IBM Watson Advertising and the 4As explore how AI can be used to combat bias in advertising – Ad Exchange

AI can help mitigate bias in advertising, but it’s still just the beginning.

On Wednesday, IBM Watson Advertising and the 4As released the results of a six-month research project that used AI to analyze data from the “It’s Up to You” COVID-19 vaccine education campaign. The Ad Council.

The dataset consisted of 10 million impressions and over 108 ad creative variations derived from predictive dynamic creative optimization.

IBM Watson Advertising used the data to develop an open-source toolkit called AI Fairness 360 as part of an incubation project with LF AI and Data Foundation, a nonprofit organization within the Linux Foundation that supports supports open source work related to AI, machine learning, and deepening. learning.

AI Fairness 360 uses 10 bias mitigation algorithms to measure ad campaigns and campaign decision-making against over 70 fairness metrics. For this application, the toolkit was used to apply post-processing bias mitigation so researchers can understand how to equalize results from a similar campaign in the future.

Rather than just looking at creative or audience, the toolkit aims to take into account the complexities of the digital advertising ecosystem.

For example, it looks for biases in the processes that drive an advertising campaign and assesses factors such as the advertiser’s development process for the campaign, the publisher’s methodology for delivering ads to its audience, and the processes of other parties and the machine learning workflows involved in the offering. chain.

In some cases, it may also be able to find unconscious biases programmed into AI-based solutions.

The Ad Council’s “It’s Up To You” COVID-19 campaign, for example, was geographically targeted to liberal and conservative leaning areas which were then divided into age groups. But the ad-targeting predictive model also took into account education level, gender and income, which is where the bias in the modeling became apparent over the course of the study.

It turns out that the dynamic creative prediction model more aggressively targets certain groups, including women, people between the ages of 45 and 65, and people with higher education and income. This likely led to lower conversions among other audience groups, such as those with low levels of education.

Looking back to understand where bias was present in a campaign is the first step towards using AI to create a more inclusive advertising ecosystem. The second step is to use AI to help overcome the biases that lurk in these internal processes.

In this study, the researchers sought to transform the predicted probabilities of the campaign’s machine learning predictive models based on observed imbalances in targeting criteria.

For groups identified as underserved by campaign targeting, the actual conversion rate was less than 0.01%. This represented a great imbalance between the number of people who did not convert and those who did.

In order to allow the bias mitigation algorithm to learn more about people from underserved groups who converted, the researchers reduced the number of non-converters to nearly equal the number of converters. . According to the study, adopting this type of common practice in cases with a large imbalance of data can help reduce bias between several disadvantaged groups.

This type of mitigation could be used to adjust a campaign’s targeting in future iterations.

“We’ve been playing around with mitigation strategies, and we’ve seen some mitigation opportunities and the propensity of that mitigation to potentially increase performance, or at least allow a wider audience to have access to the intent of the model to deliver the right message to them,” said Robert Redmond, Head of AI Advertising Product Design at IBM Watson Advertising. are present – and that we can manage them.”

The plan is to expand the analysis IBM has done with the Ad Council to enable more actionable examination of campaign data during execution, Redmond said.

To that end, IBM and the 4As are issuing a call to action for the ad tech industry to share data that can aid efforts to mitigate bias in advertising in the future.

“It’s about getting people to think differently about this campaign journey, about the people who are engaged and what they’re trying to deliver, not just in terms of DEI for their talent, but how they’re making business across the company,” said 4A President and CEO Marla Kaplowitz.

With actionable, AI-powered data at its disposal, the ad tech industry can be better able to deliver on its diversity, equity and inclusion commitments, Kaplowitz said.

“It has to go beyond the intention,” she said. “We have to start seeing the impact, and there has to be real accountability.”

About Deborah Wilson

Check Also

Advertising function | How the BMW M3 Competition Touring was born

Convince the bosses Despite a whirlwind of rumors, the BMW M3 Touring Concept remained resolutely …