Algorithms of Oppression

Home > Nonfiction > Algorithms of Oppression > Page 9
Algorithms of Oppression Page 9

by Safiya Umoja Noble


  In order to fully interrogate this persistent phenomenon, a lesson on race and racialization is in order, as these processes are structured into every aspect of American work, culture, and knowledge production. To understand representations of race and gender in new media, it is necessary to draw on research about how race is constituted as a social, economic, and political hierarchy based on racial categories, how people are racialized, how this can shift over time without much disruption to the hierarchical order, and how White American identity functions as an invisible “norm” or “nothingness” on which all others are made aberrant.

  Figure 2.5. Google search on “Asian girls,” 2011.

  Figure 2.6. Google search on “Asian Indian” girls in 2011.

  Figure 2.7. Google search on “Hispanic girls” in 2011.

  Figure 2.8. Google search on “Latina girls” in 2011.

  Figure 2.9. Google search on “American Indian girls” in 2011.

  Figure 2.10. Google search on “white girls” in 2011.

  Figure 2.11. Google search on “African American girls” in 2011.

  The leading thinking about race online has been organized along either theories of racial formation6 or theories of hierarchical and structural White supremacy.7 Scholars who study race point to the aggressive economic and social policies in the U.S. that have been organized around ideological conceptions of race as “an effort to reorganize and redistribute resources along particular racial lines.”8 Vilna Bashi Treitler, a professor of sociology and chair of the Department of Black Studies at the University of California, Santa Barbara, has written extensively about the processes of racialization that occur among ethnic groups in the United States, all of which are structured through a racial hierarchy that maintains Whiteness at the top of the social, political, and economic order. For Treitler, theories of racial formation are less salient—it does not matter whether one believes in race or not, because it is a governing paradigm that structures social logics. Race, then, is a hierarchical system of privilege and power that is meted out to people on the basis of perceived phenotype and heritage, and ethnic groups work within the already existent racial hierarchy to achieve more power, often at the expense of other ethnic groups. In Treitler’s careful study of racialization, she notes that the racial binary of White versus Black is the system within which race has been codified through legislation and economic and public policy, which are designed to benefit White Americans. It is this system of affording more or less privileges to ethnic groups, including White Americans as the penultimate beneficiaries of power and privilege, that constitutes race. Ethnic groups are then “racialized” in the hierarchical system and vie for power within it. Treitler explains the social construction of race and the processes of racialization this way:

  Racial identities are obtained not because one is unaware of the choice of ethnic labels with which to call oneself, but because one is not allowed to be without a race in a racialized society. Race is a sociocultural hierarchy, and racial categories are social spaces, or positions, that are carved out of that racial hierarchy. The study of racial categories is important, because categories change labels and meanings, and we may monitor changes in the racial hierarchy by monitoring changes in the meaning and manifestations of racial categories.9

  Treitler’s work is essential to understanding that the reproduction of racial hierarchies of power online are manifestations of the same kinds of power systems that we are attempting to dismantle and intervene in—namely, eliminating discrimination and racism as fundamental organizing logics in our society. Tanya Golash-Boza, chair of sociology at the University of California, Merced, argues that critical race scholarship should expand the boundaries of simply marking where racialization and injustice occur but also must press the boundaries of public policy so that the understanding of the complex ways that marginalization is maintained can substantially shift.10 Michael Omi and Howard Winant, two key scholars of race in the United States, distinguish the ways that racial rule has moved “from dictatorship to democracy” as a means of masking domination over racialized groups in the United States.11 In the context of the web, we see the absolving of workplace practices such as the low level of employment of African Americans in Silicon Valley and the products that stem from it, such as algorithms that organize information for the public, not as matters of domination that persist in these realms but as democratic and fair projects, many of which mask the racism at play. Certainly, we cannot intervene if we cannot see or acknowledge these types of discriminatory practices. To help the reader see these practices, I offer here more examples of how racial algorithmic oppression works in Google Search.

  On June 6, 2016, Kabir Ali, an African American teenager from Clover High School in Midlothian, Virginia, tweeting under the handle @iBeKabir, posted a video to Twitter of his Google Images search on the keywords “three black teenagers.” The results that Google offered were of African American teenagers’ mug shots, insinuating that the image of Black teens is that of criminality. Next, he changed one word—“black” to “white”—with very different results. “Three white teenagers” were represented as wholesome and all-American. The video went viral within forty-eight hours, and Jessica Guynn, from USA Today, contacted me about the story. In typical fashion, Google reported these search results as an anomaly, beyond its control, to which I responded again, “If Google isn’t responsible for its algorithm, then who is?” One of Ali’s Twitter followers later posted a tweak to the algorithm made by Google on a search for “three white teens” that now included a newly introduced “criminal” image of a White teen and more “wholesome” images of Black teens.

