A year ago, IBM Watson Advertising launched a research initiative to understand and mitigate unintended bias in digital advertising using artificial intelligence. In January 2022 the results of that research revealed that, while unwanted bias in advertising is prevalent, AI can be used as an effective tool to alleviate it.
Now, in an effort to rally industry participation around the issue, IBM is joining forces with an industry coalition of leaders to pledge their commitment to generating awareness around the issue — and driving action.
At the Cannes Lions Festival of International Creativity on Monday (June 20), IBM announced that Delta Air Lines, WPP, The 4A’s, the IAB and The Ad Council have pledged their support to mitigate bias in advertising, recognising it as an industry-wide issue.
“I don’t think any of us should be precious about DEIB. We all need to share our learnings and make sure we’re driving change across the industry,” Marla Kaplowitz, CEO and president of the 4A’s, told Campaign US. “The pledge is a way to not only acknowledge that bias exists, but also that we are going to work as much as we can to unearth bias, talk to clients about it, and while we’ll never fully eradicate it, make sure we’re aware of biases and try to reduce them.”
IBM also released a free Advertising Toolkit for AI Fairness 360, an open-sourced product that includes 75 “fairness” metrics and 13 ready-made algorithms and sample code to help companies identify bias in digital ads.
WPP-owned media agency Mindshare has been using the toolkit in three areas: locating bias in existing audience strategies, mitigating bias in data science applications and validating that curtailing bias has an impact on business outcomes, said Brian DeCicco, U.S. chief data strategy and analytics officer at Mindshare.
“If we can execute campaigns with bias mitigation strategies in place, can we prove that fairness and performance can coexist?” he said. “Can we not only demonstrate [that eliminating bias] doesn’t negatively impact campaigns, but that it actually will also drive greater outcomes in the long term?”
One common practice at high risk for bias that Mindshare has been examining, for instance, is using historical data to create audience segments and predict future outcomes.
“If there is inherent bias in those datasets — if they skew toward one population vs. another — those biases can be perpetuated in modeling, and then you have humans injecting their own biases into the data science process,” DeCicco said.
IBM’s initial research into bias, released in January, was based on a test on The Ad Council’s “It’s Up To You” vaccine education awareness campaign that analyzed 10 million impressions over 108 creative variations to uncover biases. According to a white paper IBM released on the study, the AI found biases in the campaign against people with lower levels of education.
For now, IBM’s research and algorithms focus on age and gender bias as proxy characteristics. But as more industry players join the cause, it hopes to expand its research and tools to all forms of bias.