Our lives are increasingly becoming algorithmic. Algorithms automatically extract our data, categorize us, profile us, and have deep and often tangible effects on our lives (Zuboff 2019; Couldry and Mejias 2019; Eubanks 2018; Noble 2018; O’Neal 2016). Our “algorithmic profiles” affect the content that we see online, our chances of getting a job or a loan, our relationships with our friends, colleagues, and bosses, and even how we express and understand ourselves (Karakayali et al. 2018). That is, our life chances are increasingly being affected by the algorithmic gaze. We are all perpetually seen by this gaze, and are deeply influenced by it. But who is “we”?

While in the last decade the power of algorithms has been thoroughly theorized (Bucher 2018; Gillespie 2016; Neff et al. 2012; Zuboff 2019; Couldry and Mejias 2019; Cheney-Lippold 2011; Beer 2009), and while the social consequences of algorithms have been systematically discussed (Leese 2014; Turow and Couldry 2018; O’Neal 2016; Eubanks 2018), research on algorithms tends to assume that the companies that write and run big data algorithms are Euro-American, and accordingly, that the global spread of such algorithms is unidirectional—from “the West” onwards. That is, while recent research has shown that algorithms stem from specific socio-cultural contexts (Seaver 2018, 2015; Wilf 2013; Christin 2018; Ribak 2019; Kotliar 2020a), and that data tends to mirror the social surroundings from which it was extracted (Angwin et al. 2016; Crawford 2013; Hargittai 2018; Noble 2018), the geographical and cultural distances between those who develop profiling algorithms and those who are being profiled remains overlooked. Moreover, while recent research tends to compare algorithmic powers with colonial ones (Couldry and Mejias 2019; Mann and Daly 2019; Thatcher, et al. 2016), the move from the colonial gaze (Yegenoglu 1998) to the algorithmic gaze (Graham 2010) has yet to be fully discussed. This article aims to fill these gaps, based on an ethnographic examination of Israeli companies’ attempts to get their profiling algorithms into East Asia. I ask here: What happens when the geographic, epistemic, and cultural distances between those who surveille and those who are being surveilled are exceptionally wide? How do tech companies deal with such distances, with such cultural differences, and how do they get their algorithms to “see” their “Other”?

Algorithmic power

While the meaning of the term “algorithm” changes from one social actor to the next (Gillespie 2016; Seaver 2017), and while countless types of algorithms, as well as countless non-algorithmic actors (Morris 2015; Kitchin 2017; Neyland and Möllers 2017), take part in the algorithmization of social life, algorithms’ effects on people’s lives are becoming unmistakable. Based on perpetual digital surveillance, and on algorithms’ capacity to categorize very large datasets swiftly, algorithms have become very efficient and very ubiquitous social sorting mechanisms (Lyon 2003; Willson 2014)—socio-technical sieves (Kockelman 2013) that perpetually categorize people into “clusters” (Kotliar 2020b.), and deeply affect their lives (Rifkin 2001; Eubanks 2018; Graham 2005). In other words, algorithms have become a dominant means of organizing—and governing—people’s actions (Turow and Couldry 2018, p. 419), as their choices (Kotliar 2020a; Graham and Henman 2019; Cohn 2019; Yeung 2017) and even autonomy and freedom (Christin et al. 2015; Rouvroy 2013) are incessantly mediated, restricted, and afforded by algorithmic systems.

Algorithms also play a role in reality construction (Couldry and Hepp 2018; Just and Latzer 2017): With the growing ubiquity of personalization engines, recommender systems, and user analytics algorithms, the ways in which algorithms are being used to “see us” increasingly affect how we come to see the world. Accordingly, our sense of self (Karakayali et al. 2018) and our perceptions of our community (Gillespie 2014), friends (Bucher 2012), and even lovers (Bivens and Hoque 2018) are increasingly mediated by the algorithmic gaze—by the attempts to characterize, profile, and affect people algorithmically.Footnote 1

Like previous types of social classification (Bowker and Leigh Star 1999, pp. 195–227; Uvin 2002), this algorithmic way of seeing can also have harsh consequences—algorithms have been shown to mirror and at times exacerbate social inequalities (Graham 2005; Eubanks 2018; Noble 2018; O’Neal 2016), increase individuation (Rouvroy 2013; Prey 2018), and cause political polarization (Tufekci 2014; Vaidhyanathan 2018). They have also been accused of spreading hate speech (Neff and Nagy 2016) and of depriving people of their free will, liberty, and autonomy (Cooper 2020, 12). That is, in the last decade, critical algorithm scholars have provided ample evidence of algorithms’ effects on peoples’ lives, exposing how interwoven society and algorithms have become and how harmful these systems can be.

The power of algorithms has also been widely theorized. Algorithmic power has been described as a sovereign power (Zuboff 2019), as an institutional power (Napoli 2014; Napoli and McGannon 2013), as soft biopower (Cheney-Lippold 2011), or as a Foucauldian disciplinary power (Bucher 2012). It has also been described as post hegemonic (Beer 2009; Lash 2007) and post textual power (Andrejevic, et al. 2015), and as an info-political power (Rogers 2009). Algorithmic power has even been compared to the work of pastors (Cooper 2020) and even God (Gillespie 2014, p. 168). More recently, however, in light of algorithms’ insatiable expansion, scholars have begun to refer to their power as “data colonialism.”

