This text is part of the special section Génie québécois
By determining what content people are exposed to on social media, recommendation algorithms shape today’s information landscape to benefit big business, often to the detriment of the public interest. With an application used as a small research laboratory, a graduate from Polytechnique Montréal is trying to bring together different actors in society to develop “democratic” algorithms.
The trend is documented: most Internet users use their smartphones more than their computers to get information through platforms like YouTube and TikTok. “People will click and scroll more; these are the most common modes of navigation on the Internet,” observes Lê Nguyên Hoang, French mathematician and popularizer of science.
The content suggested on these platforms is chosen by recommendation algorithms, an information filtering system aimed at presenting elements of information that are likely to interest the user. The algorithms are designed by companies with the objective of optimizing user attention, allowing them to reap more profits from advertising, explains Lê Nguyên Hoang.
But this mode of operation generates various problems. The principle is to lock users into opinion bubbles, to more often suggest controversial content that diverts attention from real social issues or even to spread false news. These algorithms are also easily manipulated by companies creating fake accounts, which allow the number of views to be artificially inflated, thus creating artificial popularity which, by being shared by the algorithm, will become truly popular, explains Lê Nguyên Hoang. To illustrate the scale of the phenomenon, he gives the example of Meta, which has deleted billions of fake accounts since its creation.
“These algorithms determine the information regime of the population,” warns the mathematician, who underlines the threats of such a system for democracy. The problem is also at the heart of Lê Nguyên Hoang’s projects. After winning a prize for the best doctoral thesis in mathematics at his university in 2014 and being a finalist in the My thesis in 180 seconds competition, the Polytechnique Montréal graduate is focusing his research attention around the ethics of artificial intelligence . In 2021, he launched the collaborative platform Tournesol, which aims to produce beneficial content recommendations using more democratic algorithms. Over the years, Tournesol has emerged as a platform studied by scientists from all kinds of disciplines and which could potentially contribute to changing the paradigm on the Web.
An app to change the game
The Tournesol application works on the principle of video votes on YouTube. Users upload content that they consider to be of quality. They are then asked to vote, between two videos, for the one they prefer. The subjects are varied, ranging from the effects of pesticides to the decarbonization of France, including the action of Rohingya refugees denouncing the biases of the Meta algorithm or the operating principle of wind turbines. Users can vote on how much they like the video, but also on different aspects, such as its reliability, clarity, fun or its promotion of diversity and inclusion. Since its creation, the application has brought together tens of thousands of participants, but Tournesol is above all a testing platform which could make it possible to deploy a strategy on a much larger scale, hopes Lê Nguyên Hoang.
“We also seek to have an influence on academic thinking and we try to have an influence on the discussion on the regulation of artificial intelligence,” he says. “It raises lots of questions and challenges, from a philosophical or security perspective. Over the past year, more researchers have become involved in the project; sociologists are very interested in this type of initiative, we also have partnerships with philosophers and researchers from Polytechnique or the National Institute for Research and Security. »
The research questions are numerous, lists the researcher, who is also co-founder and CEO of the cybersecurity company Calicarpa. As in democratic elections, what would be the right ways to allocate voting rights on the Internet? Or how can the state of mind when voting influence the algorithm?
“Many reports show that democracies are in decline and I think it is urgent to protect our democracies,” says Lê Nguyên Hoang, who devotes much of his time to popularizing science. Since 2016, he has broadcast scientific content on his YouTube channel Science4All, which has over 240,000 subscribers. He is also the author and co-author of several books dealing in particular with issues surrounding artificial intelligence and algorithms.
“I would like there to be more mobilization of journalists, politicians, lawyers and researchers,” said the researcher. We need more citizen involvement to be able to clean up the discussion for better democratic representation. If we want to make algorithms democratic, we must import public institutions into the digital world, he believes. It’s going to take a lot of work and time to really build this digital democracy. »
This content was produced by the Special Publications team at Duty, relating to marketing. The writing of the Duty did not take part.