Knowledge is power, and today, the apex of knowledge production and dispersal is through the Internet. However, not all data is equally accessible. In the information age, accessibility of knowledge is deeply dependent on algorithmic methods of organization. There are too many articles, think pieces, videos, and databases for any one person to manually sort. Instead, we rely on digital methods of categorization and data retrieval to navigate the massive supply of available information. These methods of organization, however, necessarily privilege some content over others. It is important, then to recognize how the structural biases present in day-to-day society are reproduced and reinforced algorithmically in digital culture. Kimberlee Crenshaw’s theory of intersectionality and Jasbir Puar’s assemblage help to expose the inherent prejudices of dominant search engines’ methods of data sorting. The algorithms of Google search display the same structural biases exposed by Crenshaw and Puar, stabilizing privileged identities and invisibilizing those with multiply marginalized identities. One must examine the institutional prejudice inherent in Internet culture through the methods of indexing, linking, and applying key terms in searches.
Google search engine’s method of indexing contradicts the very root of intersectional theory. Kimberlee Crenshaw’s intersectionality is a theory in which the constituent parts of a person’s identity do not equal the whole. She defines intersectional experience as, “Greater than the sum of racism and sexism” (Crenshaw, 140). It is “greater than the sum” of these two “single-axis” oppressions because the marginalization faced by Black women is more than that of Black men and white women combined (Crenshaw, 139). She faces oppressions unique to her indivisible Black, female identity. In this theory, two words describe one fluid, distinct experience. Indexing, conversely, is all about breaking down the whole into constituent parts. For example, when you search the phrase, “Brad Pitt,” the search engine looks for “Brad” and “Pitt” independently before recombining the terms. Similarly, when one searches for content through an intersectional lens, for example with the phrase, “Movies featuring disabled women of color,” Google breaks the query down word by word. This deconstructive means of categorizing data reinforces societal conceptions that identity can be broken down by category. As Crenshaw notes, the societal blind-spot towards, “Compoundedness… is also due to the influence of a way of thinking about discrimination which structures politics so that struggles are categorized as singular issues” (Crenshaw, 166-167). The “singular issues” approach to data retrieval does not hurt those with singularly marginalized identities. A search for, “Gay Movies,” will effectively produce a list of movies featuring gay themes, and for a white, cis, able-bodied, gay man, this may be sufficient for finding representation in the search results. The separation of, “disabled, “women,” and “of color” by means of indexing, however, entirely changes the meaning and ranking of results produced by the search. Results featuring only partial combinations of the search terms do not necessarily reflect the experience of the sought out identity.
Search engines marginalize intersectionally oppressed voices not only in how they break up information, but also in how they rank it, specifically through linking. This function explains why when you search “Gay Movie,” results for more popular gay movies may outrank those with a higher frequency of the phrase, “Gay” and “Movie,” even though this contradicts the “Location/Frequency Method” of ranking search results (Sullivan, 3). Any SEO resource will stress that the most valuable way to keep your content at the top of the web, is through high-quality linking. If your website is “linked,” that just means that other websites have referenced and shared it. The implications of linking for status are analogous to structures in non-digital society, where high-profile connections can help one achieve equal higher rank. In fact, this unofficial fact of life is coded into the algorithm of search engines. Inequality in linking practices can be found by examining what gives a website high status. One study of SEO explains that, “The link pattern for the site can give good hints regarding the quality of the site, if good sites link to it, and it links to good sites, it probably is a good site on its own as well” (Johansson, 9). Despite attempting to explain the inner workings of an algorithm, Johansson’s use of subjective language offers insight into the prejudices coded into web-ranking. “Good” abbreviates for a multitude of statistics, such as web traffic, “reputable” domain extensions (.com/.org), and many other factors used to establish digital hierarchy. “Good” sites are often the most frequently and broadly visited, which rarely reflects sites that are targeted towards specific marginalized identities, instead catering to privilege as the norm, expecting further oppressed people to find their experience within these narratives.
Linking as a ranking method reproduces a, “Top-down approach to discrimination,” against which Crenshaw explicitly warns (Crenshaw, 167). The “top-down approach” means that in discussions of identity, the most privileged members of a category are considered “neutral,” and therefore, given the most attention. Multiply marginalized individuals are, in this method, considered less pure representatives of the overarching experience of any given identity. Just like how conversations of feminism not specifically mentioning Black women default to the white female experience, the search results for any given identity default to the most privileged version of the category. This phenomenon is evidenced by my Google search results to general terms such as “woman” and “black person.”
