Meta to reduce ad discrimination through AI technology named Variance Reduction System in collaboration with DOJ

Meta has implemented some changes to ensure that the actual audience for an ad closely matches the target audience to address concerns that its ads are discriminatory. Meta has promised to reduce ad discrimination through AI technology in the US and is designed to reduce bias and increase fair ad distribution.

Meta has rolled out a new ad distribution system as part of the company's settlement with the Department of Justice called Variance Reduction System (VRS). This system uses new machine learning technology in ad delivery for the actual audience to see more relatable ads. VRS will conduct measures to combine demographic distribution to people who have seen the ad once the ad has been shown to several users. The Variance Reduction System ensures to reduce bias and increases fair ad distribution. Meta is placing more restrictions on ads targeting teens on Facebook and Instagram which is the second major ad policy change.

Meta allows advertisers to target people based on interests, demographics, and behavior, through an ad auction, it was then decided what ad to show a user at a particular time. The company received complaints about advertisers using ad tools to exclude people from certain services, including housing, employment opportunities, and more, although it had rules in place against discriminatory ads. The tech giant, Meta back in 2019 faced charges for letting advertisers violate the fair Housing Act using the Special Ad Audiences feature. Later, Meta removed this feature and promised to change its algorithms to combat the problem. The company said it will be extending the use of VRS to US employment and credit ads in the coming year. "Additionally, we discontinued the use of Special Ad Audiences, an additional commitment in the settlement," said Roy L. Austin Jr, Vice president of civil rights and deputy general counsel at Meta, in a blog. It was replaced with a new algorithm called Variance Reduction System. The invention "will impact the way housing ads are distributed to people resident in the U.S. across different demographic groupings," according to spokeswoman Ashley Settle. An investigation by ProPublica in 2016, revealed that advertisers were able to create Facebook ads that excluded people based on ethnic affinities. VRS came into effect in the United States and this technology sends a strong message against discriminatory advertising. The system allows advertisers to reach the target audience while avoiding unintentional discrimination against other groups. VRS is Meta's remedy for ad-based discrimination. The AI technology runs as soon as an ad goes live and comes into action once the ad has reached a certain number of users. Meta aims to ensure that users who see various ads closely reflect the targeted audience with VRS. The company does this by achieving by measuring the audience for a particular as and then comparing it with the demographic distribution like age, gender, race, etc. To determine the estimated race or ethnicity distribution, VRS relies on a Bayesian Improved Surname Geocoding (BISG) method with added privacy enhancements. BISG is a method developed by RAND cooperation that uses geocoded addresses and surnames, refining census data to estimate race and ethnicity. The new system is based on aggregate or statistical data. "This method is built with added privacy enhancements including differential privacy, a technique that can help protect against re-identification of individuals within aggregated datasets," Meta said. The VRS system will use the latest aggregate demographic measurements to distribute the ad to an audience that more closely reflects the eligible target audience every next time the ad is displayed. As more people view the ad, the VRS will remeasure and update accordingly, adjusting the pacing of ads to help ensure an equitable distribution of ads. "Before the VRS ever makes adjustments in our ad system, it is trained through a form of machine learning called reinforcement learning; here the VRS learns how to use pacing effectively to reduce demographic differences," Meta said. Once the VRS system has data, it will adjust the auction value of the ad to display it more or less to a specific group. To test the new system, Meta has started with housing ads. In the next year, the company expands to include job and credit ads. This will not jeopardize privacy since the system is based on aggregate data. According to Forrester Principal Analyst, Brandon Purcell, the VRS and the elimination of Special Ad Audiences helped Meta rebuild trust with users, and advertisers, but there are caveats. To limit harmful ad campaigns, Mets is placing more restrictions on advertisers' ability to target teens. Meta is expanding ad preferences on Facebook and Instagram to let teens see fewer ads. They could already hide the ads from specific advertisers, but this gives them the choice of automatically downplaying whole categories. In 2021, Facebook and Instagram barred advertisers from using teens' interests to target ads.