Data colonialism: The global expansion of the algorithmic gaze

With the rapid proliferation of algorithmic systems across social fields, and with the expansion of the surveillant assemblage (Haggerty and Ericson 2000) across the globe, scholars have begun to highlight the similarities between the power of big data algorithms and that of a previous practice of domination—colonialism. Thatcher, O’Sullivan, and Mahmoudi described the extraction of personal data as an “accumulation by dispossession and colonization of the life-world” (Thatcher et al. 2016). Couldry and Mejias similarly argued that data colonialism “combines the predatory extractive practices of […] colonialism with the abstract quantification methods of computing” (Couldry and Mejias 2019, p. 1). Ricaurte argued that the coloniality of data collection is manifested in a “violent imposition of ways of being, thinking, and feeling that […] denies the existence of alternative worlds and epistemologies” (Ricaurte 2019, p. 381), and Mann and Daly discussed the “double nature of data colonialism,” highlighting the fact that data practices can affect everyone, but are particularly harmful to People of Color (Mann and Daly 2019, p. 381).

More specifically, Couldry and Mejias recently focused on data colonialism’s tendency for expansion. According to them, what makes this phenomenon colonial is “the scale and scope of this worldwide network of extraction and distribution” (Couldry and Mejias 2019, p. 38). They accordingly explain that the expansion of data colonialism is both external (geographic) and internal (social). That is, data colonialism simultaneously seeks new territories to set its algorithmic eyes on, as well as new aspects of sociality that have previously been missing from its purview. Couldry and Mejias also remind us that while capitalism has been expansionary from its onset (based on the exploitation of human production through labor relations), the ability to appropriate human life in the form of data offers new expansionary possibilities. With data colonialism, capitalism extends its capacity to exploit life by assimilating new or reconfigured human activities as its direct inputs (Couldry and Mejias 2019, p. xii). In that, human beings are becoming raw material that can be transformed into value in ways that resonate with colonialism’s long history of exploitation. As mentioned above, this expansion necessarily entails an expansion in the prevalence and sophistication of surveillant assemblages (Haggerty and Ericson 2000)—both geographically and internally—as companies incessantly seek new datafiable sources of revenue (see also Zuboff 2019).

Nevertheless, while the term colonialism indeed refers to practices of domination and exploitation, it also pertains to movement in space—to the transfer of populations to new territories and their permanent settlement there (Kohn and Reddy 2017), as well as to the often-violent intercultural and interracial encounters between colonizers and the colonized. However, although this is a central aspect of colonial power, the study of algorithms has largely overlooked the role algorithms play in such intercultural “encounters,” as well as the specific ways in which algorithms move across space and “colonize” distant lands. Namely, while the role of culture in algorithmic systems is beginning to unfold (Christin 2018; Ribak 2019; Seaver 2018; Takhteyev 2012; Kotliar 2020a), the ways in which algorithms are being programmed to cross intercultural boundaries—the ways companies make algorithms that would “fit” (and affect) people from places thousands of miles away from their site of production—remains underexplored.

As Shapin argued more than two decades ago: “we need to understand not only how knowledge is made in specific places, but also how transactions occur between places (Shapin 1998, p. 7). Put in the context of profiling algorithms, we need to understand not only how local cultures affect the creation of algorithmic systems, but also how algorithmic attempts to understand users occur between two very distant localities. Such a focus will provide us with a more nuanced understanding of algorithms and their power, as well as of their global expansion. This article aims to fill this gap by offering an empirical account of Israeli companies’ attempts to get their algorithms into East Asia. I do so by focusing on the algorithmic version of a very central colonial practice—the effort to conceptualize, categorize, and construct the Other.

From the colonial gaze to the algorithmic gaze

Comparing the power of algorithms to that of colonialism is a radical theoretical move: Equating a centuries-long succession of conquests, subjugations, and appropriations that resulted in mass death, suffering, and forced displacements to an automated collection and analysis of personal data may well be an overstatement. However, the involvement of algorithmic systems in instigating ethnic conflicts (Hagerty and Rubinov 2019), in destabilizing democracies (González 2017), and as “computational theocracies” (Bogost 2015, Cooper 2020)Footnote 2is gradually unfolding, and algorithms’ tendency to reinforce racism (Noble 2018; Benjamin 2019) and cause deep social suffering is becoming almost unequivocal (Eubanks 2018; O’Neal 2016; Dencik et al. 2018). At the same time, the global expansion of algorithmic powers and their gradual instilment into societies across the globe is indisputable. Taken from this perspective, the concept of data colonialism can offer a useful theoretical framework to examine the expansion of algorithmic powers into new cultural and geographic territories. But to do that, we need to compare the specific power practices that sustain these two very different types of colonization and, particularly, the role knowledge plays in these two expansionary global powers.

After all, colonialism stemmed not only from the West’s outstanding military abilities or from unstoppable processes of modernization or capitalization, but it was also reinforced by the creation of knowledge about colonized people. As Dirks argued, referring to the work of Bernard Cohen: “in certain important ways, knowledge was what colonialism was all about” (Cohen 1996, p. ix). That is, the perpetual effort to count people, collect information about them, and categorize them into racial, ethnic, and cultural categories has played a crucial role in the creation, sustainment, and justification of colonial powers (Kertzer and Arel 2002; Kukutai and Broman 2016; Ittmann et al. 2010). Extending colonial rule into new territories necessitated the creation of clear distinctions between the settlers and the indigenous—between the colonizers and the Others, and to that aim, the Others had to become “knowable” (Kertzer and Arel 2002, p. 9).

