Drawing the Algorithmic Self: On Our Becoming Statistical

Author: Curtis Roth

Illustrating a radically new model of collective identity formed by the platform economy, researcher Curtis Roth examines how the notion of the self is evolving in the age of surveillance capitalism.

Drawing has perhaps always been associated with some form of self-realization—where turning intellection into graphic marks shares its English etymological origin with drawing water from a well. Which is only to say that if drawing was first imagined as the production of a self, it’s a self as archaic as well water; one defined by enduring interior essences drawn-out onto delicate surfaces.

But if the self isn’t what it used to be, neither is drawing the simple dragging of ideas into marks. The contemporary drawing is less of an interior essence made manifest, than it is a public arena for the negotiation of selves construed within a marketplace of power relations.

Today’s cloud-based Building Information Management (BIM) platforms produce selves through the global allocation of permissions—modulating the agencies of designers, consultants, and laborers in real-time. However, if the model of the self currently being fabricated within networked drawing platforms substitutes the stability of interior essences with the stability of an external market, elsewhere in the most proprietary corners of the internet, an entirely alien self is being stabilized through its industrial isolation.

What follows are the first three exploits of the rendering engine, a networked drawing device, created by Curtis Roth in 2019, to chase this evolving fabrication of the self after the internet. The engine is one, in a collection of custom fabrication instruments operating as interfaces between physical objects and online laborers. These instruments posit contemporary design as a managerial process in which objects are produced through the uneven allocation of agency across vast geographies of production. These interfaces rely on an expanding catalog of profiling techniques; proxies through which our digital selves are not only made to appear, but are increasingly transformed into targets for the distribution of various forms of freedoms, surveillance, and violence. Against these regimes of control, each instrument stages alternative forms of collaboration and collectivity.

The Rendering Engine, Columbus, United States, 2019

The rendering engine came online in early May of 2019 and consists of the repurposed components of a CheXpress CX30 inkjet check printer mounted to a three-axis motorized rack. The engine is located in a small warehouse on the outskirts of Columbus, Ohio, in the American Midwest. Much like contemporary cloud-based drawing platforms, the engine is a public arena in which the power-relationships between various actors are managed in order to coordinate graphic compositions. Posted as a task on Amazon’s micro-labor platform Mechanical Turk, it appears online via an Apache web server broadcasting the engine’s activities in real time to thousands of anonymous digital workers. But unlike its BIM equivalents, the engine provides these workers access not by sorting their expertise, but by construing them through a machine fingerprint.

First developed by state-sponsored researchers in order to identify web users surfing the internet on anonymous onion routers, and later adopted by corporations interested in circumventing the increasingly strict legislation of browser cookies, the machine fingerprint is a way to passively identify users by stringing together parameters such as screen resolution, default time zone, system fonts, graphics card specifications, and installed plugins into a unique portrait of an individual user. Each time an anonymous user activates the rendering engine, the printhead’s XYZ coordination, speed, and ink color are calibrated by the user’s unique fingerprint, and a non-repeating string of numbers is printed at 96 dpi.

As the machine is activated thousands of times by thousands of discrete machine signatures over the course of successive days, these unique strings bleed into compositional fields weighted by statistically significant identifiers shared between workers. The rendering engine reimagines fabrication hardware as a form of social media, an online public space for collective encounters that document the reconfiguration of the self in an age of surveillance capitalism—a reconfiguration first intimated at the Serpentine Gallery’s 89plus marathon over half a decade ago.


From Mass-Indie to Industrial Isolation

Youth Mode: A Report on Freedom was delivered by the trend forecasting group K-Hole at the Serpentine on October 19, 2013. A cultural manifesto presented like a market report, Youth Mode was a post-critical treatise on the self after the first decade of the twentieth century. While it’s remembered today primarily as the mom jeans manifesto responsible for introducing the neologism Normcore, Youth Mode’s appearance in 2013 was symptomatic of the growing sense that the techniques of contemporary life through which we have long-endeavored to design ourselves suddenly seem strikingly impotent. But in the years to follow, this growing unease would lead less to the acid washed jean jackets atop evening dresses first forecasted by K-Hole, than it would to the slow realization that today the construction of ourselves has already been outsourced elsewhere.

K-Hole’s analysis of contemporary identity construction relied on an incisive genealogy of cultural formations, which opposed the alternative culture of the 1990s with a subsequent period described by the group as Mass Indie. Where 90s alternative was characterized by an evasion of sameness, opposing a monolithic mainstream with an equally monolithic counter-culture, Mass Indie was defined by a post-critical logic that celebrates difference without oppositional antagonism. According to K-Hole, in Mass Indie times, there is no norm to rebel against, only aggregated differences, and one is compelled to mobilize the diverse spheres of the self as arenas for infinitely additive acts of self-differentiation.

