“The Scientist of the Holy Ghost”: Sunspring and Reading Nonsense
Sunspring (Oscar Sharp, 2016) is the first film written entirely by an artificial intelligence program. Like many experimental films, Sunspring is fascinating not in spite of but rather because of its incoherence. While the vast majority of scholars and journalists simply dismiss incoherent AI output as a sign of poor programming ability, I am interested in how incoherence functions within Sunspring for what it suggests about the relationship between the current reading methodologies in the humanities and how they intersect with the algorithmic technologies and culture that surround us.
Sunspring (Oscar Sharp, 2016) is the first film written entirely by an artificial intelligence (AI) program. Filmmaker Oscar Sharp and AI researcher Ross Goodwin created this nine-minute short for the Sci-Fi-London 48 Hour Film Challenge. Every year, this festival provides a few prompts, including some lines of dialogue and a couple of props, that producers must use to make a film in forty-eight hours. To write the script for Sun-spring, Sharp and Goodwin input dozens of science fiction film and television scripts into their AI program and told it to generate a script that mimicked them.1 These scripts varied widely and included, among many others, the entire run of The X-Files (Fox, 1993–2002, 2016, 2018) and Star [End Page 1]
Trek: The Next Generation (1987–1994) (see Figure 1). As explained through title cards at the beginning of the film, the AI technology that generated the script is roughly analogous to those common machine learning algorithms which try to guess the next word users will type based on words they have typed in the past. These algorithms are used on many sites across the web including Google and Facebook and smartphones employ them in their texting applications. The theory is that if the AI program thinks everything the user writes is a science fiction script, then it should be able to generate a science fiction script based on this history.
One might reasonably imagine that this process would result in a quintessentially average science fiction film. If genre is based on similarities between texts, then a computer program designed to create a science fiction film based solely on previous scripts should output something that plays up these similarities and exposes the clichés of the genre. Such a film should be able to reveal the constraints and connective tissue that hold science fiction together and perhaps also illustrate how mechanistic the genre has always been.
But that is not what the Sunspring script does at all. Instead, it is almost entirely incoherent, uncanny, surreal, and surprising. When the filmmakers showed it to colleagues, several expressed shock at its oddity. For instance, two of the characters share the same name, H, and in the middle of a scene, one of them coughs up a plastic eyeball and no one else seems to notice (see Figure 2). Based on this script, Sunspring—like many experimental films—is fascinating not in spite of but rather because of its incoherence, its non-sense.2 [End Page 2]
But first, what is there to say about a nonsensical text like the film Sunspring beyond simply pointing out that it is intriguingly weird or that it doesn’t appear to mean anything at all? Regardless of whether that incoherent text is a film, a novel, an essay, or the ramblings of Donald Trump, why write an essay on such a text as a text? One could certainly focus on analyzing the production culture that created Sunspring, the reception it received, and the discourses that surround it in order to examine what its cultural presence says about artificial intelligence more generally. While that could certainly be important and helpful work—and to some extent, I will do that—these methods cannot help one figure out how to read the film as a text, or even know if it is readable at all. With this quandary as my starting place, I am interested in Sunspring for how it might help us think through the relationship between incoherence and legibility—for how it might help us examine the assumptions, politics, and stakes of the way we read now.3
I begin by discussing current “postcritical” reading methods—from reparative to surface to distant readings—in the humanities and the field’s turn away from close reading. Here I am primarily interested in the general inability of current reading methods to, in practice, actually differentiate themselves from close reading itself. I consider the various institutional forces that have aligned to make critiques of methodology into profitable ventures and question why humanities scholars care about methodological precision at all. Don’t our best readings and analytical strategies typically transcend or just [End Page 3] ignore the limits of the methods we pretend to set for them? To call for an anti-methodology of reading, I turn to Sunspring, a film that helps us think through the problem of reading and respecting a text that makes no sense—or non-sense. It also illustrates the potentials and stakes of trying to read a text for its lack of sense. This is not simply a call for reading a text against the grain, but rather for addressing those instances where it is not at all clear which direction the grain of the text is going, or whether there is one at all.
As a piece of experimental media, Sunspring illustrates the actors’ and human filmmakers’ reasonable and rational attempts at reading and analyzing the predominantly nonsensical writing of an AI machine. In the process, they present the film as if the AI’s outputted script is meaningful and coherent even when it makes no sense—at least not human sense—at all. I argue that their reading strategies are exemplary of how we are expected to read and interpret the algorithmic output we receive every day via Google, Amazon, and Netflix among many sites and technologies. These sites and their advertising train us to minimize if not completely ignore all signs of otherness, alterity, or incoherence in their output.4
Indeed, there is a long history to the belief that algorithmic technologies obscure or entirely erase otherness. By otherness, I do not mean marginalized bodies (though algorithmic technologies do certainly obscure them as well), but rather those experiences that register as being utterly different to the degree that one cannot even be sure of what they are experiencing at all. Like many contemporary pundits, Jean Baudrillard feared that digital technologies and globalism make us insular, arrogant, and narcissistic; he felt that otherness was becoming a precious and scarce element.5 Otherness is not inherent in an object, but rather always a relation. Recognizing something as other from yourself requires that you look for its difference. We can simply look for what is familiar in a film like Sunspring, but in the process, we would miss out on what makes it truly interesting. In the same way, if we only see ourselves reflected back by digital media, that may be our fault, not the technology’s. Forcing texts to make sense when they do not is potentially unethical in that it reduces difference to sameness. Reading only for what makes sense obscures and devalues otherness of all sorts. In the process, when we only look for ourselves in the data that surrounds us, we too often dismiss and discard everything that doesn’t remind us of ourselves as meaningless excess or noise. The mere act of attempting to consider Sunspring as a text at all (rather than dismiss it as noise) allows for the possibility of an alternative ethics of algorithmic culture—one grounded in an attention to otherness rather than to one’s extant expectations and desires. [End Page 4]
THE POSTCRITICAL TURN
If the 1970s and 1980s are nostalgically remembered in academia for the many productive theoretical paradigms generated in those years, the 2000s and 2010s may one day be fondly recalled for broadening the methodologies of the humanities. The main difference may be that whereas the 1980s has (rightly or wrongly) been framed as a fight to find the most useful, most grand, or “best” theory, the fight over method, at least presently, has the much more modest aim of suggesting that close reading may not be the only legitimate method in the humanities.6 From Bruno Latour to Heather Love, many have questioned (if not fully attacked) close reading and critique more generally for being ultimately incapable of accomplishing what the field hoped it might: illuminating the deeper and hidden meanings of a text and fomenting political change in the process.7 Stephen Best and Sharon Marcus argue that close reading is unnecessary or unhelpful during a period when the failures and treacheries of governments and economies are so explicit.8 And following Rita Felski, Anne Anlin Cheng points out that close reading has too often encouraged a phallogocentric desire to master texts, a desire made explicit by anyone even implicitly claiming to know a text’s one true meaning.9 Latour has gone even further by arguing that while close reading is meant to explain oppression and free us from it, it has instead become a tool of the paranoid political right, who use it to argue away climate change and claim that “no plane ever crashed into the Pentagon.”10 As a whole, this scholarship focuses on the worst examples of close reading, framing it as a straw man and a tool only useful for a kind of gotcha analysis that expounds on the best or worst aspects of a text in order to dismiss everything else.
