technology

Online advertising: Facebook will use machine learning to prevent discrimination


More equity, less discrimination. Facebook sets off in the United States a technology aimed at reducing the discriminatory biases of the advertisements broadcast on its platforms. Called Variance Reduction System (VRS, for system of reduction of the variations), it was developed under the impulse of the ministry of Justice after the signature of an agreement. The platform was accused of discriminatory practices by offering the possibility of targeting certain advertisements with disconcerting precision, in particular for housing.

More fairness thanks to algorithms?

Across the Atlantic, Meta’s VRS applies for the time being to the latter. Advertisements related to job offers and credit will come next. To serve ads more fairly, the tool is based on a new knowledge machine learning technique.

In detail, the advertisement is first displayed to a test group of users. The profiles that have viewed the ad are then analyzed and compared “to the demographic distribution of the eligible target audience selected by the advertiser”. If any shortcomings are found by the program, the ad targeting algorithm helps to “reduce the difference between audiences” The VRS is thus based on the “United States Census public statistics to measure estimated ethnicity”, says Facebook. The system will operate throughout the advertising campaign broadcast period.

A class action in Canada

The US Department of Justice welcomed this advance. “This development marks a crucial step in the Department of Justice’s efforts to hold Meta accountable for unlawful algorithmic bias and the serving of discriminatory ads on its platforms.”reacted Kristen Clarke, assistant attorney general of the civil rights division relayed by the wall street journal.

In parallel with the development of the VRS, Facebook recalls that it is now impossible for advertisers in Canada, the United States and certain European countries to target advertisements for housing, employment or credit according to the age, sex or postal code. “We know that we cannot wait for a consensus to move forward in addressing important concerns about potential discrimination”explains the social platform.

The announcement comes a few days after the validation by the Quebec Court of Appeal of a class action against Facebook for advertising discrimination. “Algorithmic discrimination, which excludes people like women and older workers from receiving job advertisements, is just a modern form [et] illegal under the Quebec Charter”, then denounced Audrey Boctor, lawyer at the origin of the complaint. Facebook had not commented on the substance of the case, but assured “to promote non-discrimination and equity in [ses] advertising systems”.

Related Articles

Back to top button