Importantly, this reframing of self-design, from the oppositional culture wars of the 90s to the endless cultivation of differences marking Mass Indie explicitly refashions the self as a market commodity—identical to the market-based identities managed by contemporary cloud-based drawing platforms. In Mass Indie times, one no longer considers the design of the self as a means of actualizing fragile authenticities suffocated by mass culture, but rather, as a strategy for cultivating their own scarcity within a marketplace of similarly differentiating selves. Browsing Revit libraries or Tinder for prefabs or lovers ultimately mobilizes similar faculties of differentiation in order to discern desirable outcomes from increasingly crowded fields of content.

But in diagnosing this millennial cultural turn, K-Hole also stressed the impending collapse of this accelerating market of the self, arguing that the cultural markers of difference necessary for the production of our own scarcities are themselves exhaustible resources. Mass Indie entails a logic in which each act of differentiation both exhausts the future market value of that differentiator while driving each of us toward evermore obscure forms of isolation. As the competition for individuation becomes ever more intense, the mining of increasingly scarce social capital becomes all the more cutthroat. Against the exhausting isolation of Mass-Indie, K-Hole posited the utopian swerve of Normcore: an attempt to find freedom through a rediscovery of our non-exclusive selves.

Seen more than half a decade in retrospect, K-Hole’s desire to achieve meaningful connections by opting in to shared values blurs with the pre-bailout optimism of Barack Obama’s first term, Hong Kong’s Umbrella Movement, and the spirit of Occupy Wall Street. These collectively performed political acts of solidarity were motivated, in part, by a growing exhaustion with the labor of maintaining the devaluing markers of difference that had once seemed to imbue our constructed selves with uniqueness. But ultimately Youth Mode proved to be less of a forecast than a eulogy for a closing window of solidarity intimated by 2013’s second report on freedom.

“In the absence of specific information regarding whether a target is a US person, a person reasonably believed to be located outside the US or whose location is not known will be presumed to be a non-US person unless such person can be positively identified as a US person, or the nature of circumstances of the person’s communications give rise to a reasonable belief that such a person is a US person.”

Attorney General Eric Holder, NSA FAA Minimization Procedures A, 2009

In response to former US Attorney General Eric Holder’s notion of postponed state citizenship, contained within the trove of documents leaked by Edward Snowden in the summer of 2013, theorist John Chenney-Lippold has offered the neologism jus algoritmi. For Chenney-Lippold, where citizenship has historically been conferred by the state, either due to parental ancestry (jus sanguinis) or the site of one’s birth (jus soli), the post-9/11 surveillance apparatus presided over by Holder reframed citizenship as a perpetually deferred status, continually reassessed based on one’s internet identifiable distinctions. According to the model of jus algoritmi, one possesses the rights of a US citizen so long as their online activities meet the predetermined characteristics of a citizen, such as speaking in English or communicating with other known citizens. This status is forfeited by the subject in question once their activities deviate from this ready-made model. Chenney-Lippold’s theorization of jus algoritmi documents an often overlooked aspect of the Snowden revelations: aside from the egregious violations of human rights contained within, Snowden’s disclosures suggest that such violations were made possible by a radical reconceptualization of the self. While our total adoption of social media self-construction compelled each of us toward the peak difference of a Mass Indie age, an almost unimaginably vast catalog of differentiators was being coined in the data centers of telecom corporations and the US Defense Department alike.

While K-Hole anticipated the exhaustion of our ability to convincingly differentiate ourselves from one another, perhaps what was being exhausted was merely our capacity to believe that we possessed meaningful agency over these processes. Your ability to realize any particularly desirable identity through your own conscious acts of distinction are virtually meaningless in comparison to the colossal catalog of mouse movements, speech patterns, search histories, and credit ratings through which your freedoms are now conferred or withheld. K-Hole was correct in imagining these processes of self-actualization as the refinement of commodities, but altogether too optimistic in imagining that we were all the executives of ourselves, rather than the market’s raw material. In short, our being still relies on our relentless isolation, only now we’ve learned that this isolation is being manufactured elsewhere.