The ways in which colonial powers came to “see” the Other served to (re)construct and reify group boundaries and relations (Hirschman 1986), to determine ethnic and racial divisions, and at times, to turn the newly categorized ethnic groups against one another (Kertzer and Arel 2002; Kukutai and Broman 2016). This knowledge production also played a role in justifying the colonial rule in the eyes of the colonizers, and in naturalizing the social order that was created by this rule. Moreover, conceptualizing the Other was also the West’s way of defining itself “by way of a detour through the Other” (Yegenoglu 1998, p. 1; Said 1995). In fact, contemporary surveillance practices are in large part direct legacies of such colonial knowledge generation practices (Berda 2013).

But the effort to understand and categorize the Other was by no means restricted to explicit surveillance. Throughout the colonial era, “The West” used various epistemic means to conceptualize, depict, and (re)construct the Other. This type of cultural differentiation was famously described by Edward Said as Orientalism—“a style of thought based upon an ontological and epistemological distinction made between ‘the Orient’ and […] ‘the Occident’” (Said 1995, p. 2). According to Said, the Orientalist gaze offers “a Western style for dominating, restructuring, and having authority over the Orient” (ibid., p. 3). Following Foucault (1977), Said predominantly saw Orientalism as a discourse (produced by a wide range of actors—poets, novelists, philosophers, theorists, economists, administrators, and others) that systematically devalues the Oriental Others, depicting them as backward, primitive, or weak. Through such discourses, Others are seen as belonging to a “subject race” that is differentiated from and dominated by a supposedly superior race that allegedly knows them better than they could possibly know themselves (Said 1995, p. 35). Said’s successors have similarly pointed at how different forms of knowledge—censuses (Kertzer and Arel 2002; Ittmann et al. 2010; Kukutai and Broman 2016; Anderson 1991), sciences (Seth 2009), statistics (Asad 1994), art (Benjamin 2003), broadcast media (Morley and Robins 1995), and more—are being used to “see,” constitute, and dominate the Other. Thanks to these scholars, we can now chart the ties between knowledge production and colonial power when artists use their brush to paint the Other (Benjamin 2003), when novelists write novels about the Other (Aravamudan 2011), or when ethnographers muse about the Other (Sax 1998). But what does it mean when programmers use their algorithms to “see” and profile the Other? What happens to this Foucauldian knowledge/power nexus when knowledge about the Other is algorithmically produced? And what is the role of the algorithmic gaze in the expansion of data colonialism, in the perpetual effort to push back the “data frontiers” (Beer 2018)? This article answers these questions by focusing on a case study of an Israeli data analytics company and its attempts to sell its profiling algorithms to companies in East Asia.

This article is part of a larger ethnographic study that explores the Israeli data analytics industry and, particularly, high-tech companies that produce user-profiling and user-analytics services. Conducted between 2013 and 2018, this multi-sited ethnography (Marcus 1995) included forty semi-structured interviews with Israeli data scientists, programmers, marketers, CEOs, and investors; participant observations in various data-analytics’ events (conferences, hackathons, professional meetings, meetups, and more); and participation in data scientists’ Facebook and WhatsApp groups. I also collected and analyzed data analytics firms’ websites, promotional videos, and other publicly available data, and participated in four data-mining courses. The different data were logged into MaxQDA18 for clustering and analysis: Using the thematic analysis method (Braun and Clarke 2006) I read and reread the data, identified recurrent themes and major concepts, and clustered similar segments together. Subsequently, I highlighted illustrative quotations from each category, according to an interpretive-constructivist approach (Heikkinen et al. 2000).Footnote 3

Accordingly, this article sees knowledge production as a socio-technical process that is inherently political. Knowledge about people is never objective, nor neutral, but is always the result of diverse interpretive processes, created in changing socio-technical contexts (Bowker and Leigh Star 1999; Bucher 2018; Hacking 1995). Thus, this article sees the knowledge that algorithms create as socio-algorithmic constructions that stem from complex socio-technical assemblages (Kitchin 2017, p. 16) and operate through “distributed agency,” alongside a wide array of people and things (Neyland and Möllers 2017; Seaver 2017). However, while the majority of critical algorithm studies focus on how people see, understand, or are being affected by algorithms, this article offers an empirical exploration of the social drama behind the production of algorithmic systems and of the ways in which people use their algorithms to “see” and construct the Other.

For the sake of brevity, this article focuses on one such company—ExtractiveFootnote 4—and its attempts to sell its algorithmic products to companies in East Asia. I start by discussing Extractive’s attempts to “code against culture”—or, design data analytics algorithms that are “culture agnostic”. I then continue by discussing how Extractive labels its algorithmically produced user categories. I next explore the non-algorithmic view of East Asian clients and users, as described by Extractive’s senior executives. Lastly, I argue that Extractive’s Algorithmic gaze paradoxically conjoins three very different but deeply interrelated views of the Other and thus paves Extractive’s way into new territories.Footnote 5

The case of Extractive

Extractive is an Israeli start-up company that provides user profiling algorithms to companies in East Asia. From their offices in Israel, and based on multiple data sources, Extractive generates user analytics—datafied characterizations of users that allegedly help their clients—companies in China, Singapore, and the Philippines—form a “better understanding” of their customers.Footnote 6 As Doron, the company’s chief data scientist, explained in an interview:

