This post was written by Prem Sylvester, a graduate student at the Digital Democracies Institute working on the Beyond Verification research stream. All images are courtesy of Charlton McIlwain.
Charlton McIlwains’ rich and deeply important presentation as part of the Fall Speaker Series 2022 raised a crucial but underappreciated question: how do we think about racial targeting as a particular form of harm in the digital economy? Through critical reflections on how racial targeting has historically proceeded in newspaper advertising, to how we might extend some — but not all — of those lessons to understanding digital advertising, McIlwain demonstrated how advertising and predation often go hand in hand. Simultaneously, his blend of theoretically-informed qualitative and quantitative methods help clarify the stakes of racially targeted digital advertising.
Drawing on his experiences as an expert witness, McIlwain explained that targeted advertising has been used to sell predatory mortgage products to Black and Latinx persons. The extent of such targeting could be interrogated through the advertisers’ choice of displaying certain content in certain geographical locations, alongside examining their choice of newspapers, duration of advertising, and cost of ad purchase. However, these cases raised an important complexity to the question of targeted advertising. Unlike older forms of racial segregation such as employment discrimination and reverse redlining, racial targeting seemed to be inclusionary of racial populations who were prewvioulsy denied access to credit and financial services. Racial targeting, then, is standard practice in advertising. However, it becomes a matter of concern when the product being sold is harmful and the advertising tactics are predatory.
The questions that McIlwain asks (and that are useful for other scholars engaged in similar work to ask), then, revolve around what might be the adequate frameworks for “framing, identifying and addressing the impacts and harms produced by race-targeted advertising”, and if the legal norms of fairness and/or disparate impact remain appropriate. These questions become especially pressing in what McIlwain notes as “the complex, automated, and algorithmically driven digital advertising ecosystem.” In particular, the distortion of impact in the digital environment may necessitate a revaluation of the importance of intentionality in the production of racially-targeted harms.
The historical juncture that introduced these complexities — and therefore sets the terms for challenging the harms of racial targeting — can be located in the development of audience research that came out of the proliferation of advertising and marketing through radio and other communication technologies beginning with the 1920s. By measuring and segmenting audiences as commodities, advertisers could determine how to communicate messages about their products to their intended audience. The computational turn in advertising, or “when computation meets the audience” per McIlwain, began gathering steam in the 1950s and 1960s. Ithiel de Sola Pool, who was professor of political science at MIT, sought to simulate and predict human behaviour with the data and computational tools that the university made accessible, with the view to ultimately manipulating that behaviour. Such prediction and manipulation was based not on who people were as individuals, but on individuals as they were connected to particular groups with which they powerfully identified. Historically, then, McIlwain notes that “interlocking racial, computational, and spatial logics fuel the advertising and marketing industry’s development,” which in turn “allow racial discrimination writ large to persist, particularly in the forms of segregation.”
Payday lending, which often functions as a known predatory practice along axes of race and class, offers an example of how digital advertising continues to target and harm racialized populations, such as Black persons in the US, and therefore how we may identify and define an approach to the problems of digital ad targeting. Using zip codes as proxies for the racial character of specific neighbourhoods, McIlwain analyzed display advertisements as well as paid and organic search results in order to ascertain the geographies of targeting. Furthermore, based on the keywords used by those advertising in the payday loan space, the study sought to excavate the sources and targets of potentially predatory advertising.
In other words, the study sought to model targeting intent. With the erosion in the usefulness of disparate impact as a legal and normative framework, the question becomes how we might gather instances of intent as evidence of predatory racial targeting. McIlwain proposes that we understand targeting intent along three indices:
- Ad infrastructure presence on a website as a measure of targeting intent
- Differential representation of, for example, African-American-dominant neighborhoods in the targeted zip codes indicates race-based targeting intent
- Differential reach in terms of, say, breadth of keyword targeting may be used as a measure of the degree of targeting interest, or race-based targeting intent
These indices could further be used to produce ‘ad tech index scores’ along categories of ad targeting, such as A/B testing and retargeting — the density of adtech infrastructure deployed across such categories index the penetration of targeted advertising. In McIlwain’s analyses, they also demonstrated targeting intent through overrepresentation of some keywords in Black zip codes, or the exclusion of others from being used to target audiences in those zipcodes. Financial service providers advertising predatory loans (especially with high interest rates) at Black-exclusive/overrepresented zip codes also tended to have lower levels of trust among general audiences, but claimed more authority in the distribution of such services.
Racially targeted advertising and geographies of discrimination and exclusion, then, are mutually reinforcing. While the explicitly racist norms of the 20th century may have shifted, their historical and cumulative effects remain, as well as different norms that produce similar results (including racially differentiated neighbourhoods). However, it remains difficult to make reliable claims about racial targeting merely using zip codes as proxies for race because of the confluence of adtech platforms, the frequency of data and financial movement through such systems, and the opacity of auction houses that manage ads. There remains a complex and often unpredictable relationship between searches and ads in digital space and their outcomes in physical geographical space. However, this reiterates the usefulness of emphasizing, as McIlwain did succinctly, that “targeting is an intentional practice” for digital advertising. Thinking about intent — especially wherein the targeting of advertising is also the targeting of harms, as with predatory loans — also enables us to follow the traces of advertising to find evidence of racial targeting. It also allows us to move past frameworks of fairness that tread the line of being ahistorical and relying on tabula rasa methodologies and norms.
The leitmotif of Charlton McIlwain’s presentation, then, could be the refusal to discard histories in the digital now. Deep social and geographical histories mark algorithmically constructed online environments. Simultaneously however, McIlwain also cautions us against considering the present a mere reproduction of the past. This carries important lessons for examining the production and circulation of mis- and dis- information, which have often been entwined with the infrastructures of adtech. The embedded and social histories that make the former transmissible shape the targeting of anti-democratic messaging. The networked infrastructures and business sectors that threaten to make of history a flat line are openable to investigation and intervention, and work such as that of McIlwain’s offers us powerful tools to do just that.