The duty reported on Wednesday that, according to several sources, the federal government will not support the COVI application offered by the Quebec Institute for Artificial Intelligence (Mila), preferring another solution. The Press for his part reported that, according to the president and chief executive officer of the research center, Valérie Pisano, Mila will not “go ahead” with her project since any application for smart phone distributed on Apple platforms and Google in connection with COVID-19 must first obtain the approval of the government.
The COVI project has raised controversy because it requires more data to run than other applications and, like The duty revealed it last week, Mila planned to entrust them to an opaque structure, a non-profit organization. As the Institute has close ties to private companies like Facebook and Google, several voices have been raised against the project.
Other proposed contact tracking apps would not need that much data to work. How was that of Mila different?
In data | Our interactive content on COVID-19
Mila’s COVI application intended to use the same contact search procedure as most of the other protocols offered. By using the Bluetooth wireless feature of its users’ phones, these protocols allow devices to identify each other, while maintaining user anonymity. By default, this data is kept on the device. When a user is confirmed to be diagnosed with the virus, his device sends a message to all users he has encountered previously, when he was contagious. They can then isolate themselves and be screened.
But, according to the article published by the Institute when its project was announced, Mila’s application intended to go much further: it had to calculate the risk posed by COVID-19 for each user.
To do this, an algorithm would have taken into account, among other things, the basic profile of users (for example, if they suffer from heart problems or asthma), their movements and the level of risk associated with the users with whom they would have been in contact. Each time this data is updated, such as when contacting a person at high risk of being infected, the user’s risk level would have been recalculated by the algorithm on their device.
But to feed and develop such an algorithm, you need data. If artificial intelligence can look like a mysterious black box, its principle remains quite simple: we ask a computer to find links and correlations between different data in order to draw inferences. For example, if an algorithm is given enough photos of cats, it will be able, by dint of noticing similarities between these photos, to recognize a cat photo among other images. Users of the COVI application who have given their consent could have anonymously transmitted their data to external servers in order to train Mila’s algorithm to better assess the risk level of each user.
Mila had bet that this approach would have been the government’s favorite. According to the Institute, this gave greater responsibility to its users, because it enabled them to make informed choices thanks to advice provided by the application according to their level of risk.
Many of you have submitted questions for this newsletter to email@example.com. Thank you very much, and most importantly, keep doing it.