Our work primarily focuses on Telcos [telecommunication companies] in the East. The biggest problem these companies have is that people there often buy prepaid [SIM cards] because they don’t have enough money [to pay for contractual deals]. And then they [the companies] say: ‘I don’t know anything about them!’ But they downloaded our app, right? And they are logging into it through Facebook … and then—behold—I can link a phone number to a person. I can suddenly know who they are, and what they do—including their name, their picture—everything! And that’s very significant—before that, they knew absolutely nothing about them.Footnote 7

Doron explains how Extractive’s data analytics algorithms are being used to see, and know, customers in “the East.” According to him, these companies’ customers tend to use prepaid SIM cards, and hence, remain unknowable and, in effect, inaccessible to telecommunication companies. Extractive’s services aim to solve this problem by gaining direct access to users’ Facebook accounts, as well as to their phones: By getting users to log into their app using their Facebook credentials, Extractive makes them (unknowingly) agree to transfer their data right into their hands. From that data, Extractive then creates detailed user profiles—algorithmic characterizations that turn anonymous customers into identifiable persons. These profiles can later be used by the telecommunication companies to segment their customers and approach them with more personalized offers, services, contents, or prices, and hence turn them into more profitable or more “loyal” customers. In other words, Extractive offers its algorithmic gaze to clients in various East Asian countries, who can then turn their anonymous users into seeable, knowable, and accessible customers.

I asked Dalia, Extractive’s VP, how they generate user analytics if they are not familiar with the language or culture of the users they analyze. She answered:

Our algorithms allow us to understand people in a "language-agnostic way"—we can understand Chinese speakers, Russian speakers, and practically any other speaker. Psychologists would have needed to speak their language in order to understand them, and use psychological tools to conceptualize them. For them, language is a crucial tool, but for us? With our algorithms, we use only one tool to understand many languages.

Dalia describes the algorithms Extractive creates as something that can conceptualize people regardless of their language, nationality, or culture. She highlights the universalizing potential of her algorithms by contrasting it with another highly universalistic episteme—that of psychology. While psychology traditionally assumes that its theories, discourses, and categories are universal—namely, that they can fit people from almost any culture—Dalia points at what she sees as the basic flaw in this type of universalism—its dependence on language. According to her, while psychologists need to actively learn the language of people to approach them and offer translated diagnostic or therapeutic tools to understand them, Extractive’s algorithms can allegedly categorize people “in a language-agnostic way.” That is, according to Dalia, her algorithms offer a highly universalizing gaze that can allegedly, and in contrast to previous, modernistic types of human classification, transcend lingual or cultural boundaries. This is what I term “coding against culture”—the creation and use of algorithms that presumably generate knowledge about people while overlooking their lingual or cultural characteristics.Footnote 8

When asked how they achieve such universalizing abilities, Dalia responded with an example:

Our Facebook-based profiling algorithm is completely language agnostic. We demonstrated it recently to one of the top managers in Singapore—we did a demo, and he was very happy with it! He said: ‘right! I love sports! I’m this, I’m that—you’re right!’. You see, whether [users’ Facebook] profiles are in Russian, German, Chinese, or Hebrew, our system can understand people in a language-agnostic way, with absolutely no relation to their language or culture. So basically, thank god for metadata, and kudos to Facebook for their structuring.

As described above, Extractive collects Facebook data using their app and algorithmically interprets the data into detailed user profiles. As Dalia explains, the data they get from Facebook is very well structured, and thus, Extractive can categorize users based on their metadata rather than the actual content they post online. As a case in point, the Singaporean executive Dalia describes was categorized as “Sporty” not because he discussed sports in his Facebook posts, nor because he self-identified as such, but because his actions on Facebook—his likes, shares, and comments—were automatically classified as sports-related. Thus, Extractive’s algorithms do not need to “read” people’s posts in order to “understand” and categorize them—they only need to gain access to Facebook’s data, and a basic understanding of the structure of this database.Footnote 9

As van Dijk explained: “Metadata are not merely a by-product of user-generated content, they are a prime resource for profiling real people with real interests” (van Dijk 2009, p. 49). In this case, metadata functions as an inter-cultural bridge—one that gives this Israeli company access to East Asian users, and hence, to East Asian markets. That is, the use of metadata, alongside Facebook’s well-structured database, are among the technological factors that make Extractive’s algorithms “language agnostic,” and that accordingly afford it its hyper-expansionary potential. The use of metadata also exemplifies what Antoinette Rouvroy described as “data behaviorism”—“[a] way of producing knowledge about [people’s] preferences, attitudes, behaviors or events, without considering the subject’s psychological motivations, speeches or narratives” (Rouvroy 2013, p. 1). It is a way of categorizing people based on their online actions (their likes or shares), rather than on their demographic or biographical details, and a categorization that is no longer based on predetermined categories of meaning, but on automatic pattern-recognition within troves of user-data (Fisher and Mehozay 2019; Beer 2009; Andrejevic 2013). Such a categorization also creates mutable, hyper-individuated identities (Lake 2017; Cheney-Lippold 2017), instead of more traditional, stable ones, as people can get dynamically and automatically sorted into countless “clusters” instead of into a limited number of known categories (Kotliar 2020b). In this case, based on their use of Facebook data, Extractive can focus on the online actions of their Others rather than on seemingly more stable characteristics such as their ethnicity, nationality, or race. In this manner, their system can “understand” their Others while disregarding their language or culture.