In response, an array of new humanities reading has sprung forth.11 Some of the more famous examples are Eve Kosofsky Sedgwick’s reparative reading, Best and Marcus’s surface reading, Franco Moretti’s distant reading, Leah Price’s not-reading, James Sosnoski’s hyper-reading, N. Katherine Hayles’s speed reading, D. A. Miller’s too-close reading, Love’s close-but-not-deep reading, and Ed Finn’s algorithmic reading.12 There is not enough [End Page 5] space in this article to explain the nuances of each of these methods and what makes them distinct. What interests me is that however different these methods may be in theory, they often end up being quite similar in practice. Indeed, the analysis that these methods produce often sound not just similar to one another but also to what close readings were supposed to be: a careful analysis of a text that is in some way helpful for understanding what the text does. Even the works in which these scholars explain their methods read from a similar script. They largely start by suggesting that in considering close reading the only worthy academic method for understanding a text, we have too often neglected or actively thumbed our noses at other, more common ways of reading and the types of knowledge they can create. In the process, they argue, we have also blinded ourselves to the politics of close reading itself, with a special focus on Fredric Jameson’s assumption that you cannot assume a “text means just what it says.”13 This assumption frames close reading as a primary tool for “discovering” the unconscious ideologies in a text and thereby performs the political act of explaining how oppression works so we might combat it. In making this argument, postcritical scholars paradoxically justify their new reading strategies with close, symptomatic readings of the close reading method.14 Often, these close readings adopt the same gotcha form that they critique.
The impulse to question and critique close reading is valuable, and the desire to find new methods for approaching problems that close reading fails to unravel is certainly important. Anything that might move us away from a desire to simply claim mastery over the meaning of a text must be celebrated. Yet this impulse is hardly the only reason that these alternative reading methods have become so popular among humanities scholars. They are worth critiquing as well, especially since their supporters, while excellent at articulating the politics of close reading, too often present their own new strategies simply as pragmatic responses to the problems of close reading. In the process, they neglect the political and institutional demands that have impelled them. There are many potential reasons for this upsurge of interest in method, reasons less motivated by scholarship itself than with its funding models.15 If anything, the drive toward methodological specificity illustrates [End Page 6] not only how messy and ill-defined method is in the humanities but also how productive this mess can be. In the humanities, we can afford to have a more intuitive relationship to method than those in the sciences and in engineering; we are free to pull data from various sources and analyze them in ways that show the nuanced contradictions at play.
Indeed, perhaps methodology is the problem. With its suggestion that to best approach a text, one must follow certain specific procedures and not others, methods may obscure more than they reveal. As Pauline Kael famously argued, “Criticism is an art, not a science, and a critic who follows rules will fail in one of his most important functions: perceiving what is original and important in new work and helping others to see.”16 Here Kael is skewering Andrew Sarris’s methodical approach to auteur theory, but she could just as easily be discussing any of the many, more recent textual analysis and close reading methods, from Rick Altman’s “A Semantic/Syntactic Approach to Film Genre” to the big data approaches popularized by Franco Moretti to those associated with the Arclight data mining project and many others.17
Kael argues that a reliance on ‘rigid’ theories or methods with “objective standards” of any kind does not make it easier for critics to perform their craft, but rather leads to a situation where critics are no longer necessary at all. She bemoans how this focus on theory and method led to Sight and Sound becoming “a good, dull, informative, well-written, safe magazine, the best film magazine in English, but it doesn’t satisfy desires for an excitement of the senses.”18 Such approaches can certainly help us to see something new in the films and the world around us, but Kael laments how often, they just show us what we already know, and, in so doing, also uphold an often harmful (not to mention terribly boring) status quo.
Instead, Kael believes “that we respond most and best to work in any art form (and to other experience as well) if we are pluralistic, flexible, relative in our judgements, if we are eclectic.”19 This is a call for critique sans method and even though she points out that “criticism is an art, not a science,” her essay anticipates an influential movement in the 1970s and 1980s in the sciences devoted to challenging methodological precision.20 For instance, in Against Method, Paul Feyerabend famously argues that obeying a method actually impairs scientific progress. He proclaims that those scientists who [End Page 7] have impacted our world the most through their experiments were methodological opportunists, using whatever moves or tricks of the trade were close at hand, whether they violated ideals of empiricism or not; “A complex medium containing surprising and unforeseen developments demands complex procedures and defies analysis on the basis of rules which have been set up in advance and without regard to the ever-changing conditions of history.”21 While much of the power of methods lies in their ability to be employed continually in many different contexts, Feyerabend argues that this quality is also what makes them ill-equipped for examining anything new or for discovering new information about anything old.
