8 million people, 14 alerts: why some covid-19 apps are staying silent

When France launched its app for digital contact tracing, it looked like a possible breakthrough for the virus-ravaged country. After going live in June, StopCovid was downloaded by 2 million people in a short time, and digital affairs minister Cedric O said that from the first downloads, the app helps avoid contamination, illness, and so deaths." But officials soon had to walk their enthusiasm back after it emerged that in its first three weeks the app had alerted only 14 people to tell them they might have been exposed to the coronavirus.

This isn't the end of the story," said O in its defense. We keep improving the application."

In Australia, meanwhile, things were even worse. The country's Covidsafe app launched in April and got far greater adoption-6 million downloads in a country of 25 million. And yet it had even less of an impact: in the state of Victoria, it failed to identify a single contact that hadn't already been uncovered by manual tracers, according to Gizmodo.

But while it doesn't sound great, the lack of pings may not necessarily be a sign of failure in and of itself.

Part of the criticism may be due to too much hype. The early focus on contact tracing apps was understandable: a vaccine is still many months away, assuming we can even find one that will work. Apps stepped into the breach as a potential panacea-even though many insiders have consistently argued that they are only one of a number of tools we have to fight the virus.

On a mathematical level, too, the low level of notifications might be expected, according to Jon Crowcroft, professor of communications systems at the University of Cambridge. In a situation where there are low numbers of covid-19 cases, people are observing social distancing, and the density of app users is not high, you would not expect to see many notifications, he says.

It's simple math for the numbers of notifications: if 1% of people have covid-19 and they are all tested, and only 1% of people run the app, you have a 1 in 10,000 chance of having both the tested person and the exposed person having the app, so your notification rate will be 10,000 times lower than the case rate," Crowcroft explains. (For example, during the period in which Victoria issued 21 notifications, the state registered just 350 cases of covid-19.)

However, even with the most optimistic lens, it's clear there's a gulf between what was promised and what these apps are delivering. So what went wrong?

Technically awkward

First, it's worth looking at the similarities between the two services. Both France and Australia shunned the model put forward by Google and Apple-where data is kept on the user's phone to maintain privacy-in favor of a centralized approach, where user information is sent to remote servers. This is problematic because Google and Apple have restricted how much Bluetooth scanning centralized apps can do in the background.

Michael Veale, a digital policy lecturer at University College London, sums up the issue: They aren't detecting many phones because the background Bluetooth does not function. That's because they aren't using a decentralized approach."

This situation has created a series of other technical difficulties. Australia's app works only 25% of the time on some devices, in particular iPhones. That's because the Bluetooth handshake" necessary to register proximity between two phones doesn't work if the phone screen is locked. This was the exact problem that caused the UK to abandon its app last month (it's not clear when it will launch a replacement).

This effectively means for a contact tracing app to work without using their system, a user has to walk around like a Pokemon Go player, with their phone out, the app open, and not use their phone for anything else," says one researcher not directly involved in development for either app, who requested anonymity.

Too conservative

All this may have been exacerbated by adopting an overly conservative approach to avoid the risk of over-notifying" users, says Crowcroft. Worries that oversensitive alerts could create panic means the apps only consider people who are extremely likely to have been in close contact with each other for extended periods of time-not just people you brushed past for a few seconds in the store. A lot of care went into trying to avoid a lot of false positive notifications in some apps. This may make them super conservative," he says.

In addition, both Australia's and France's apps have been blighted with performance issues and bugs. Users have complained that France's app drains their phone's battery life-possibly the reason that hundreds of thousands of people have uninstalled it.

This is the prime risk for developers: you make one mistake and wipe out somebody's battery," says Andrew Eland, who until recently worked as an engineering director at Google and then DeepMind Health. Some users say the StopCovid app regularly crashes, and has to be reactivated every time you switch your phone back on.

Aiming for improvement

So what are the lessons? Bluetooth is a very complex technology, but it's fiendishly difficult to build a contact tracing app without using Apple and Google's system. So for the sake of building an app rapidly, perhaps it's best that governments don't adopt a centralized system or something else that creates technical difficulties. If possible, countries should consider reusing the code for another country's app that has proved to be a success-for example, Germany's open-source Corona-Warn App, which has been downloaded by over 15 million people in a population of 83 million since it launched on June 15. Secrecy and clinging to exceptionalism are a poor combination when it comes to building contact tracing apps.

And ultimately, the public needs to bear in mind that contact tracing apps are likely to be only a small part of the fight against the coronavirus-rather than a magic answer to the problem.

If you want to know the best way to spend time and money on technology to track and trace coronavirus infections, it would probably be better to focus on making manual contact tracing more efficient," says Eland.