But Extractive’s work is not limited to Facebook data. Extractive’s cooperation with telecommunication companies also grants them access to another source of data which is quite controversial—Deep Pocket Inspection (DPI).Footnote 10 DPI allows companies to monitor the specific ways in which users use their smartphones and hence gain access to their location, their text messages, their apps, and more. I asked Gavriel, a data scientist at Extractive, how they profile Singaporean users with DPI, and he replied:

We don’t need to make special adjustments to do that [make their algorithms fit Singaporean users] because we can now access lots of data, and virtually work with any vertical. So we don’t care if you’re [using] a news app, a coupon app, or a parking app, we just don’t care—we are entirely detached from context. Our methods are the same methods, the model is the same, even if the content isn’t.

At first glance, it seems like Gavriel conflates cultural heterogeneity with heterogenous content—as he answers a question about adjusting an algorithm to fit a specific group by referring to the types of markets (or “verticals”) his algorithms can operate in. But in fact, Gavriel’s answer highlights how deeply intertwined the two fields are for Extractive: For them, having a content-agnostic algorithm is crucial to having a culture agnostic one. The plethora of data they receive from using DPI, alongside the ability to use metadata as their main source of data, is essential for their pretention to overlook culture.

This type of categorization, alongside the previously mentioned attributes of the algorithmic gaze, underscores these algorithms’ expansionary logic. It is a logic that aims to exceed the logic of modern classification systems—like psychology or demography—and hence go far beyond the “traditional” colonialist gaze. As Doron, Dalia, and Gavriel explain, their culture agnostic algorithms, with their data-behaviorist, hyper-individuating gaze, allow them to conceptualize people with no relation to their locality, race, or culture and, hence, to quickly expand from their offices in Israel far into “the East.”

In other words, while the colonial expansionary logic was based on the ability to make populations “knowable” (Kertzer and Arel 2002) and differentiate between people based on their alleged membership in racial, ethnic, or cultural groups, algorithms offer a different expansionary logic—one that favors dynamic behaviors over seemingly stable identities, mutable categories over bounded groups. It is a gaze that views and characterizes the Other while rejecting the “groupist way of thought” (Brubaker 2004, p. 11) that traditionally favors more stable categories, like ethnicity or race. Thus, while colonial census makers were passionate about having unambiguous, complete categories (Anderson 1991, pp. 164–165), algorithmic categories are often far from complete. However, while “language agnostic algorithms” may overlook cultural differences, they do so only to point at other, allegedly more fine-grained differences. After all, difference is the bedrock of the data-analytics industry—knowing users, producing analytics about them, necessarily entails differentiating between them. And hence, these algorithms offer a universalizing perspective, no doubt, but one that favors inter-personal differences over inter-cultural ones.

But how are these differences characterized? How does this company name its algorithmically-formed categories? As I have shown elsewhere (Kotliar 2020b) while algorithmic categories may stem from a quantitative algorithmic calculation, the names of the categories are often determined by specifically positioned human actors, in specific socio-cultural contexts (Kotliar 2020b). In this context, and as their demo for the Singaporean executive reveals, Extractive’s activity in East Asia depends on their clients’ satisfaction with the specific categories that they use—with the specific names Extractive chooses for their algorithmically induced categories.

Westernizing the other: The categories’ names

While Extractive’s algorithms can potentially produce endless categories, Extractive in fact offers their clients much fewer ones. As Jacob, another VP at Extractive, explained in an interview:

If we aim to sell our technology, we can’t really have a million tiny groups, we have to generalize. And that’s what we do. We’ve developed 30 groups. We divided the entire world into 30 groups. This [categorization] can help news companies in the West personalize their contents, or give banks in developing countries an ability to assess their customers’ creditworthiness when they don’t have enough information about them.

Jacob’s words reveal another aspect of Extractive’s expansionary logic. While they present their algorithms as culture agnostic, they also believe that the same set of thirty categories can describe people in “the entire world.” That is, Extractive translates their algorithmic outputs into merely thirty lingual categories, which, according to Jacob, are entirely universal: the same set of categories can be used by companies in the West, as well as by companies in the East or South (in what he refers to as “developing countries”). That is, Extractive sees itself as capable of conceptualizing people from any culture using only thirty named categories. This epistemological view is another means of overlooking culture—not only in the collection of data and its algorithmic interpretation, but also in its analysis and presentation—in how they name their algorithmic categories. In other words, the names Extractive choose for their algorithmic categories are merely another aspect of the company’s universalist, expansionary vision.

However, a closer examination of their categories tells a different story. The categories Extractive offers to their clients include categories like “techie,” “hipster,” “alternative,” “globetrotter,” “digital socializer,” “artsy,” or “sporty.” The names are presented in English, and are never translated into other languages. That is, while their algorithms can allegedly disregard language and categorize people from any culture, race, or locality, the categories they eventually sort people into stem from a very culture-specific perspective—from a Westernized view of people that echoes a globalist, consumerist, techno-elitist ethos. This view also represents a very specific global class (that includes, for example, people who “globetrot”), as well as a very particular generation (that includes, for example, people who “digitally-socialize” and self-identify as “hipsters”). Notably, this categorization does not include more culture-specific categories nor does it include any racial, ethnic, or cultural categories. After all, as Noble and Roberts recently argued (2019), postracialism goes hand in hand with neoliberal, technolibertarian worldviews. Moreover, in this case, and in opposition to the colonial categorization, highlighting cultural or ethnic differences might be perceived as overly political, and hence, impinge the company’s global marketing efforts and put its global expansion at risk. And so, while Extractive allegedly creates culture-agnostic algorithms, the labels they use for their algorithmic categories stem from a very specific, almost ethnocentric worldview, and from an equally specific socio-cultural position. While Extractive’s employees highlight the culture-agnosticism of their algorithms, the categories they actually use make it seem like they are, in fact, culture blind—ignorant to the fact that different cultures may perceive themselves through different sets of categories or have radically different interpretations to the same categories.