Along with Feyerabend, Sandra Harding and Helen Longino have long challenged the usefulness of strict adherence to scientific and social science methodologies from a feminist perspective.22 They particularly critique the central scientific methods for ignoring the experiences, knowledge, and problems of women and minorities and thereby encouraging findings that support patriarchal understandings of the world. Like Kael, they argue that method too often only makes it harder to see and explain what is new. Unlike Feyerabend, they do not want to throw away methodology altogether; rather, like Love and the various other humanities scholars I have discussed, they propose methods that they hope will lead to more equitable conclusions.
Within the context of humanities reading methods, the drive toward different reading methodologies has led to a profusion of oppositional terms, such as surface over depth, distant over close, and normative versus queer. This discourse recreates binaries that make it seem as if the new methods should at the very least be quite distinct and lead to different conclusions. They also imply that there is an objective way of figuring out whether your analysis is superficial or deep, queer or not. Yet these judgments are subjective and dependent on education, discipline, and familiarity with extant scholarship, among other factors. Such terms are more a value judgment than anything else and make it nigh impossible to tell whether one is following one reading method or another. If doing a surface reading requires that you only focus on the most obvious details and interpretations, how do you even begin to decide which ones those are? This hermeneutic conundrum leads to many more: When exactly is symbolism too obscure or the doubling of an entendre too obvious—and to whom? When is rupture or excess in a text actually the dominant meaning? Can something in plain sight “elude observation”?23 Why are these scholars going to such trouble to confuse or redefine surface as depth, distance as connection? Would it perhaps be more productive to abandon this drive toward methodological definition and specificity in favor of flexibility yoked to feminist, antiracist ethics and goals?
Even as many of these scholars begin by defining their methods, they also seem aware of just how exploratory and partial those methods are. First, they [End Page 8] routinely backpedal by suggesting that they aren’t really against close reading even as they attack it.24 Many of these authors including Felski, Latour, Sedgwick, Best, and Marcus describe their efforts not as a complete turning-away from close reading, but rather, as a speculative exercise in trying to imagine, as Sedgwick queries, “How are we to understand paranoia in such a way as to situate it as one kind of epistemological practice among other, alternative ones?”25 For all of these scholars, paranoia is intrinsically connected to the dominant close reading practices popularized by Jameson, which brings me to my second point: when explaining the binary they are creating, they also continually use terms in ways that have little in common with their vernacular usage. This is often due to these authors searching for alternative historical or scientific models to base their hermeneutic methods on. Sedgwick’s work on reparative reading is a great example of this tendency. As she sketches out a psychoanalytic analysis based more on the work of Melanie Klein and Silvan Tomkins than Sigmund Freud and Jacques Lacan, Sedgwick also rearticulates a variety of terms: reparative readings are depressive, and paranoid readings are hyper-active; neither appear to repair or destroy; both critique.26 Sedgwick worries over how overpowering and viral paranoia can be, but isn’t that also a quality of the depressive mode as well? If paranoia and depression are also often linked, why are they here put in direct opposition? Even if I make the very reasonable assumption that the confusion is on my end, wouldn’t any attempt to better understand this distinction be based in a paranoid desire to figure out what Sedgwick actually means? Is asking these questions and admitting to confusion a paranoid or reparative practice?27 Regardless, it is nigh impossible except in the most extreme cases to determine whether or when a reading strategy falls into one basket or the other.28
And third, very few of the calls for new reading methods do more than gesture toward actual examples of these methods in practice.29 Sedgwick argues that the essays collected in Novel Gazing are examples of the creative [End Page 9] anti-paranoic reading strategies she is calling for, but one would be hard-pressed to identify specific instances where the analysis could not just as easily be categorized as (very good) traditional close readings. Like Sedgwick, Best and Marcus mostly avoid examples. This lack of examples has led to a great deal of confusion over how to implement these methods. When these analyses do include examples, they are rarely clearly distinct from the close readings they are set against. Sometimes, as in the case of Sedgwick, Latour, or Best and Marcus, the examples that do appear suggest that these new methods simply lead back to very nuanced (i.e., good) close readings. For example, Love asserts that her “close but not deep” reading strategy is necessary to avoid making unsubstantiatable arguments about what a book does or does not mean, only to then make a sweeping argument about the meaning of Toni Morrison’s Beloved based on a close reading of two infrequently discussed pages.30 In the process, Love ends up illustrating (ironically or not) just how hard it is to imagine the value of a humanities paper that does not claim on some level to understand and be able to explain its text. While quite diverse in the way they do so, all of these various methods of reading are united in their shared desire to locate or divine the meaning of a text. These scholars each at least implicitly believe that texts do have some meaning that can be derived, but instead of framing this meaning as deep, they frame it as obvious. As Love’s reading illustrates, even this claim is belied by the continual desire to develop original, surprising, and convincing readings of rich texts. Indeed, if Love’s reading of Beloved wasn’t deep, it wouldn’t be as surprising and impressive (not to mention publishable) as it is.
This desire to read deeply is so naturalized that it is often considered part of the definition of reading itself. Even the term reading implies a desire to divine meaning.31 As Best and Marcus state, they and the vast majority of “post-1983 English and comparative literature professors” equate reading with “interpretation: with assigning a meaning to a text or a set of texts,” in which texts are also defined as coherent objects of some sort.32 The authors are here defining texts extremely broadly (as almost any object or event), but in the process, they are excluding those things, which by design or not, do not make sense (that is, cannot be read). According to Best and Marcus’s definition, you cannot read something that does not make sense, since to read is to make the text make sense.
