IBM explores AI tools to spot and cut bias in online ad targeting

IBM

IBM Corp is developing tools that would ensure online advertising algorithms do not unfairly show ads to only specific groups such as mostly men or wealthy people, aiming to address discrimination concerns that have drawn industrywide scrutiny.

The company announced on Thursday that a team of 14 will research “fairness” in ads over the next six months, exploring ways to spot and mitigate unintended bias, including in audiences and the messages themselves.

Academic researchers and civil rights groups have found for a decade that some audiences including Black people and women can be excluded from seeing job, housing and other ads because of potentially unlawful choices made by advertisers or automated systems they use.

Following complaints by U.S. anti-discrimination regulators and activists, Facebook Inc and Alphabet Inc’s Google, which are the world’s biggest digital ads sellers, enacted some changes.

But problems remain, while greater concern about data privacy has already begun reshaping internet marketing.

IBM’s Watson Advertising unit sells ads on its The Weather Company properties and also provides ad software to other businesses. Clients eventually could be pitched on anti-bias offerings.

IBM’s initial analysis of its own ad purchases suggests ads can be shown fairly equally to all groups without affecting measures such as the percentage of users who click an ad, said Robert Redmond, head of AI ad product design.

Redmond offered as a hypothetical example a system told by an advertiser to direct an ad to men across the United States. It may start showing the ad to only those who are in big cities and white because that group clicks more. IBM is researching whether artificial intelligence programs can spot that and then redirect the ads more equally.

The company next plans to analyze data from the nonprofit Ad Council’s public service announcements about COVID-19 vaccines.

Share: