Algorithmic Audience Modeling and the Fate of African American Audiences Timothy J. Havens (bio) Algorithmic data processing has transformed commercial media audience research from behavioral measurement based on limited data to behavioral prediction based on information glut. Today's streaming media giants like Netflix and Amazon employ teams of engineers who design AI-enabled algorithms that scour this glut of user information, discerning behavioral patterns that they use to group subscribers into countless "taste clusters." Based on this clustering, the algorithms recommend content to subscribers while also guiding program production and acquisition practices.1 Ultimately, the engineers who design algorithms really aren't sure how they work, and, as scholars have shown, the algorithms aren't necessarily effective at predicting audience taste preferences.2 Regardless, algorithmically derived taste clusters shape a range of practices in the contemporary media industries. In this essay, I examine some of the implications of algorithmic audience modeling on the fundamental questions of race and media scholarship. The best guess is that algorithmic processing and prediction do not take into account demographic features like race and ethnicity.3 It is not that they misrecognize [End Page 158] or under sample nonwhite viewers, nor that they poorly predict these viewers' taste preference. Rather, demographic characteristics seemingly are not even represented in algorithmic processing. As such, recommendation algorithms are the perfect tool for an industry seeking to become post-racial. I say that race is "seemingly" not represented because, given the proprietary nature of these algorithms, very little is known about how they work. Numerous scholars have begun algorithm "audits," where they bombard a platform like Netflix with a range of requests and see how the recommendation algorithm responds.4 Others have examined the public statements of engineers and executives who work for streaming platforms for clues about how the algorithms work.5 However, to my knowledge, no one has audited streaming algorithms to examine whether they take into account demographic characteristics like race. The likelihood that demographic information is not used in algorithmic processing is reinforced by public statements made by Netflix executives that, as Evan Elkins shows, consistently claim that their taste cluster analysis does not include demographic information about subscribers.6 Audience measurement has been a long-standing issue among activists and scholars interested in the connections between media and racial justice, particularly African Americans, because of measurement's direct impact on what kinds of shows get produced and who gets hired to work on those shows as writers, producers, and actors. As early as 1977, the US Commission on Civil Rights' report Window Dressing on the Set: Women and Minorities in Television observed that the broadcasting industry's reliance on gross ratings points, or the percentage of Nielsen homes watching a particular channel or show, made it tough for non-mainstream (including nonwhite) content to get through the production development process.7 In the fallout from this report, Nielsen began over-representing Black households as a percentage of their panels in an effort to more fully represent their cultural tastes.8 As the 1980s and 1990s progressed and the networks began to lose viewers to cable, they started to focus more on 18-to 49-year-old white audiences. Since then, as Herman Gray argues, the networks have tended to think of African American viewers as political subjects capable of causing turmoil rather than as economic subjects worth targeting with relevant programming. This shift created a predictable programming cycle that recurred multiple times in the 1980s and 1990s as scholars and activists decried the absence of Blacks and Latinos in prime-time series. The networks responded by temporarily adding more diverse series and characters but inevitably dropped them due to poor performance among the 18-to 49-year-old white viewers. One such cycle occurred at the end of the last millennium, when Fox canceled a number of shows popular among African Americans, such as Living Single (1993–1998), Martin (1992–1997), and New York Undercover (1994–1999), as it shifted [End Page 159] its programming focus to young white men.9 for poor overall ratings, followed by political agitation on the part of African American and other minority-based political groups, which led...
Read full abstract