The vast majority of “woman” results were white, while men dominate the search results for “black person.” These results are not neutral – they clearly demonstrate a “top-down” prejudice. There is not, in my knowledge, something overtly written into Google’s code that defines “woman” as white, or “black person” as male. Instead user interfacing and linking practices generate these oppressive results. Stock image websites link a majority of white women to the search term “woman,” articles on women further bolster the linking statuses of these images of white women as desirable results. Through this digital networking process, privilege pulls up privilege, maintaining systems of oppression and ignoring intersectional identity. “Good sites,” that can provide the most valuable boost in Internet rank, therefore, reinforce white/straight/able/cis supremacy, and push multiply marginalized results to the bottom of the Internet.
The third element that under-serves intersectionally marginalized identities in the organization of the Internet is the use of “keywords.” Key terms force multiply marginalized voices to represent, claim stable identities, and ultimately get drowned out by more popular, single-axis applications of the individual terms. When you search for something, keywords are what you type in the search bar, to match content, “In the title, meta description, main heading, and body text,” of your results (Johansson, 8). These are the primary places to saturate a site with the most relevant, popular keywords that a searcher might use if they are interested in the page’s content. The pressure, then, for those with intersectionally oppressed identities, is that to make one’s work findable to those with similar experiences, one must place their societal labels at the forefront of their content. This may be entirely acceptable to activism-based web pages all about reclaiming labels, however, for other content this poses an oppressive dilemma. In order to stand out on the Internet, one must elevate categories of identity over all else, eliminating the ability to transcend labels, while still addressing their marginalized experience. When explaining the legal issues of single-axis discrimination policies, Crenshaw noted that, “They (Black women) can receive protection only to the extent that their experiences are recognizably similar to those whose experiences tend to be reflected in anti discrimination doctrine” (Crenshaw, 152). Substitute “protection” for “findability” and this statement reflects the issue of keywords in Internet identity politics. Content is only findable to the degree that it is categorizable within known, defined experiences. Therefore, participants in Internet culture are forced to either take on the burden of representing their categorical labels, or embrace obscurity.
Beyond the pressure of representation within keyword use, the labeling process promotes a stabilization of identities that upholds structures of oppression. Keywords essentially function as modifiers to the norm, which as was already addressed, is not neutral. This engineers the same problematic dynamic that Puar identifies in intersectional theory; “Distinct from a frame that privileges ‘difference within,’ ‘difference from’ produces difference as a contradiction rather than as a recognizing it as a perpetual and continuous process of splitting” (Puar, 53). “Difference from,” is a critique on categorization as a negative process–defining someone by what they are not. Blackness, for example, is more than the absence of whiteness. However, because an Internet search for “Woman” produces predominantly white results, the addition of “Black” as a keyword is mostly for use as a distinction from whiteness. This function erases variation and fluidity within Black identity, algorithmically privileging “difference from” over “difference within.” Through these methods, search engines utilize, “WOC as a mere enabling prosthetic to white feminists” (Puar, 54). “Of color” solely functioning as a modifier to “women” objectifies and simplifies Black women’s experiences as tools of defining whiteness, rather than expressions of the true diversity within Black female identity. Inherent search bias contributes to the stabilization of privileged identities as both the norm, and the most important lenses through which to examine marginalized experience.
The third way in which keyword sorting fails intersectional web content, is through unfairly placing it in competition with material that uses the same terms to express single-axis oppressions. By this process, more “mainstream” websites drown out pages focusing on multiply marginalized identities. In the search, “Movies featuring disabled women of color,” the keywords “disabled” “women” and “of color” are so popularly utilized in isolation, that their joint application is buried on the Internet. The lack of search terms dedicated to intersectionality, forces multiply marginalized identities into competition with more privileged offshoots of their experience for online real estate. Though Puar postulates that, “‘women of color,’ … has now become, I would argue, simultaneously emptied of specific meaning in its ubiquitous application and yet overdetermined in its deployment,” the newfound fame of “women of color” as a term makes them searchable (Puar, 52). Without “ubiquitous applications,” other multiply marginalized identities, or those with lesser cultural cache, fall victim to indexing’s attempt to make sense of them through a breaking-down process. The lack of either alternate, solely intersectional, keywords, or an algorithmic method that understands the interconnectivity of intersectional identity forces multiply marginalized content behind that which addresses single-axis oppression.