  Figure 2.12. Kabir Ali’s tweet about his searching for “three black teenagers” shows mug shots, 2016.

  Figure 2.13. Kabir Ali’s tweet about his searching for “three white teenagers” shows wholesome teens in stock photography, 2016.

  What we know about Google’s responses to racial stereotyping in its products is that it typically denies responsibility or intent to harm, but then it is able to “tweak” or “fix” these aberrations or “glitches” in its systems. What we need to ask is why and how we get these stereotypes in the first place and what the attendant consequences of racial and gender stereotyping do in terms of public harm for people who are the targets of such misrepresentation. Images of White Americans are persistently held up in Google’s images and in its results to reinforce the superiority and mainstream acceptability of Whiteness as the default “good” to which all others are made invisible. There are many examples of this, where users of Google Search have reported online their shock or dismay at the kinds of representations that consistently occur. Some examples are shown in figures 2.14 and 2.15. Meanwhile, when users search beyond racial identities and occupations to engage concepts such as “professional hairstyles,” they have been met with the kinds of images seen in figure 2.16. The “unprofessional hairstyles for work” image search, like the one for “three black teenagers,” went viral in 2016, with multiple media outlets covering the story, again raising the question, can algorithms be racist?

  Figure 2.14. Google Images search on “doctor” featuring men, mostly White, as the dominant representation, April 7, 2016.

  Figure 2.15. Google Images search on “nurse” featuring women, mostly White, as the dominant representation, April 7, 2016.

  Figure 2.16. Tweet about Google searches on “unprofessional hairstyles for work,” which all feature Black women, while “professional hairstyles for work” feature White women, April 7, 2016.

  Understanding technological racialization as a particular form of algorithmic oppression allows us to use it as an important framework in which to critique the discourse of the Internet as a democratic landscape and to deploy alternative thinking about the practices instantiated within commercial web search. The sociologist and media studies scholar Jessie Daniels makes a similar argument in offering a key critique of those scholars who use racial formation theory as an or
ganizing principle for thinking about race on the web, arguing that, instead, it would be more potent and historically accurate to think about White supremacy as the dominant lens and structure through which sense-making of race online can occur. In short, Daniels argues that using racial formation theory to explain phenomena related to race online has been detrimental to our ability to parse how power online maps to oppression rooted in the history of White dominance over people of color.12

  Often, group identity development and recognition in the United States is guided, in part, by ongoing social experiences and interactions, typically organized around race, gender, education, and other social factors that are also ideological in nature.13 These issues are at the heart of a “politics of recognition,”14 which is an essential form of redistributive justice for marginalized groups that have been traditionally maligned, ignored, or rendered invisible by means of disinformation on the part of the dominant culture. In this work, I am claiming that you cannot have social justice and a politics of recognition without an acknowledgment of how power—often exercised simultaneously through White supremacy and sexism—can skew the delivery of credible and representative information. Because Black communities live in material conditions that are structured physically and spatially in the context of a freedom struggle for recognition and resources, the privately controlled Internet portals that function as a public space for making sense of the distribution of resources, including identity-based information, have to be interrogated thoroughly.

  In general, search engine users are doing simple searches consisting of one or more natural-language terms submitted to Google; they typically do not conduct searches in a broad or deep manner but rather with a few keywords, nor are they often looking past the first page or so of search engine results, as a general rule.15 Search results as artifacts have symbolic and material meaning. This is true for Google, but I will revisit this idea in the conclusion in an interview with a small-business owner who uses the social network Yelp for her business and also finds herself forced from view by the algorithm. Search algorithms also function within the context of education: they are embedded in schools, libraries, and educational support technologies. They function in relationship to popular culture expressions such as “just Google it,” which serves to legitimate the information and representations that are returned. Search algorithms function as an artifact of culture, akin to the ways that Cameron McCarthy describes informal and formal educational constructs:

  By emphasizing the relationality of school knowledge, one also raises the question of the ideological representation of dominant and subordinate groups in education and in the popular culture. By “representation,” I refer not only to mimesis or the presence or absence of images of minorities and third-world people in textbooks; I refer also to the question of power that resides in the specific arrangement and deployment of subjectivity in the artifacts of the formal and informal culture.16

  The Internet is an artifact, then, both as an extension of the formal educational process and as “informal culture,” and thus it is a “deployment of subjectivity.” This idea offers another vantage point from which to understand the ways that representation (and misrepresentation) in media are an expression of power relations. In the case of search engine results, McCarthy’s analysis opens up a new way of thinking about the ways in which ideology plays a role in positioning the subjectivities of communities in dominant and subordinate ways.