If what you want to study is not obviously a cohesive object, then the first step is often to make it one by explaining away the fissures and stressing the unities. Robin Wood argues that the desire to make things cohere is a symptom of the drive to dominate “through objectification and the denial of otherness, a tendency greatly encouraged by bourgeois capitalism with its [End Page 10] emphasis on possession.”33 Wood considers why so many 1970s Hollywood films are ideologically incoherent and ambiguous and analyzes Martin Scorsese’s Taxi Driver (1976) as a prime example of films “that do not know what they want to say.”34 Yet, his analysis quickly turns toward making the film cohere by explaining how it is actually symptomatic of “Hollywood cinema’s (America’s) continuing inability to resolve its dichotomies.”35 For Wood, while the film doesn’t make sense, ironically, that is what makes it coherent and meaningful.
Consequently, this definition of a text and the types of reading methodologies that have arisen from it shape not only how texts are analyzed, but also the kinds of texts generally considered readable. The scholars referenced above engage with Nobel Prize winners, the English literary canon, and a surprising number of Hitchcock films, all of which are already established as sensical texts.36 Insodoing, they implicitly dismiss the many texts that do not make sense to us and therefore do not appear to be texts at all.37 One may experience a text as incoherent for any number of reasons. For instance, the text may come from another culture, the reader may lack the requisite knowledge necessary to understand it, or the text may actively be working to resist meaning making. Notably, none of the new reading methodologies address what to do with a text that you do not understand at all and that may be more broadly incomprehensible. Yet during a period when AI generates more and more media and news (not to mention customer care for many companies) and when Donald Trump tweets his every random thought, such texts do exist and are all around us.38
While none of these new methodologies offer ways to read, interpret, or critique incoherent texts, this does not mean that we should just ignore them or that we can simply throw up our hands and decide that while they may be texts, there is nothing in them to read or analyze. Following Stuart Hall, Wendy Brown argues that to break with “monological, totalizing and linear accounts,” we must first “reckon with the incoherent, multiply sourced, and [End Page 11] unsystematic nature of political orders and rationalities.”39 For Brown, acknowledging and considering incoherence is a necessary and ethical step toward recognizing other non-eurocentric and anti-capitalist ways of being in the world.
Following Emmanuel Levinas, Judith Butler, and various others, Dorothy Hale argues that the “foundational aesthetic” of “the new ethical theory” of literature “lies in the felt encounter with alterity that it brings to the reader.”40 Referring primarily to the works of Henry James, Hale and Butler both argue that it is in those moments in his novels when readers are most stymied into incomprehension that they are also most open to the possibility of ethical connection to others. Incomprehensible moments in novels exasperate readers and, in doing so, put them in a “position to understand the limits of judgement and to cease judging, paradoxically, in the name of ethics.” 41 In other words, in these moments, we stop trying to understand or judge a character’s incomprehensible actions and instead potentially learn to care for them as other. In the process, Hale and Butler both argue, these experiences of otherness, or alterity, can trouble our sense of certainty and teach us to recognize our own assumptions and ideologies so that we might “judge less and undergo more.”42 While some might argue that focusing on one’s own subjective experience not only is inherently limiting but also keeps one from being able to embrace otherness, Hale and Butler assert that these moments make one more aware of their limitations as readers or judges in order to also teach them how to care for others and the world around them, not in spite of their lack of knowledge and understanding of it but rather because of it.
While Edward Said influentially focused on the discriminatory effects of othering, there is a long history of considering the positive creative and critical aspects of otherness, or alterity, especially within phenomenology.43 Georg Hegel argued that experiences of otherness were necessary for self-consciousness, and Edmund Husserl defined experience as a basis for intersubjectivity, or our ability to relate to others.44 He thought that others force us to challenge our solipsistic sense of self and make us more ethical and aware of alternative perspectives that we can never really know. Levinas took this further by arguing that we do not even need to encounter the Other to be in an ethical relationship with them, nor is our ethical duty dependent on them acting ethically toward us; indeed, he argues that we must prioritize our ethical relationship to the Other over our own needs and wants.45 For [End Page 12] Levinas, the possibility that the Other may not respect or value us makes our respect of them all the more vital.
For some, this ability to recognize alternative perspectives is necessary for creativity to take place. Cornelius Castoriadis went so far as to consider otherness, or radical alterity, a synonym for creativity, arguing that it is only through the interaction between unlike things that something new can come into existence.46 But the belief that digital technologies are simply reflecting ourselves back to us has led some, like Baudrillard, to fear that we are no longer coming into contact with otherness and are therefore also becoming less creative.47
Reading only for what makes sense obscures and devalues otherness of all sorts. That said, it is often quite hard to figure out what to do with a text that you truly don’t understand. How and why would you even attempt to write about such a text, especially at length for an academic audience? What would be the point? Perhaps finding that out is the point. As Marc Guillaume puts it, “Thinking without understanding may be the strength of human thought, which gains its particular power from its imperfection.”48 The rest of this article will focus on how Sunspring allows us to think through all the ways in which reading can become a way to erase incoherence—and the problems that erasure can cause.
“I DON’T KNOW ANYTHING ABOUT ANY OF THIS.”
How can we have meaningful encounters with the meaningless? How can we become more attuned to strangeness and value its seeming lack of productivity? And if we choose to do so, how can we articulate something interesting without reducing its strangeness? The following reading of Sunspring is an attempt to do just that.
With its opening titlecards briefly explaining how the film was created using the same AI technology that smartphones, computers, and many websites use to guess the next word you want to write, Sunspring sets viewers up to compare their experience of the film with that of their smartphones and computers. The opening text is broken up into several black screens cut together with jarring white noise effects and stretching text reminiscent of [End Page 13] horror film trailers and openings where text is used to establish the haunting situation the characters find themselves in and suggest that thrills and chills will follow. These opening title cards also include the entire screenplay, which viewers can pause on, read, and then consider in relation to the film itself. With these title cards in mind, I viewed the rest of the film with a focus on how the actors performed their lines and how the AI’s directions were otherwise interpreted and carried out. One can certainly think of this production as a collaboration between AI and humans, but only in a limited sense. For while the AI generated the script, it was not part of any future discussions on script revisions or production questions. Regardless, I am here interested in how the filmmakers read the AI output, how they tried to make sense of it as they determined how to best perform and shoot it, and how this reading process is evident in the resulting film itself.