The first image series produced by the rendering engine resulted in four images drawn by 1,928 unique workers over the course of 650 hours. Each composition is the product of a specific pairing of the XY and Z axes, print-head speed and ink color of the rendering engine with the default language, time zone, IP address location, screen resolution, and font count of each anonymous worker. The compositional qualities, in the form of discrete color fields or ink densities, are the index of the statistical similarities between workers as discreetly measurable types. Each composition is at once a graphic composition on the page and the managerial composing of hundreds of laborers into machine readable categories.

Drawing 01.01 (180 hours, 510 workers) pairing y-axis movement with browser type, x-axis movement with font count, ink color with IP address location, and speed with the worker’s default language.

Drawing 01.02 (155 hours, 468 workers) pairing y-axis movement with screen resolution, x-axis movement with default language, ink color with IP address location, and speed with the worker’s default language.

Drawing 01.03 (195 hours, 325) pairing y-axis movement with default language, x-axis movement with screen resolution, ink color with browser type, and speed with the worker’s font count.

Drawing 01.04 (120 hours, 625 workers) pairing y-axis movement with IP address location, x-axis movement with default language, ink color with screen resolution, and speed with the worker’s font count.


Bayesian Proxies

Returning to the US Department of Justice’s jus algoritmi, it’s crucial innovation wasn’t only that something of such urgent importance as one’s citizenship status could be determined by an abstract algorithm rather than birthplace or heritage, but rather, that the immutability of conventional citizenship could be reconfigured into an always updatable forecast pending a subject’s future activities. Holder’s deferred citizenship model relies on what is known as Bayesian statistical analysis, an innovation in statistical thinking by Thomas Bayes which was posthumously published in his 1763 work An Essay Toward Solving a Problem in the Doctrine of Chance.

Bayes’ theorem argued that the probability of an assertion being correct increases based on prior knowledge of conditions that might be related to that assertion. By way of an example, one only has to look at their own Google Ad Personalization tab for a Bayesian mirror of themselves. Chances are good that when you opened your very first Chrome browser tab, Google’s ad profiling algorithm knew little more than your name, age, and perhaps your gender identity. Today, your profile likely contains hundreds of interests, from architecture to heavy metal or television comedies, each facilitating a more targeted ad experience. And like jus algoritmi, this model of you is always deferred. If you decide to lose weight, after a few Google searches your profile might be expanded to include an interest in gym memberships or weight lifting routines.

But crucially, a Bayesian model in the era of information, whether in the form of algorithmic citizenship or personalized advertisements, is not just a technique for deriving a more accurate profile based on past performances, but a method of anticipating a profile of you to come. Such anticipatory forecasts are inferred from the millions of other lives encountered by the same algorithm simultaneously. For example, through their Similar Audiences feature, Google might anticipate that your recent interest in gym memberships will soon be followed by a desire for a slimmer wardrobe or an enthusiasm for online dating. Clothing and romance will be preemptively added to your advertising profile in anticipation of the subject that you’re suddenly more likely to become. Unnervingly, our Bayesian models depict us less like the relatively stable subjects perceived by others, and more like we might prefer to see ourselves: as disparate aggregates of faculties, histories, and impulses evolving toward an always anticipated self to come.

The difference between those numerous selves maintained by proprietary Bayesian algorithms, and K-Hole’s model of the self as a financial asset, is that while relentless differentiation leads to exhaustion in the latter model, it merely increases the efficiency of the system in the former. The infinitesimal uniqueness of my Bayesian self actively contributes to the evolving definitions of all other selves in real-time. Under this model, you are no longer a financial asset, but a renewable resource for the production of infinite differences. Where K-Hole optimistically anticipated a future collectivity born out of our exhausted exclusivities, Bayesian analytical enterprises, such as the internet advertising giant Quantcast, process one trillion discrete media consumption events per month; our increasing isolation is now not just irrelevant, it’s the point. No longer living in Mass Indie times, ours is the age of loneliness; not because we feel loneliness today any more acutely than we used to, but rather because the models of ourselves manufactured by the corporate surveillance economy attain value precisely through their industrialized isolation.

The second image series produced by the rendering engine resulted in three images drawn by 1,520 unique workers over the course of 504 hours. Each composition deviates from a pre-determined graphic figure consisting of (1) a straight horizontal line, (2) a sequential grid, and (3) four perfect circles. Each of these figures would have been realized had the browser fingerprints of the actual workers participating in the images’ production matched a Bayesian forecast of the workers, predicted through a machine analysis of the workers from Series 01. The three forecasts were produced by Oracle’s Crystal Ball, an analytics forecasting software that relies on the Monte Carlo Method, the same stochastic technique used to produce realistic light behaviors in contemporary computer renderings. Each image’s compositional qualities emerge through the unlikeness of the workers to their own simulated models.