  This concept of informal culture embodied in media representations of popular stereotypes, of which search is an instance, is also taken up by the media scholars Jessica Davis and Oscar Gandy, Jr., who note,

  Media representations of people of color, particularly African Americans, have been implicated in historical and contemporary racial projects. Such projects use stereotypic images to influence the redistribution of resources in ways that benefit dominant groups at the expense of others. However, such projects are often typified by substantial tension between control and its opposition. Racial identity becomes salient when African American audiences oppose what they see and hear from an ideological position as harmful, unpleasant, or distasteful media representations.17

  These tensions underscore the important dimensions of how search engines are used as a hegemonic device at the expense of some and to the benefit of dominant groups. The results of searches on “Jew,” as we have already seen, are a window into this phenomenon and mark only the beginning of an important series of inquiries that need to be made about how dominant groups are able to classify and organize the representations of others, all the while neutralizing and naturalizing the agency behind such representations. My hope is that this work will increase the saliency of African American women and other women of color who want to oppose the ways in which they are collectively represented.

  Google’s enviable position as the monopoly leader in the provision of information has allowed its organization of information and customization to be driven by its economic imperatives and has influenced broad swaths of society to see it as the creator and keeper of information culture online, which I am arguing is another form of American imperialism that manifests itself as a “gatekeeper”18 on the web. I make this claim on the basis of the previously detailed research of Elad Segev on the political economy of Google. The resistance to efforts by Google for furthering the international digital divide are partially predicated on the English-language and American values exported through its products to other nation-states,19 including the Google Book Project and Google Search. Google’s international position with over 770 million unique visitors across all of its properties, including YouTube, encompasses approximately half of the world’s Internet users. Undoubtedly, Google/Alphabet is a broker of cultural imperialism that is arguably the most powerful expression of media dominance on the web we have yet to see.20 It is time for the monopoly to be broken apart and for public search alternatives to be created.

  How Pornification Happened to “Black Girls” in the Search Engine

  Typically, webmasters and search engine marketers look for key phrases, words, and search terms that the public is most likely to use. Tools such as Google’s AdWords are also used to optimize searches and page indexing on the basis of terms that have a high likelihood of being queried. Information derived from tools such as AdWords is used to help web designers develop strategies to increase traffic to their websites. By studying search engine optimization (SEO) boards, I was able to develop an understanding of why certain terms are associated with a whole host of representational identities.

  First, the pornography industry closely monitors the top searches for information or content, based on search requests across a variety of demographics. The porn industry is one of the most well-informed industries with sophisticated usage of SEO. A former SEO director for FreePorn.com has blogged extensively on how to elude Google and maximize the ability to show up in the first page of search results.21 Many of these techniques include long-term strategies to co-opt particular terms and link them over time and in meaningful ways to pornographic content. Once these keywords are identified, then variations on these words, through what are called “long tail keywords,” are created. This allows the industry to have users “self-select” for a variety of fetishes or interests. For example, the SEO board SEOMoz describes this process in the following way:

  Most people use long tail keywords as an afterthought, or just assume these things will come naturally. The porn world though, actually investigates these “long tails,” then expands off them. They have the unique reality of a lot of really weird people out there, who will search for specific things. Right now, according to Wordze, the most popular search featuring the word “grandma” is “grandma sex,” with an estimated 16,148 searches per month. From there, there’s a decent variety of long tails including things like “filipino grandma sex.” For the phrase “teen sex,” there are over 1000 recorded long tails that Wordze has, and in my experience, it misses a lot (it only shows things w
ith substantial search volume). The main reason they take home as much traffic and profit at the end of the day as they do is that they actively embrace these long tail keywords, seeking them out and marketing towards them. Which brings us to reason #2. . . . When there is complete market saturation for a topic, the only way to handle it is to divide it into smaller, more easily approached niches. As stated above, they not only created sites with vague references to these things, but they targeted them specifically. If someone is ranking for a seemingly obscure phrase, it’s because they went out there and created an entire site devoted to that long tail phrase.22

  Furthermore, the U.S. dominates the number of pages of porn content, and so it exploits its ability to reach a variety of niches by linking every possible combination of words and identities (including grandmothers, as previously noted) to expand its ability to rise in the page rankings. The U.S. pornography industry is powerful and has the capital to purchase any keywords—and identities—it wants. If the U.S. has such a stronghold in supplying pornographic content, then the search for such content is deeply contextualized within a U.S.-centric framework of search terms. This provides more understanding of how a variety of words and identities that are based in the U.S. are connected in search optimization strategies, which are grounded in the development and expansion of a variety of “tails” and affiliations.

 

‹ Prev