As noted above, my expectation here was that the screenplay and film would display and reflect familiar science fiction clichés. Yet it is noticeable that those clichés that are present were put in not by the AI but rather by the human filmmakers. The first shots of the film do suggest that it will be familiar to genre fans. After the title cards, we are greeted with a close-up of a desk drawer with the film title in Futura Heavy font superimposed upon it. From off screen, a hand opens the drawer and inside is a book called Sunspring, which H (Thomas Middleditch) picks up, thumbs through, and then quickly puts down (See Figure 3). Importantly, the specific font and placement of the tile over the drawer were not referenced in the script at all, yet they comprise a direct and human-generated reference to Wes Anderson’s playful style and signals that more metacinematic clichés may be coming.
Then H says, “In a future with mass unemployment, young people are forced to sell blood. That’s the first thing I can do.” To this, the other character named H (hereafter H2; Elizabeth Gray) quickly and caustically responds from across the room, “You should see the boys and shut up. I was the one who was going to be a hundred years old.” This is the beginning of a conversation in which the characters continually appear to be responding to each other but with non sequiturs that do not make any sense. While the lines written by the AI are incoherent nonsense, the actors perform them as if they are meaningful and deeply antagonistic. The cinematography and mise-en-scène only add to this incoherence. The filmmakers’ handheld camerawork and shot-reverse shot editing are largely from H’s perspective. While this camerawork aligns viewers with his perspective, it also frames H as awkward and distant from H2.
Beyond this formal positioning, the audience is framed as an outsider many times over. They cannot understand what the dialogue means, where the scene is taking place, what is happening, or whether the characters or even the actors understand what is going on. However, the filmmakers go out of their way to also frame and contextualize this incoherence; they thus attempt to make it make sense. Both H and H2 wear gold and silver, textured fabrics (chosen by the human filmmakers rather than the AI) that connect them visually even as the tone of their argument and physical distance frames them as deeply antagonistic. Their shiny outfits also bear a passing [End Page 14]
resemblance to Sun Ra’s afrofuturist costumes. If anything, this reference puts a spotlight on the glaring absence of Black people and discussions of race and ethnicity from this film and the future that the AI and human film-makers envision. These appropriated outfits also notably clash against their bare, white, cramped office with its cheap fixtures and plain desks covered in computer detritus. As the film progresses, the space becomes only trashier as a low-fidelity graphic of stars moving in the distance replace one wall. Soon thereafter, the wall reappears with a cheap toy gun attached to it with electric tape. These aesthetic details are found within the AI-generated script and the human filmmakers are emphasizing them. In particular, the amateurish, slapped-together set works to align the script and overall film with Ed Wood’s low-budget sci-fi aesthetic.
Soon after H and H2’s conversation, C (Humphrey Ker) walks into the room, picks up a computer tablet that immediately scans his face, and laughs. The script states that C “picks up a lightscreen and fights the security force of the particles of a transmission on his face.” Of course, this action does not obviously translate into a particular acting cue; the filmmakers have chosen to reduce this nonsense phrase into something humorous that the actor can perform and we can understand. Either way, this moment has little to do with anything else that happens in the scene and is never discussed again. What is meaningful, though, is how this moment is filmed. C’s flippant attitude and positioning next to a smiling H2 in this moment keeps his character lighthearted; the filmmakers make C’s character seem knowable, as an amiable braggart who may (or may not) now be in a relationship with H2. While this scene does not make sense for us, it does for the characters. As a result, we realize that we as viewers are not watching the AI screenplay. Rather, we are watching people trying to make sense out of the nonsensical screenplay; we are watching reading in action.
In response to C’s appearance, or perhaps just randomly, H begins to choke, spits a plastic eyeball into his hand, and tosses it away. The script describes this as “to Hauk, taking his eyes from his mouth.” There is no character named Hauk in Sunspring, although the AI could be referring to H, H2, C, or a character that is there but has no lines. A different, and equally plausible, [End Page 15] way of dealing with this action would be to have H look at C’s mouth and then look away rather than to take an eyeball from his own mouth. I am tempted to argue that one of these readings would be literal versus figurative, but I’m not sure how that binary would actually apply to such a text. Figurative and literal, surface and depth: such binaries cannot function here as there is no way to know what the script means in any basic sense. Nevertheless, a sense of coherence is again created through the filmmakers’ reading of this particular moment as they turn it into an allusion to a long line of experimental and surreal films, starting with Luis Buñuel’s Un chien andalou (An Andalusian Dog, 1929). Through this choice of the human filmmakers, the film’s incoherence is contained by a sense that it is purposefully surreal.49
Much of this process of reading, or making sense of the script, also meant adding more genre markers that give the audience the feeling that they understand what is going on, even if there is nothing to understand. For instance, H2 yells at H as she romantically touches C, the other male character. The way the scene is acted suggests that the characters are in a love triangle, even though there is nothing in the script that would suggest this. Shortly thereafter, we see H standing over the prone body of C. Based on genre and acting cues, we thus are led to think that H murdered C, perhaps due to the implied love triangle. Yet the script does not really suggest this at all:
[H] is standing in the stars and sitting on the floor. He takes a seat on the counter and pulls the camera over to his back. He stares at it. He is on the phone. He cuts the shotgun from the edge of the room and puts it in his mouth. He sees a black hole in the floor leading to the man on the roof. He comes up behind him to protect him. He is still standing next to him. He looks through the door and the door closes. He looks at the bag from his backpack, and starts to cry.
