Study the impact of AI on the design of transparent and fair marketplaces
FAIRPLAY, the joint Criteo - Inria - Institut Polytechnique de Paris project-team, aims to study the impact of AI on the design of transparent and fair marketplaces. Led by Vianney Perchet, senior researcher at Criteo, researcher at CREST and professor at ENSAE Paris, and Patrick Loiseau, Inria researcher, the FAIRPLAY project-team builds a link between industry and higher education by introducing ENSAE Paris students to mathematical-economic research motivated by the concrete topics of online advertising specialist Criteo.
Less and less data, persistent discrimination
How to provide relevant and personalized information to Internet users, while having less and less access to their personal data? How to ensure that an advertisement or a job offer will be seen in a non-discriminatory way by the people to whom it is pushed? These questions are particularly topical, but they are also important issues for the advertising marketplace organizer Criteo, which relies entirely on machine learning.
"Data are less and less provided, and they are less and less accurate. At the same time, when data is exploited, algorithms naturally reinforce biases, even without having access to sensitive data from Internet users," explains Vianney Perchet, a researcher at Criteo, as well as at CREST, a professor at ENSAE Paris and scientific co-leader of the Inria FAIRPLAY project team. "The question today is how to measure the extent to which algorithms are discriminating, while respecting data protection," he adds.
But then how to explain this discrimination by machine learning systems, even though sensitive user data is protected? There are several reasons for this problem. First, not having access to private data, such as the gender of the Internet user, does not guarantee that there will be no discrimination. "There is a lot of non-sensitive data, which is highly correlated with data considered sensitive, such as the sites consulted by Internet users, which can give indications of the gender of the person consulting these sites," explains Patrick Loiseau, an Inria researcher and scientific co-leader of the project team. Another problem: the multi-agent aspects, which make things more difficult. "Typically, in an auction system, decisions are made asynchronously and decentralized. The fault does not necessarily lie with the advertiser, because the final decision to print an ad is the result of a chain of intermediate decisions made by different agents," he says.
A project team to work on the design of transparent and fair marketplaces
To address this issue, Inria, Criteo and CREST (a joint CNRS-ENSAE Paris-ENSAI-École Polytechnique research unit) have decided to work hand in hand through the creation of a joint project team. Named FAIRPLAY, this project-team is composed of five academic researchers (Inria, ENSAE Paris and École polytechnique) and four researchers and engineers from Criteo.
Its credo: to study learning problems in multi-agent systems. "In advertising, there are typically many agents learning at the same time, and this unconsciously causes discrimination, particularly with regard to the "opportunities" (job offers, financial offers, etc.) offered to Internet users," says Patrick Loiseau. It has been proven, for example, that women receive job offers that are on average less well paid than men.
The objective behind the project team's work is to improve the automatic market systems, but also to be able to know the degree of discrimination of certain algorithms, all while remaining compatible with the notions of privacy of the RGPD.
"The idea is to be proactive in the creation of these algorithms and their implementation, to force them to respect global fairness constraints," says Vianney Perchet. "We approach this with the help of game theory, which is a tool for modeling multi-agent systems that makes it possible to evaluate the quality of systems and find solutions," adds Patrick Loiseau.
As with a traditional project team, FAIRPLAY has been created for a renewable four-year period. It will be housed at both CREST and Criteo.