Drawing 02.01 (168 hours, 1080 workers) each worker draws a unique vertical line based on their default language, IP address location, screen resolution and browser type. The length of the line is determined by Oracle Crystal Ball’s average anticipated fingerprint of the worker. Length to the vertical line is added based on the worker’s deviation from this forecast.

Drawing 02.02 (168 hours, 90 workers) each worker draws a vertical or horizontal line, depending on whether they log-on to the machine before or after Oracle’s Crystal Ball anticipated their arrival, resulting in unexpected overlaps with the under-painted layer.

Drawing 02.03 (168 hours, 350 workers) each worker is polled on whether or not they possess any of the four rare fonts depicted in pink. The worker contributes a lump to each of the quadrants that represent a font they do not have. The geometry of that lump deviates from perfect circle based on the ways in which default language and IP address location differ from Oracle Crystal Ball’s forecast.


Our Becoming Statistical

This age of loneliness, experienced vicariously by our own circulating digital models is being engineered by corporations like the previously mentioned Quantcast, the San Francisco-based business founded, “to manufacture data to make advertising more relevant for consumers.” Today, Quantcast intimately models nearly every internet user on planet earth via Quantserve, a ubiquitous suite of tracking cookies gathering up to 30 petabytes of browsing information per day. The average US based web-user updates their own Quantserve profile with new information 600 hundred times per month. Increasingly, corporations like Quantcast are interested in bridging online and offline data mining endeavors. In 2015, they launched Audience Grid, a data-sharing partnership between TiVo and Oracle’s Datalogix that brought together web-browsing, television consumption, and offline purchasing histories respectively into stacked consumer profiles.

In a conventional sense, these digital profiles can be understood as avenues for the transmission of biopolitical power. That is, as specific categories of life that serve as the targets for the distribution of various resources, forms of attention, and violence. And indeed, companies like Quantcast are quick to point out that despite Quantserve’s ability to model individual web users in almost unimaginable intricacy, Quantcast, “does not intentionally collect any personally identifiable information—that is, information that could be used to uniquely identify or locate an individual.” Rather, these systems are oriented toward biopolitical models once removed from the subject herself. Services such as Quantcast’s Lookalikes, for example, allows businesses to develop sophisticated behavioral profiles of their ideal clients so that Quantcast can identify and advertise to users sharing similarities with those models.

But while sharing conventional biopolitical regimes’ circumscription of the individual within manageable types, today’s contemporary Bayesian classes also point to a profound reconfiguration of the eighteenth-century regimes first described by Michel Foucault. Where these earlier models obliterated the specificity of the individual with discursive categories such as gender, which comport unique subjects to broad classes, the Bayesian subject’s deviation from type is where their value as a resource is derived. Recalling a previous example of a web user who searched for weight loss strategies, one would hardly need an apparatus as sophisticated as Quantserve to anticipate their future searches for new jeans or new romantic companions. Rather, such systems gain intelligence precisely when subjects deviate from rational inference. Imagine an interest in weight loss routines being followed by an interest in nineteenth century model boats, color-changing fingernail polish, or lawn-care enthusiast forums. Such seemingly irrational inferences are fed back into the measurable type’s emergent definition and stored until some kind of future value can be derived. In the age of loneliness, our distinctions are mined at exponentially increasing resolutions. No longer promising us exclusivity, today these distinctions are recursively fed back into a vast biopolitical apparatus that not only approaches the modeling of all life at 1:1 resolution, but offers us an increasingly profitable portrait of all life to come.

The third image series produced by the rendering engine resulted in three images drawn by 1,255 unique workers over the course of 504 hours. Each composition consists of two layers, an underpainting produced by myself depicting (1) an arrangement of voids, (2) an instruction, and (3) a bar graph. Each of the underpaintings describes to the worker the image that would have been realized had their own browser fingerprints matched a Bayesian forecast of themselves defined through a machine analysis of the workers from Series 01 and 02. Each of the image’s compositional qualities emerge through the worker’s willingness to disrupt the rendering engine by redesigning themselves in real-time through the modification of their own browser fingerprint. Graphic deviations accumulate by irrationally shifting between browsers, camping on the server to prevent competing traffic, or logging-in to the rendering engine through the use of virtual private networks (VPNs) that modify a worker’s readable IP address.