To state the obvious, none of this makes sense, and there is no reason to think it should. But the producers and actors do try to make it make sense, and in so doing, they resort to familiar motifs of betrayal and murder, even when the script specifically states that H “protect[s] him,” stands next to him, and cries.
At other times, the human filmmakers try to make the script coherent by emphasizing the humor and oddity of its incoherence. Take, for instance, the moment when H says to H2, “I saw him again. The way you were sent to me . . . That was a big honest idea. I am not a bright light.” This series of phrases neither follow one another nor what the previous person was saying. [End Page 16]
This line is delivered with pathos as if it is a significant and meaningful moment. Yet, the moment is rendered absurd as C responds flatly, “Well, I have to go to the skull. I don’t know.” Indeed, throughout the film, characters repeat variations of “I don’t know,” including “I don’t know any of this,” “There is no answer,” and “I don’t know what you’re talking about.” While these lines are in the AI’s script, the actors read them with smirking, self-conscious grins that intensify a knowing humor that may reassure viewers that, while they do not know what the characters are talking about, neither do the characters (and actors) themselves. Ironically, these “I don’t know” lines are often the only ones that make sense in their context, as they seem to suggest that the audience isn’t wrong: the film really doesn’t make sense.
What does it mean that an AI generated such an incoherent text? Why doesn’t the AI provide any insight into the texts it “read” to make this script, and why might we hope and expect that it would? In an attempt to read Sunspring as a cliché-filled reflexive science fiction film, Ross Goodwin argues that these lines reveal the pervasive pattern in sci-fi “movies of characters trying to understand the environment.”50 Is Goodwin close reading, surface reading, too-close reading, or misreading? I can’t tell, and that is, to some extent, the point. While Goodwin is certainly right that sci-fi movies often focus on learning and discovery, the characters in Sunspring do not say “I don’t know” as an impetus to learn more but rather as a declarative statement. They do not actually seem to be particularly concerned with all the things that they do not know or with the larger incoherence that surrounds them. They perform their roles as if they do understand what is going on even if we, the audience, have no idea.
Not knowing and not understanding become a repeated motif of the generated script. To make this insight productive, I consider what this motif helps me understand about the role of incoherence more generally within [End Page 17] science fiction and algorithmic reading practices. In not making sense when I expect it to and when it itself appears to want to, Sunspring leads me to suspect that we have a very different relationship to our algorithmic technologies and their output than we think. I tend to view them mostly as mimetic technologies—reflecting ourselves back to us—but what if they are also othering devices?
How critics “read” the “reading” of the AI script is also striking for what they ignore, overlook, or simply do not notice. For instance, in spite of the film’s extreme incoherence, a quality that would seem to undermine the prevailing notions of genre and the generic, Annalee Newitz, in conversation with Sharp and Goodwin, considers the film to be “a mirror of our culture.”51 While I did not expect Sunspring to mirror our culture, I did expect it to mirror the science fiction it had been shown, but the script itself does not. If anything, Sunspring gives us a chance to question why we continually think of our relationship to digital media in terms of mirrors. This rhetoric traces back to the very beginning of AI with Alan Turing’s imitation game, wherein a computer must trick a person into thinking that it is speaking to a woman. The guiding assumption is that AI is only intelligent insofar as it thinks and acts like its human creators.
The history of AI is littered with programs that humans have tried to make in their image. One can see this reflective, narcissistic logic in Eliza, the famous AI Rogerian Therapist who would repeat back statements from patients in the form of a question. It is also present in popular representations of AI that guide the public’s expectations of what it should be capable of. In Star Trek: The Next Generation, the android Data’s greatest wish is to be human. On HBO’s Westworld (2016), the robots want to be people, and the people want to be robots. The same catoptric trope continues today, as journalists and scholars continually compare the structure of AI machine learning programs to the human brain, even though this analogy is far from accurate.52 Considering that even cognitive psychologists know shockingly little about how the human brain actually works, it is presumptuous to think that we are capable of modeling a computer after it. In attempting such metaphors, we try to model how we think our brains work, which is in turn increasingly defined by how we think computers work.
Even so, this belief that computers inherently mirror their users guides [End Page 18] how both the creators and users of these technologies currently perceive our relationship to digital technologies. It shapes our understanding of what makes digital media “digital” insofar as users now routinely expect that their technologies simply extend themselves outwards into the world while displaying them back to themselves. Digital output certainly can mirror us back to ourselves, but that is hardly the only thing it does. Here I am reminded of Guy Debord’s apocryphal quoting of Karl Marx: “People can see nothing around them that is not their own image; everything speaks to them of themselves. Their very landscape is alive. Obstacles were everywhere. And they were all interrelated, maintaining a unified reign of poverty.”53 How much of the world do we miss when we only look for ourselves? Sharp and Goodwin are hardly the only artists and scholars interested in this question. As Safiya U. Noble, John Cheney-Lippold, Taina Bucher, and many other algorithmic culture scholars attest, the stakes of this question are quite high, especially for minorities and other oppressed groups. They require us to become more attuned to the othering dimensions and interpellative effects of the digital media that surround us.54 In addition, many of the works of Ed Atkins, Zach Blas, Annie Dorsen, Jason Salavon, and Hito Steyerl (among many others) are in dialogue with this question and take the othering aspects of algorithmic technologies and media seriously.
But if Sunspring is not simply a purified form of the science fiction genre, what does that mean for scholars of algorithmic culture who typically fear that we may become generic clichés of ourselves? This fear is still reasonable, but Sunspring offers us the chance to imagine other ways to think of our relationship to the digital technologies around us. Echoing Baudrillard, Ted Striphas, Eli Pariser, Mark Andrejevic, and others argue that personalization technologies and algorithms like those used by Facebook, Amazon, Google, and their ilk take data about their users and exploit it to reflect users back to themselves through the commodities, entertainment, news, and information that they recommend.55 Time magazine celebrated this phenomenon when they named “You” the 2006 “Person of the Year” and used as the issue’s cover image a monitor that doubled as a mirror. However, reading the Sunspring script in terms of mirrors is particularly problematic given that the mirror—if there is one—is indecipherable; it had to be “decoded” by human filmmakers for us to even look into it.
