New York, USA – Global technology company IBM, through its AI advertising solutions arm IBM Watson Advertising, is announcing the launch of a new research initiative in order to apply open-source AI technology, developed by IBM, to better understand how prevalent unwanted bias is in advertising, while simultaneously laying the foundation for potential mitigation.
Said work aligns with IBM Watson Advertising’s overarching mission to make AI the catalyst for improving solutions, services and trust in the advertising ecosystem, and builds upon IBM’s leadership in transforming business across industries with AI.
A part of the research initiative’s inspiration stems from a study conducted by the Geena Davis Institute on Gender in Media, which found male characters appeared in advertisements 12% more than females. This while women-led and gender-balanced videos yielded 30% more views than other videos, revealing a demand for more inclusive content.
For IBM Watson Advertising, the goal is to drive a deeper understanding of the factors leading to unwanted bias in advertising and how AI can help, spanning audience segmentation, creative messaging, performance optimization and campaign impact. Armed with this information, marketers and IT vendors may be able to develop a roadmap for the responsible use of AI to reduce unwanted bias and more effectively create and execute campaigns.
“We find ourselves at a moment in history where a long-overdue discourse on socioeconomic inequalities is dominating the national agenda and compelling action. Our hope is that AI can be the catalyst for reducing unwanted bias in advertising, just as it is helping to transform the advertising industry as it rebuilds for an era without third party cookies. Through this research we are taking an important first step toward that goal, applying scientific rigor to determine just how big of an impact AI can have,” said Bob Lord, senior vice president for Worldwide Ecosystems at IBM.
Areas of the research initiative include:
- Incidence of bias in advertising – The prevalence and frequency of bias in campaigns through the analysis of performance data. For instance, using the AI Fairness 360 toolkit, a suite of open-source AI tools developed by IBM and donated to the Linux Foundation, the study will look at how certain audiences of past and active campaigns are being targeted with creative content to assess whether bias was present.
- Role of signals in determining bias – How heavily signals, which refer to the context in which an advertisement is delivered, impact bias. As an example, if a creative message is deemed to be unbiased on its own yet is delivered on a digital channel alongside an inherently biased signal, the advertisement may be perceived as biased.
- Capabilities of AI to potentially mitigate bias – How useful AI can be in identifying instances of bias, and what can be done to fully capture the power of AI to potentially reduce occurrences of bias in advertisements.
As a start, data from the Ad Council’s “It’s Up to You” COVID-19 Vaccine Education initiative will be used for this initial research phase. The AI Fairness 360 toolkit will be used to examine the data and determine whether there are ways AI can be useful to help mitigate discrimination and bias.
“Collective bias has been prevalent in our industry for far too long, and the need to study its origins and impact is critical so that we can effectively work together to create progress. The Ad Council is proud to be the first industry partner of many working alongside IBM to help fuel this compelling research mission,” said Lisa Sherman, president and CEO at the Ad Council.