Drawing 03.01 (168 hours, 648 workers) parcels workers into particular cells depending on 9 measurable qualities. Each worker’s unique qualities is absorbed into a highly saturated black. Workers can print unlimited passes. Individual workers who remained logged onto the machine for an extended period of time would register their presence in the form of colorful overspray.

Drawing 03.02 (168 hours, 196 workers) workers in UTC +2 using the browser Firefox could activate the printer in order to print one vertical green line. All other workers logging in to the engine would gradually de-calibrate its resolution, causing lines to appear less as discrete figures and instead, as fields.

Drawing 03.03 (168 hours, 411 workers) a graph, wherein the x-axis corresponds to browser type by hour of the day, and the y-axis corresponds to the day of the week. The pink underpainting indicates the number of workers using a particular browser anticipated by Oracle’s Crystal Ball forecast. Workers could print a new line every two minutes as long as they remain logged on to the engine.


Self-Design in the Age of Loneliness

In his well-known treatise Ornament and Crime, Adolf Loos infamously claimed that “the modern person who tattoos himself is either a criminal or a degenerate.” For Loos, the surface, whether of a pair of shoes or a human being, demanded reduction; an erasure of aesthetic artifice so as to reveal the essences presumed to lie just beneath. Elsewhere, theorist Boris Groys has observed that not only was Loos’ aversion to tattoos symptomatic of modern design’s ambition to liberate the essences beneath encultured surfaces, but argues that in the twentieth century, the liberation of an authentic self is the apotheosis of this ambition. Living with an obligation to self-design that stretches from Nietzsche’s Übermensch to K-hole’s Mass Indie, one continually re-fashions their various manifestations so as to present “him or herself to the gaze of the Other as an honest, plain, unornamented, ‘un-designed’ object.”

But today the self is neither the stable interior essence implied by drawing’s etymological origin or Loos’ aversion to decoration, nor is it the market-stabilized self, managed by our Mass-Indie aesthetic choices or our networked drafting platforms. Our ever-multiplying selves are now constructed as much through our curated tastes as they are through obscure algorithms reprocessing our rights of citizenship from a Nevada desert. We are both a citizen and not, an actualized self and a statistical data point allowing Oracle to more accurately forecast the next anonymous worker to log on to an ink-jet printer in a Columbus suburb. Self-design can no longer be understood as the obligation to wrest essences out from under appearances, but rather, as the exploitation of a vast catalog of selves now constituted by design briefs that are utterly indifferent to one another.

During the 34th hour of the production of drawing 03.03 an anonymous Mechanical Turk worker with the Worker ID A1V2H0UF94ATWY logged on to the rendering engine through a Chrome browser tab from an IP address located in New York and printed a small line segment in red ink. Two minutes later, an anonymous worker with an identical Worker ID logged on to the engine through a Safari browser tab from an IP address located in Amsterdam and printed another small line segment in red ink on an alternative portion of the page. Two minutes after that, an anonymous worker with an identical Worker ID logged on to the engine through a Firefox browser tab from an IP address located in San Diego and printed a third small line segment in red ink on a third portion of the page. Over the next hour, Worker ID A1V2H0UF94ATWY oscillated between a Chrome, Safari and Firefox user working in New York, Amsterdam, and San Diego, preventing other users from accessing the engine, and resulting in three abnormally long vertical lines that far exceeded Oracle’s forecast for the 34th hour of the drawing.

Over the course of this 34th hour, A1V2H0UF94ATWY was drawn-out into a heterogeneous array of 'both/and' subjects that never cohered: they were an individual Amazon worker and a trans-national collaborative, a Chrome and a Firefox user, a resident of San Diego and Amsterdam, an Oracle model and its error, a worker and an author at once. Suggesting that perhaps in our age of industrialized loneliness, the closest thing to freedom might be found in accelerating the isolation of our circulating profiles. In fabricating selves that can never be profitably folded back into the sense-making machinery of surveillance capitalism. In learning to live incongruent fragments of lives at once, all while hoping that this age of loneliness is more fragile than it seems.

The 'both/and' subjects performed by A1V2H0UF94ATWY between 33:05:04 and 33:53:47

Curtis Roth

An assistant professor at the Knowlton School of Architecture at The Ohio State University. His research consists of images, objects, and texts that examine the relationship between computation, subjectivity, and distance as they pertain to questions of authorship and labor. His work has been published in e-flux, Thresholds, and Volume, and has been exhibited at the ZKM Karlsruhe, the 2014 Venice Biennale, and elsewhere.

If you noticed a typo or mistake, highlight it and send to us by pressing Ctrl+Enter.