I came away from watching Sunspring with the distinct impression that if AI does not imitate genre, perhaps personalization algorithms (which use the same technology) do not actually reflect their users either. This would [End Page 19] explain why so many of them are so terrible at doing so. I have rarely clicked on a Facebook, Twitter, or Google recommended advertisement, and while I may occasionally purchase a recommended book on Amazon or watch a film suggestion on Netflix, these sites present me with hundreds of recommendations every time I log on; the odds are close to random that I will click on something and like it. Yet the advertising of these sites, the reporting on them, their interfaces, and the expectations they create often lead users to only pay attention to the recommendations that they take and completely ignore and forget the vast majority that make no sense at all.56 We are, like Goodwin “reading” Sunspring, trying to prove the algorithm reflects us by pointing to one detail in a vast amount of data.
By focusing on the reflective aspects of AI algorithms, we end up neglecting the various ways in which they are nothing like us. By not taking the incoherence in these algorithms’ output seriously, or even really noticing it, we risk limiting the potentials of algorithmic production. Instead of mirroring us, Sunspring challenges our values and our assumptions about how the world works and what is possible. It allows me, at least, to pay attention to and consider the presence of the incoherence in digital culture. If Sunspring is a mirror at all, it is one that reflects back far more than what stands before it. Thus, it calls for a reading strategy that goes beyond looking for ourselves in the text or for what we already recognize.
To that end, Sunspring offers us a chance to rethink how we “read” the output of digital technologies more generally. Rather than focus on the familiar, we may attempt to bring the intensely incoherent to the fore. I do not think of this strategy as being aligned with any one particular reading method. Marking off and strictly following one particular method creates more problems than it solves. Instead of modeling a set of rules that might simply turn into a new method, I have tried to model here the necessity of consciously rejecting methodological oppositions in reading practices in order to better recognize and appreciate the otherness in texts. In contrast to those who may desire to only recognize and celebrate the familiar in Sun-spring, I use the film to ponder what it would really mean to consider the otherness of artificial intelligence as valuable despite the fact that we do not see ourselves in it. This appreciation calls for us to consider what could result if we approach AI with an attitude of respect and care as a collaborator rather than simply as a tool of neoliberal capital.57
During the Sci-Fi-London 48 Hour Film Challenge, the AI behind Sun-spring was interviewed about the future of machine-written entertainment, and it responded by stating, “It’s a bit sudden. I was thinking of the spirit of the men who found me and the children who were all manipulated and full of children. I was worried about my command. I was the scientist of the Holy [End Page 20] Ghost.”58 This clearly makes no sense to us, but what a great line! In trying to read, interpret, or analyze Sunspring while still respecting its incoherence, I, too, feel like a scientist of the Holy Ghost—an entity often defined as an impossible mystery, an incoherence that holds “the mind in a state of wonder and as a reminder that human intellect could never understand the nature of God.”59 AI may not be trying to reflect us, but that should not stop us from trying to reflect on it. [End Page 21]
Jonathan Cohn is an assistant professor in the Department of English and Film Studies and Director of Digital Humanities at the University of Alberta. He is the author of The Burden of Choice: Recommendations, Subversion, and Algorithmic Culture (Rutgers University Press, 2019).
1. The program is specifically a long short-term memory (LSTM) neural network, which is “trained” on a certain data set and then is given a certain prompt. Based on that prompt, the LSTM uses its training data to predict what should come next; e.g., if the training data includes multiple instances of an obscure word like deictic, the output will be more likely to also contain that word.
2. Throughout this article, I use incoherence and nonsense interchangeably, if provocatively. Whereas incoherence indicates a lack of narrative logic, nonsense emphasizes a lack of semiotic rationality. In Sunspring, however, these two levels of balderdash overlap and are inseparable.
3. This phrase is the title of Sharon Marcus and Stephen Best’s special issue of Representations 108, no. 1, (Fall, 2009) which originally featured their introduction essay, “Surface Readings: An Introduction” (p. 1–21). While Representations is a multidisciplinary journal, Best and Marcus argue that the “we” they refer to is still primarily those with English and Comparative Literature PhDs. While the readership of JCMS may primarily be those with a humanities-based Cinema and Media Studies PhD, I hope this new online, open-access format will draw a broader (if also perhaps less coherent) readership.
4. This habit is pernicious in academia as well; see Frances Ferguson, “Now It’s Personal: D. A. Miller and Too-Close Reading,” Critical Inquiry 41, no. 3 (2015): 52, https://doi.org/10.1086/680084. Ferguson starts with a discussion of how personalization algorithms are constantly hailing users to engage them in a form of too-close reading. For her, this is the normative, if not only, way to “read” this output. I am instead interested in what this particular reading strategy occludes: How would reading them differently push us to notice all those recommendations and ads that illustrate how unintimate the vast majority of algorithmic culture tends to be? How we are constantly surrounded by otherness hidden in plain sight?
5. Jean Baudrillard quoted in Jean Baudrillard and Marc Guillaume, Radical Alterity, trans. Ames Hodges (Los Angeles: Semiotext(e), 2008).
6. For many, including Stephen Best and Sharon Marcus, Rita Felski, Heather Love, and others, close reading (and perhaps critique more generally) is far too synonymous with Fredric Jameson’s symptomatic reading. This simplification helps them to more easily argue against close reading entirely. See Stephen Best and Sharon Marcus, “Surface Reading: An Introduction,” Representations 108, no. 1 (2009): 1–21, https://doi.org/10.1525/rep.2009.108.1.1; Rita Felski, The Limits of Critique (Chicago: University of Chicago Press, 2015); and Heather Love, “Close but not Deep: Literary Ethics and the Descriptive Turn,” New Literary History 41, no. 2 (2010): 371–391.
