A recent platform developed by computer researchers tries to collect as much data as possible in order to identify human preferences in terms of content. Its final objective: to fight against disinformation and misinformation on the Web. Futura spoke with the president of the Tournesol association, Lê Nguyên Hoang, computer researcher and scientific popularizer at the Federal Polytechnic School of Lausanne.
Have you ever thought in retrospect on the quality of the content you consume on the Internet and more particularly on video platforms? What do you prefer to watch when you want inform you on a subject ? How do you rate the information ? Swiss academics have developed an algorithm whose ultimate goal is to improve recommendations by matter information. They are currently working on these issues and need you. For its readers, Futura interviewed the president of the association Sunflower, Lê Nguyên Hoang, computer scientist and scientific popularizer at the Federal Polytechnic School of Lausanne.
Futura: When and how was the Tournesol project born and what role did the informational context play in this project?
Le Nguyen Hoang: The issue of information algorithms, in general, and content recommendation algorithms are questions that have been running through my head for almost a decade, when I started as a videographer on YouTube. For about five years, I have familiarized myself with academic research in this area and have reflected on it further with colleagues in my laboratory. Many events came to fuel our reflection and strengthen our conviction that there was a problem to be solved, such as the Cambridge Analytica scandal, and led us to publish the book The Fabulous Building Site (edp sciences) and 2019.
The algorithms of social networks and platforms (…) have a key role to play in the fight for the dissemination of information of public utility on a large scale
But the click was really there pandemic from Covid-19. Never so much misinformation and mass misinformation had not circulated on a given subject. From then on, our conviction that popularization of science was of crucial importance grew stronger. It was necessary that the essential information could be transmitted without problem and not be diluted in an ocean of false information or less newsworthy information at the time.
Since the genesis of the project, we believe that the algorithms of social networks and the platforms that distribute this content have a key role to play in the fight for diffusion large-scale public utility information.
At the start of the pandemic, I contacted a friend who worked at Google and I told him that it was absolutely necessary to modify the algorithms to perfect the content recommendations. He told me he agreed with me but he asked me this question: ” Which videos should be featured in your opinion? “ It was then that I realized that this question which had seemed so simple to me for years was in fact of an abysmal complexity.
Futura: How does Calculus work?
Le Nguyen Hoang: Before talking about how Tournesol works, we must briefly recall its current objective. The platform is currently a kind of microscope on preferences and how humans make judgments. The current goal is simply to collect as much data as possible on what content people deem best to recommend overwhelmingly. For the operation of Tournesol, my last video just talks about this in detail. Very briefly, it allows you to compare two videos with each other according to several criteria that we deem relevant to consider content of public utility or not. We made this choice because theories in psychology suggest that our judgments are more reliable and have more informational value when we make comparisons than when we evaluate something individually. From the reviews of contributors, the algorithm generates a score that transcribes the usefulness of the video based on user preferences.
Futura: We tested Tournesol. Is there not a problem with the fact that all opinions have the same weight, for example on the criterion of reliability?
Le Nguyen Hoang: Like I said earlier, Tournesol’s current goal is really to understand how people rate content. Beside that, we invite our contributors to fill in personal information. For example, their age or level of education. When we have built a database solid which can already be downloaded from the home page of Sunflower, we hope that other researchers will look into the analysis of these data and that we can, collectively, design more precise models in order to overcome the various biases such as that of subjectivity and equal weight of the rating.
Futura: When will we be able to see Tournesol “rampant” on a large scale?
Le Nguyen Hoang: The first version of Tournesol is already active. On Twitter, for example, you can already challenge the Sunflower bot to ask him his opinion on a video. We are also going to develop an extension compatible with Firefox and Chrome so that information about a given video appears every time a user consumes content.
In order for Tournesol to produce reliable recommendations at scale, what we really need are contributors
It is a way of reminding people that paying attention to the reliability of information is very important. Getting back to the question, in order for Tournesol to produce reliable recommendations at scale, what we really need are contributors.
The problem right now is that we have a strict account validation policy to protect Tournesol against what is commonly referred to as the 51% attack, which reflects the influence that fake accounts can have that are currently in the spotlight. standard on the internet. Therefore, we only accept trusted email domains. We are considering alternatives to solve this problem.
However, I invite everyone reading the article, if they wish, to take a tour of our site and rate the videos. Each contribution allows us to move a little further towards a world where information recommendations will, I hope, be more robustly beneficial.
Interested in what you just read?