These categories are also reminiscent of more traditional types of categorization that were highly popular in late twentieth-century marketing—namely, psychographic or lifestyle segmentation. Such segmentation gained prominence in the 1960s, offering to expand beyond demographic categories by categorizing consumers according to more detailed criteria such as lifestyle, values or psychographic tendencies (Vyncke 2002; Arvidsson 2004). However, like Extractive’s categories, these too are mainly Western-born categories, that were devised and used in specific North American environments. For example, one of the most prominent methods for lifestyle segmentation originated from the Stanford Research Institute (SRI). Developed throughout the 1960s and 1970s, it became known as the “Values and American Lifestyles” (VALS) project (Swanson 2013). This project drew upon Maslow’s “Hierarchy of Human Needs,” yielding the influential 1983 book by Arnold Mitchell, “The Nine American Lifestyles” (ibid.). That is, this methodology was created for, and was primarily used on American consumers, but, at the same time, it also relied on highly universalistic assumptions (as the reliance on Maslow’s hierarchy demonstrates). And so, from its onset, this categorization embodied the assumption that different cultures share the same “lifestyles,” and that people from different geographies can be categorized according to similar sets of categories.

And indeed, such a universalistic view, alongside the practice of categorizing the Other using the categories of the West, is a longstanding one. As Prasad wrote, colonial discourses often render the non-Western Other knowable by using Western categories and epistemologies, and in doing so, they actively “drew the non-West into the West” (Prasad 2009, p. 21). That is, categorizing the Other according to the categories of the Colonizers is a way of making Others knowable and, hence, governable. In this case, however, in their attempts to make their algorithms sell across the globe, this Israeli company chose to give their categories the most universal names they could think of. Choosing benign, consumerist categories that focus on people’s “interests” instead of on their allegedly-more-stable racial, national, or cultural characteristics makes this categorization seem neutral, a-political, and hence, more readily saleable. Thus, this set of names is but another strategy for disregarding cultural differences by focusing instead on inter-personal differences; it is another aspect of the expansionary logic that is coded into Extractive’s algorithmic gaze.

However, it is worth noting that in this case, the East Asian Other is being categorized by the algorithms of a company from the Middle East, using Western (albeit “global”) categories—a company that is located at the periphery of what Couldry and Mejias describe as the “Cloud Empire” (Couldry and Mejias 2019, p. 40), far from the center of data colonialism.Footnote 11 That is, rather than using the categories of their own culture, this Israeli company chose to see the Other through a globalizing perspective, believing that such a categorization would better help their expansion Eastward. As I show below, this globalizing view of the Other also hints at how Extractive wishes to see itself—it is part of their efforts to see themselves as “Western” and position themselves as central actors in the global tech scene.

That is, in Extractive’s case, the algorithmic view of the Other includes the attempt to code against culture, as well as the specific globalized (and hence, easily expandable) names they pick for their algorithmic categories. But to understand algorithms, one has to examine the socio-technical assemblage that surrounds them (Kitchin 2017; Gillespie 2016; Seaver 2018; Kotliar 2020a)—not only how data gets analyzed or how the categories get named, but also the perceptions, worldviews, and narratives of the people who write and sell these algorithmic systems. In this case, examining the ways in which Extractive’s employees see their so-called “Eastern” partners, as well as their partners’ “Eastern” users, would better contextualize this algorithmic gaze.

The non-algorithmic gaze: Difference as a resource

In one of our interviews, I asked Dalia about their meetings with their partners in East Asia. This is when her narrative took an unexpected turn. She said:

The business culture there [in East Asia] is totally unpretentious—they are open to new things, and they’re essentially leapfrogging now. But until recently, they had nothing. While Europe and the US have [evolved] very gradually, they have these huge gaps. They totally lack local expertise, they don’t have universities, nor technological institutions. Nothing. But they do have A LOT of money, because they have lots of customers. For example, we’ve been talking to this type of Amazon in China, […] and it’s just … it’s a crazy country, […] an outdated country. There are no computers, no devices, there’s just nothing there. You can … you can send a cat there, and if you’re connected to the right people, and everything’s organized, you can sell them analytics, shmanalytics, whatever you want. As long as it works, no one will ask questions. But you have to be very well connected: they have a tradition in China—they’re very money-oriented and very control-oriented.

While Extractive’s algorithms supposedly overlook culture, and while their categories represent a globalizing view of the Other, this quotation shows that culture does play a significant role in how Dalia and her company see their Others. In this sweeping Orientalist generalization, Dalia describes the Asian economies in which they operate, and essentially, East Asia as a whole, as outdated, backward, and “crazy.” For her, the Asian Others lack expertise, universities, and technologies, and while they may be rich (and hence, offer lucrative funding for her startup), their wealth is explained as a naturally occurring phenomenon—as something that stems from the size of East Asian populations, not by their technological or entrepreneurial merits. That is, while Extractive’s algorithms can allegedly disregard culture, locality, or race, and while the company’s executives seem to believe that people from “the entire world” can be accurately described using universal, culture-agnostic categories, in this interview Dalia demonstrates a highly hierarchical, even evolutionary view of culture—a racist narrative that highlights differences rather than disregards them. In this view, “the East” is contrasted with “Europe and the US” only to present the former as weaker, more backward, and more ignorant than the latter. The “Asians” themselves are similarly perceived as gullible (“you can sell them shmanalytics”), and as potentially corrupt (“you have to be very well connected” to work with them).

