, in some cases, make things more fair.
Even the most technophobic of our fellow citizens could acknowledge the three essential virtues of algorithms:
First, there is no better way to manipulate numbers and optimize quantities.
Second, their time-saving advantage is undeniable (take, for example, Waze, Google, banking applications, etc.).
And third, they are rigid and difficult to hijack, making fraud much harder (planes are always hijacked with a Kalashnikov — not by writing lines of code).
Since the COVID crisis began, the countries that have made the best use of software devices for tracing contacts, organizing quarantines and testing are among those which register the lowest number of casualties — specifically the Far East countries, including the least liberticidal among them.
It's easy to understand why, considering the first two virtues listed just above. This crisis is full of numbers to be minimized (the reproduction rate, the number of serious cases and contacts) or to be maximized (social distance, the number of tests...) and a race against the virus and its variants is still underway. As the French President Emmanuel Macron said, virus is the master of time.
A good algorithm would allowed last doses to be given to the people who needed them the most.
I have already criticized the use that has been made of these software devices and the extraordinary flaws revealed, during the crisis, when public services exploited these same devices. There was the failure of the Bluetooth contact tracing app and the initial difficulties to organize the tests. Later, and despite the fact we've been in this crisis for a year, there were the unacceptable hiccups when people booked vaccination appointments, toward the start of the campaign.
We saw young people, in good health and teleworking, lifting their shirt sleeves to get the vaccine, instead of the at-risk elderly people who need it most. And what a peculiar idea Belgium had, unlike many European countries, to send vaccination invitations without first allowing people to state whether they wanted to be vaccinated or not ... The result? An unbearable number of unanswered invitations and, today, having to compensate with a flawed software called QVAX.
The reality, though, is that the vaccination campaign actually constitutes the perfect example of a crisis situation where algorithmic assistance is vital. For a successful rollout, it is necessary to make calculations by combining at least three different quantities: the available vaccine doses, which vary all the time; the cohorts of patients who need to be vaccinated; and the capacity of vaccination centers.
We need to vaccinate as many people as quickly as possible while also preventing fraud and preferential treatments — something that is inevitable as long as people are in charge and that, in some countries, has brought down ministers and other influential figures.
In Belgium, some of these priority reversals were justified by the so-called last doses that need to be administered promptly. But this is not a good reason because, again, using a good algorithm would have allowed these last doses to be given to the people who needed them the most, while respecting the priority order, instead of favoring friends.
A perfect match
Computer scientists are familiar with the Gale-Shapley algorithms: They are used, for example, in organ transplants or in France's Parcoursup, an automated system for university enrollments, and what they're able to achieve is a perfect association between several essential elements — in this case, patients and vaccines, even though that latter is subject to shortages.
The algorithm, or at least the logic behind it, should be known to all.
These algorithms start by sequencing the first elements in relation to the second. This way, each patient can arrange his or her potential vaccines and vaccination centers in order of preference. At the same time, each available vaccine can do the same for the patients (with the centers able to sort out the patients by geographical proximity).
The algorithm achieves the perfect association, for example, matching the high-priority patients with the adequate vaccines and directing these patients to easily accessible locations. Also, any type of fraud becomes almost impossible with the algorithm, as your phone informs you, and no one else, about the and time and place of the vaccination appointment.
The algorithm, or at least the logic behind it, should be known to all; better still, it should be decided by all, in particular when it comes to the priority levels we grant patients.
Common sense has already made it possible to give priority to the elderly and/or people suffering from comorbidities. But when it is time to decide which job is more essential than another in the eyes of the virus, that's a whole different story!
We fear a rough and tumble kind of situation. This is where the citizens' assembly that President Macron wanted to create in France could have really made all the difference. Once these citizens reached an agreement, priorities would have been set in algorithmic stone, clearly and transparently. And nothing and no one could have made it diverge from that.
Finally, we shouldn't let privacy concerns stand in the way of this process: A lot was said about that in the newspapers, especially for people with comorbidities who suddenly became more than reluctant to expose their obesity or hypertension to the public. It is easy to see the absurdity of it when lives are at stake.
*Hugues Bersini is a computer science professor at the Université Libre de Bruxelles in Belgium.
**This article was translated with permission from the author.