Despite all these instances of algorithmic oppression, search engines and their methods of organization are not hopeless with regard to reform. Rather than wholly rejecting the Internet as biased, it is instead important to recognize that machines are not neutral. Megan Garcia defines algorithmic bias as, “When seemingly innocuous programming takes on the prejudices either of its creators or the data it is fed” (Garcia, 112). Coders who write for Google and other prominent search engines’ ignorance informs their code. The lack of coders with multiply marginalized identities, and the absence of intersectionality theory form coding culture produces single-axis methods of sorting digital data. Conversely, if representation and awareness of multifaceted experience were a greater priority in computer culture, new manners of organizing information could elevate intersectional materials from the depths of the Internet. In fact, it is necessary that these measures are taken, because otherwise we are, “seeding self-teaching AI with the discriminatory undertones of our society in ways that will be hard to rein in because of the often self-reinforcing nature of machine learning” (Garcia, 112). Here, Garcia references the increasingly common reality that code is iterative. One coded instruction implements itself broadly, basing future developments on its foundations. Even if human society evolves to embrace intersectionality, the single-axis organization of data on marginalized identities will fight to maintain old structures of oppression.
Part of the potential for an intersectionally inclusive digital space stems from the Internet’s analogousness to an assemblage. Puar defines assemblage when she explains that, “Categories—race, gender, sexuality—are considered events, actions, and encounters between bodies, rather than simply entities and attributes of subjects” (Puar, 58). The Internet naturally functions in “events, actions, and encounters,” because, as is implied in the word “web,” the Internet manufactures meaning through connection and time. Real-life events create surges in the popularity of topics, and therefore forge new connections within the World Wide Web. This ephemeral, iterative process of sorting and creating information, in conjunction with relevance, forces search engines to transcend simple grid-structures of organizing data. Code accounts for the flexibility of importance and meaning of words over time. With this in mind, indexing, linking, and keyword sorting are not static functions. The results of an indexing process, the value of specific links, and the relationships between keywords and content evolve with our society. This fluidity of categories and meaning through event relationships, makes the structure of the Internet resemble and assemblage. The inherent flexibility of meaning on the web serves as potential for a more complex, flexible presentation of identity. To summarize, the assemblage of digital meaning allows room for a better way of categorizing intersectional content. This would require an adjustment of the coded mechanisms within search engines, but if prioritized, has great potential for improvement.
So much is written on the importance of intersectional representation, but few focus on the significance of what happens to that content after its creation. In a digital society where people shape their worldview based on Internet-generated answers, the methods used to organize that data are undeniably significant. Machines do what they are taught, so if we do not teach them to see certain identities, those experiences become invisible. Our practices of breaking down concepts into constituent parts, in conjunction with a rating system that favors the mainstream, push down the most marginalized members of our society.
True, reprogramming our computers to prioritize content that reflects intersectional experience would, as a result, lower the rank of some single-axis content, but that would not be detrimental. As Crenshaw famously noted on the topic of white feminists, “If their efforts instead began with addressing the needs and problems of those who are most disadvantaged and with restructuring and remaking the world where necessary, then others who are singularly disadvantaged would also benefit” (Crenshaw, 167). Applying Crenshaw’s logic, pushing up intersectional content on the Internet would not harm those with singly oppressed identities. In fact, it would even benefit them by increasing awareness of societal structures of marginalization. It is easy to recognize your own oppression inside that of someone who faces more axis of marginalization. This is why white women can go to benefits promoting the education of African girls and sympathize with the systemic sexism. The privilege of identification, however, does not work the other way around. When one is intersectionally oppressed, seeing only those with single-axis oppressions is not sufficient in encapsulating their experience. With this knowledge, we need to restructure how online culture categorizes data. If intersectional thought is, indeed, to be a societal priority, we must program it into our computers. Conversely, by coding intersectional theory into our digital sphere, perhaps it will restructure our real-life society.
Works Cited
Crenshaw, Kimberle (1989). Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics. The University of Chicago Legal Forum 140:139-167.
Garcia, Megan. “Racist in the Machine.” World Policy Journal. N.p., 01 Jan. 1970. Web. 27 Apr. 2017.
Johansson, Dennis. “Search Engine Optimization and the Long Tail of Web Search” Beginning Joomla! (n.d.): 415-35. 16 June 2016. Web.
Nash, Jennifer C. “Rethinking Intersectionality.” Feminist Review 89.1 (2008): 1-15. Web.
Puar, J. K. ““I would rather be a cyborg than a goddess”: Becoming-Intersectional in Assemblage Theory.” philoSOPHIA, vol. 2 no. 1, 2012, pp. 49-66. Project MUSE, muse.jhu.edu/article/486621.
Sullivan, Danny. “How Search Engines Work.” N.p., 14 Oct. 2002. Web