Dalia’s racist remarks express what Frenkel described as the “colonial imagination”—a tendency “to take colonial hierarchies and distinctions for granted” (Frenkel 2014, p. 36). That is, while the algorithmic gaze could potentially offer a post-hegemonic, post-structural, and perhaps even postcolonial way of seeing, Dalia sees the world through the old colonial lens—through a language of essentialist distinctions and differences that can, in a complete reversal of the real power structure she operates in, depict giant corporations (from “the East”) as weak subalterns. This techno-orientalist discourse (Roh et al. 2015, p. 2; See also Marchart 1998; Morley and Robins 1995) is also characterized by the completely unrealistic depiction of the Asian Other in hypo-technological terms (“There are no computers, no devices, there’s just nothing there”), while simultaneously imagining her Asian partners as collusive—ones she can work with. After all, for Dalia, the combination between these companies’ imagined backwardness, weakness, and wealth merely highlights their business potential.

That is, Dalia’s Orientalist remarks are more than an expression of her personal worldview. It is part of Extractive’s business plan; it is how they aim to expand the reach of their algorithmic gaze Eastwards. And so, if the expansionary logic of Extractive’s algorithms is based on their agnosticism—on their alleged ability to overlook language or culture, Extractive’s business expansionism stems from a completely different worldview—one that sees cultural difference as an important and even crucial resource for expansion.

A similar perspective can also be found in Extractive’s view of its partners’ users. In the words of Doron, Extractive’s chief data scientist:

Around there [in East Asia], the problem of privacy doesn’t exist. There’s just no such thing. Nada! And people there are also very open to spam, but they don’t call it that, they call it fun! They also love deals and they love notifications—people actually want it! You know, if you or I started living on coupons like they do there, then maybe we could have saved some money, but we wouldn’t bother, would we? But they … they have this psychological, cultural background. It’s just … that’s the kind of people that they are—uncomplicated. They use [social media] all the time, […] and telcos [telecommunication companies] are everything for them. Telcos are their fathers and mothers, it’s the establishment. They have communities that have no one to trust but their telcos.

In Doron’s eyes, “the East” is seen as the solution to what is, for such companies, “the problem of privacy.” This solution lies in cultural difference—In the perceived attributes of these imagined Asian Others, and primarily, in their alleged disregard for privacy. Doron contrasts the “uncomplicated” behavior of his Others with his own behavior and explains that these differences are hardwired—they stem from their “psychology and culture.” Hence, highlighting the Others’ alleged dependency on coupons, notifications, and deals serves to highlight and reify the cultural differences between the Eastern Others and himself and, at the same time, to justify the vast data collection that is at the base of his company’s technology.

Much like the previous quotation, what we see here is an extremely differentiating view of the Other, an extremely essentializing view of culture and, in fact, the age-old Orientalist gaze: Like Dalia’s racist perspective, Doron sees “Asian users” as simple, uncomplicated, gullible users who childishly cling to their telecommunication companies (“they have no one to trust but their telcos”).These users’ alleged naïveté is also depicted as fertile grounds for the surveillance-based profiling that Extractive enables. And so, this “Asian culture” is seen as an excellent basis for what companies like Extractive do—extracting, analyzing, and selling user data. And “The East” is once again seen through an essentializing, racist gaze—as an Other that can easily be exploited. Ironically, Doron fails to recognize the discrepancy between his essentializing gaze and the culture-agnostic gaze of his algorithms, which, as we have seen, favors inter-personal differences over inter-cultural ones. In this case, Otherness and cultural difference are seen as crucial resources that can help this small Israeli company expand into East Asian markets and potentially turn this startup into an emerging technological power.

But how can we explain this discrepancy? And how can we explain the fact that it is not visible to Extractive’s executives? As Frenkel argued, peripheries often deal with the immanent paradox of trying to join the economic center while remaining on the geopolitical periphery (Frenkel 2014). Lyan and Frenkel similarly argued that peripheral companies attempt to rid themselves of peripheral stigma by attributing it to other companies from other peripheries (Lyan and Frenkel 2020 ). In this case, Extractive’s Orientalism attributes the stigmas of developing economies to their East Asian partners in an attempt to present themselves as a developed, Western actor. That is, while Extractive’s efforts focus on algorithmically defining the Other, their actions also point at their own attempts to discursively self-identify with the eternally-elusive category of “the West.” After all, as Said famously wrote: Orientalism “helps define ... the West in [the Orient’s] contrasting image” (Said 1995, pp. 1–2). And so, in this case, in describing their business partners in East Asia as gullible, exploitable, and technologically backward, Extractive is, in fact, drawing the line between those who see and those who are being seen—those who surveille and those who are being surveilled, and are making a considerable effort to stay in what they see as the right side of that line.

