Mindshare's new AI tool measures how toxic news is to help brands decide where to advertise

GroupM's Mindshare has launched an analytics tool that aims to evaluate the social impact of editorial content on minority communities. Tyson Foods will be the first client to utilize the Impact Index, which was created in partnership with socialcontext.ai, a contextual intelligence startup supported by the University of Colorado Boulder. After a media planner inputs a client’s media and creative content, the Impact Index monitors, analyzes and categorizes the content into four levels of public impact: positive, neutral, negative and toxic. The ‘evolution of brand safety’ The idea of the Impact Index was born over a year ago, according to Rachel Lowenstein, global head of inclusive innovation at Mindshare, who came up with the idea alongside Jared Greene, director of invention+ at Mindshare. “The goal of this project is to create a human-safety tool,” Lowenstein said, noting that the algorithm “addresses the pervasive toxicity online and the role that advertisers might play in monetizing that rhetoric.” “The social and the media climate during the spring and summer of 2020 related to the deaths of George Floyd and Breonna Taylor, along with some of those negative depictions in the media of individuals in the Black community as a whole, really made us take a really critical eye in terms of the media that our brands were surrounding,” Greene added. The algorithm itself was built by a tech team of fewer than 10 members from socialcontext.ai, with the goal of creating something that is an “evolution of brand safety,” Lowenstein said. Chris Vargo, co-founder and CEO of socialcontext.ai and associate professor of advertising at the University of Colorado Boulder, said he has noticed other brand safety tools tend to “over-block content.” “Brand safety focuses on topics like does this article mention crime or not?” Vargo said. “Does it mention violence or not? And those are things that advertisers are sensitive about and they have the right to be…but there's actually a lot of really good news that accidentally gets swept up into this system, and we wanted to create a signal for conscious brands to intentionally say 'We're going to actually sponsor this stuff because it aligns with our brand.'” Lowenstein said one of the biggest challenges was finding the right partner to execute the idea. Vargo said Mindshare helped elevate its thinking on the topic to focus on what impact the article has versus simply what the article is saying. How it works To start, the Impact Index will focus on topics around Black Americans, but will eventually be expanded to include other marginalized groups such as the LGBTQ+ community, women, Latinos, and more. The software will also be expanded to account for intersectionality. To create the algorithm, Vargo said there had to be agreed-upon definitions of what makes editorial content-neutral, negative, positive or toxic. Vargo consulted not only with the professors that work on soxialcontext.ai but also a university advisory board made up of female academic leaders. “Toxic content has a clear aim to persuade the reader that an individual should be judged by their race,” Vargo said. "So [the system evaluates] content that says 'This Black person did this thing' and then the next step is [to determine whether] it invokes a stereotype or a general generalization or disinformation and then say that these two things are linked. Negative in this example would be coverage of Black-on-Black crime that was objective and factual. It's a negative article because it's portraying a Black person as being associated with bad behavior, but it's not toxic because it was a factual piece of reporting.” Once the definitions were solidified, the team had to then feed the algorithm thousands of articles over a few months so it could “learn what concepts are related to the thing we're interested in and which ones we are not so it can make a final prediction. And it can say 'I looked at this article and I think it's 99% likely that it's pro-racial diversity or an article that talks about African Americans in a positive light.'" The algorithm can also compare publishers rather than articles themselves. “There are certain sites that collect all of the Black crime that they can find and they run every single article and they make it a core part of their offering,” Vargo said. "That’s an over-representation of reality. One of the things that makes the Impact Index so awesome is we ingest thousands of publishers. So we know each week what the average outlet covered in terms of negative African American coverage. So if someone is three standard deviations away from the mean, they're doing something that everyone else isn't doing.” A growing need for marketers While the tool isn’t meant to tackle disinformation specifically, Lowenstein says she understands that disinformation is often used to create “toxic content.” Daniel Rogers, co-founder and executive director of the Global Disinformation Index, said he has seen an “increasingly concerted effort over the last few years by the industry to “stop advertising alongside disinformation, including content that is adversarial toward minorities.” “That said, this disinformation remains an enormous problem, as evidenced by GDI's current observations like in our recent report about ads appearing alongside anti-LGTBQ+ content during Pride month,” Rogers said. “It's great that GroupM is contributing to that industry-wide concerted effort, and GDI certainly supports the use of AI as part of that solution," noted Rogers. That said, "GDI also strongly believes that there needs to be a consistent and dedicated effort to eliminate the bias inherent in AI and that any solution must contain some element of human review. Human eyes on this content help eliminate false positives and ensure these efforts don't inadvertently block vital support to media that serves minority and at-risk communities." Weekly audits That human review will be a continued part of growing the Impact Index moving forward as it is audited weekly by members of the university to make sure the algorithm maintains at least a 90% precision rate. This process involves trained graduate students and professors reading and labeling news according to the specified definitions. The tool will eventually be used by other Mindshare clients as well, according to Lowenstein, and will be expanded to focus on more than just editorial content online. Making sure brands are part of the right conversations online is becoming a “growing need for all marketers,” according to Steve Boehler, founder of Mercer Island Group. “Media agencies need to take a proactive role in helping manage that challenge.” “Over time, at ID Comms we are seeing advertisers apply more scrutiny to where their dollars go and they are becoming more conscious about the impact of those dollars, beyond the impressions they buy,” said Tom Denford, CEO, ID Comms Group. “We expect to see more advertisers optimize in this area and will likely prefer to work with agencies equipped with these insights." The Impact Index is GroupM’s most recent initiative that aims to change the way advertising interacts with minority and underrepresented communities. Last May, the agency made a pledge to invest 2% of client budgets in diverse and Black-owned media, and launched a “diverse voice accelerator” to provide funds for diverse and Black content creators. The following month, the agency partnered with ad tech firm TripleLift and signed on to the “Underrepresented Voices” initiative, a stylized exchange traded deal to increase programmatic ad spending with minority publishers. This June, the agency struck a deal with influencer marketing company Influential to connect Black creators with brands as companies look to uphold promises to diversify media spending. At Cannes this year, Mindshare was one of parent WPP's agencies to partner with IBM to test out an AI toolkit dedicated to tackling bias in data. Another benefit of the Impact Index down the line has to do with data privacy. “I don't ever want ad tech to release [information about] audiences that could be exploited," Vargo said. “We don't know anything about the person reading the article. We don't know if they're Black or if they're a supporter, if they're sympathetic, we just know that they're reading this type of news. And I think that that's a better way to think about reaching audiences. We're not advertising to them based on assumptions. Instead, we're creating a tool that is supportive and inclusive and we're saying 'This is the kind of news we want to sponsor'…it's a nice way to avoid a lot of the issues that have come up in the cookie world.”

Mindshare's new AI tool measures how toxic news is to help brands decide where to advertise

Featured Stories