Thus, this essentializing gaze on the Other and the company’s agnostic algorithmic gaze are two contrasting-but-interrelated ways of imagining the Other, and together with Extractive’s way of naming their categories, these three perspectives underscore the expansionary logic of this type of data colonialism. That is, with this trifurcated, multi-focal gaze, Extractive can imagine the potential of their technology as limitless, and their algorithmic power as one that can quickly push the “data frontiers” (Beer 2018) further into “the East.”

Conclusion

Algorithms offer new ways of seeing (Kitchin 2014), and their ability to profile, categorize, and tangibly affect people plays a crucial role in their global expansion. But to better our understanding of this expansion, and to provide a more nuanced understanding of the algorithmic gaze, we need to see it as a multi-focal one—as a gaze that stems from a complex combination of very different types of lenses: The ways in which algorithms are being programmed to see the Other, the ways algorithmic categories are being named to depict the Other, and the ways in which the people who design such algorithms describe and understand the Other are all different but deeply interrelated factors in how algorithms “see.” In the case before us, the attempt to “code against culture,” the attempts to create a unifying global categorization, and the seemingly paradoxical attempts to capitalize from cultural difference all play a role in how the algorithmic gaze moves Eastwards, these are all factors in the global expansion of “data colonialism.”

Thus, this algorithmic gaze simultaneously seems like a continuation of the colonial gaze and its exact opposite. It is a gaze that disregards culture, but at the same time highlights cultural differences; a gaze that can generate countless hyper-individuated identities but can also categorize people into only thirty supposedly-universal categories. It is a gaze that can see without groups but that also deeply relies on the “groupist way of thought” (Brubaker 2004); a gaze that potentially offers new, softer, post-hegemonic ways of seeing (Cheney-Lippold 2011; Lash 2007), but that falls back into and is based on “traditional” racist, orientalist worldviews. These opposing-but-complimentary perspectives work together to create Extractive’s expansionary vision, to pave Extractive’s way into Other territories, or at the very least, to help them secure their place on the right side of the “big data divide” (Andrejevic 2014).

After all, Extractive’s view of the Other also plays a role in how this company sees itself: through their “data orientalist gaze,” as well as through their pretention to “code against culture,” Extractive reimagines itself as Western—a category that signifies to them the possibility of being a developed and potentially rising technological power. That is, through their multi-focal view of the Other, this company is creating its own “imaginary cartography of the internet” (Marchart 1998) as a territory that is divided between data collectors and data subalterns, those who see and those who are being seen, while technologically and discursively placing themselves on the “right” side of that map. In other words, while this company attempts to algorithmically conceptualize the Other, they, in fact, try to escape their own Otherness by joining those who see.

The case of Extractive, and particularly, their peripheral location, also points to the fact that the flow of algorithmic power is far from unidirectional. As Takhteyev wrote: “Being peripheral often means being connected tentatively and being dependent on resources that can be withdrawn” (Takhteyev 2012, p. 24). In this case, the emergence of this algorithmic gaze can be traced back to Extractive’s dependence on bigger global forces, in the East as well as in the West: their East Asian partners, their access to Facebook data, their access to users’ phones through DPI, and more. That is, in this case, the algorithmic gaze is simultaneously dependent upon factors in Israel, Singapore, the Philippines, China, and the United States. Accordingly, the power of these algorithms does not reside solely in one of these locations, but it flows back and forth—between the East and the West, the South and the North.

Thus, this article follows Milan and Treré’s call to move past the universalist view of datafication (Milan and Treré 2019), but it also challenges the presumption that algorithmic power flows from its centers in the West out to the global peripheries. In fact, as I have shown above, the algorithmic gaze is dependent upon different actors across different geographic locations, as well as on the dynamic interactions between them. Similarly, while recent academic and popular discussions on algorithms and their power overwhelmingly focus on Euro-American companies, and particularly on American platforms, we should keep in mind the multi-directional flow of such powers, as well as the multi-cultural relations that sustain them.

Moreover, while recent research has insightfully shown how algorithms reinforce racism and mirror racial biases (Noble 2018; Benjamin 2019; Sweeney 2013), this article also shows that the discussion around algorithms and race can go beyond questions of biased algorithms—racism can be found in different forms and in different stages of algorithmic production. Similarly, while research on algorithmic racism tends to focus on race relations in the United States, this article serves as a reminder that the spread of data colonialism across the globe opens new and important avenues for research on the ties between algorithms, race, and culture.

This article contributes to the critical study of algorithms by highlighting the complex, multifocal structure of the algorithmic gaze, and the ways in which its different lenses, constituted by various epistemologies, worldviews, and ideologies afford the global expansion of algorithmic powers. This article similarly highlights how algorithms move across space, and the importance of identifying the specific intercultural contexts in which algorithms operate. Extractive’s case study also contributes to our understanding of contemporary processes of Othering and of the socio-technical mechanisms behind contemporary constructions of the Other.

Future research should consider not only the ways in which algorithms are influenced by specific socio-cultural contexts (Seaver 2015; Christin 2018), but also how they are affected by intercultural encounters and by intercontinental flows—of data, of ideas, of manpower, and of funds. Moreover, when examining tech companies’ attempts to point their gaze at far-away populations, and to localize their platforms or contents based on this gaze (Mohan and Punathambekar 2018), we must see this as a complex, multifocal gaze that includes various and often seemingly contradictory epistemologies, worldviews, and discourses. Identifying these lenses, and how they are combined, could help form a better understanding of the power of algorithms, of their global expansion, and of the particular ways in which algorithms see.