8. Best and Marcus present the torture photos at Abu Ghraib and news coverage of Hurricane Katrina as two examples that require no close reading. One may wonder if the Vietnam War, Nixon’s resignation, or Reaganomics really required a close reading to understand either. Best and Marcus, “Surface Reading,” 2.
10. Latour, “Why Has Critique?,” 228.
11. Many of these methods are discussed in Elizabeth S. Anker and Rita Felski, eds., Critique and Postcritique (Durham, NC: Duke University Press, 2017).
12. N. Katherine Hayles, “How We Read: Close, Hyper, Machine,” ADE Bulletin, no. 150 (2010): 62–79, https://doi.org/10.1632/ade.150.62; Franco Moretti, Distant Reading (London: Verso, 2013); Leah Price, How to Do Things with Books in Victorian Britain (Princeton, NJ: Princeton University Press, 2012); and Eve Kosofsky Sedgwick, “Paranoid Reading and Reparative Reading; or, You’re so Paranoid, You Probably Think This Essay is About You,” in Novel Gazing: Queer Reading in Fiction (Durham, NC: Duke University Press, 1997), 1–40. One might assume that algorithmic readings may be a particularly useful concept here, but it is impossible to tell exactly what this method consists of from Finn’s brief and vague statements on the subject. See Ed Finn, What Algorithms Want: Imagination in the Age of Computing (Cambridge, MA: MIT Press, 2017), 8.
13. Fredric Jameson quoted in Best and Marcus, 2.
14. Felski’s and Best and Marcus’s works are perhaps the best examples of these tendencies, but they are widely evident.
15. The reasons for this are legion, but many are related to a focus on scientism in humanities research funding models. First, many humanities grants now ask for multi-page explanations on method. This may make sense for certain studies, but if you are mainly reading and analyzing books, the sheer length of these sections asks for a level of precision and novelty to our methodologies that disincentivize both traditional and looser reading methods in favor of new and niche ones that require lengthy descriptions. Second, these funding models also may generate a desire to implement digital technologies in research to make it appear more “cutting-edge” and/or objective (but too often also less critical) while also using and studying these technologies in classrooms with the ostensible belief that it will generate more majors (or at least appeal to those in engineering). Third, the inclusion of “new media” in the humanities has led to many (often essentialist) media-specificity discussions around the need to develop new methods to study them.
17. Rick Altman, “A Semantic/Syntactic Approach to Film Genre,” Cinema Journal 23, no. 3 (1984): 6–18, https://doi.org/10.2307/1225093; and Franco Moretti, “Planet Hollywood,” in Distant Reading, 91–107; Project Arclight, accessed May 1, 2019, http://projectarclight.org/.
18. Kael, 22. We may now be in a similar moment as film studios use algorithms to help them decide what films to make, when some viewers watch whatever appears first on their Netflix “for you” lists, and where scholars use software to qualitatively analyze their data for them.
19. Kael, 21.
20. Kael, 14.
21. Paul Feyerabend, Against Method: Outline of an Anarchistic Theory of Knowledge, 3rd ed. (London: Verso, 1993): 10–11.
22. Sandra Harding, The Science Question in Feminism (Ithaca, NY: Cornell University Press, 1986); and Helen E. Longino, The Fate of Knowledge (Princeton, NJ: Princeton University Press, 2002).
23. Best and Marcus, “Surface Reading,” 18.
24. This tendency is perhaps most clearly articulated in the above mentioned 2009 issue of Representations on “The Way We Read Now,” which begins with the suggestion that “we” no longer read symptomatically (closely) but then contains a set of essays that all, to varying degrees, perform symptomatic readings themselves (if not of their actual texts, then of the books as objects or of Jameson’s or other scholars’ work). See Best and Marcus, “Surface Reading,” 6. Other examples include Moretti, Distant Reading, 48; Felski, Limits of Critique, 9; Hayles, “How We Read,” 63–64; and Sedgwick, “Paranoid Reading,” 7.
25. Sedgwick, 7.
26. Heather Love points out just how confusing many have found Sedgwick’s distinction and how this has led to wildly varied and contradictory readings of this essay. Yet Love still argues that there is indeed a correct reading, an argument she renders through a close reading. See Love, “Truth and Consequences: On Paranoid Reading and Reparative Reading” Criticism, 52, no. 2 (2010): 239.
27. One could also seriously ask whether taking Sedgwick’s subtitle, “You’re So Paranoid, You Probably Think This Introduction is About You” at face value is a reparative surface reading or actually generates the paranoia that it critiques. Is this paranoia what generates the reparative reading, or vice versa?
28. For an example of how this method is typically implemented to declare whether a text is empowering or oppressive, good or bad, see Katrin Röder, “Reparative Reading, Post-structuralist Hermeneutics and T. S. Eliot’s Four Quartets,” Anglia 132, no. 1 (2014): 58–77, https://doi.org/10.1515/anglia-2014-0004.
29. Sedgwick primarily critiques examples of paranoid readings and devotes only a page to a consideration of how reparative readings might open up a discussion of camp texts in a very general sense. Sedgwick, 25.
30. Love, Heather, “Close but Not Deep: Literary Ethics and the Descriptive Turn.” New Literary History 41, no. 2 (2010): 383–6.
31. For instance, The Oxford English Dictionary’s first definition for “read, v” is “to consider, interpret, discern.” Many of their other definitions include terms like “guess, make out, or tell by conjecture,” and “foresee, fortell, predict.” See “read, v.” OED Online. September 2020. Oxford University Press.
32. Best and Marcus do not explain why 1983 is their cut-off, but this may be the point at which they believe their field fully embraced Jameson’s understanding of the political unconscious and symptomatic reading methods. “Surface Reading,” 1.
33. Robin Wood, Hollywood from Vietnam to Reagan . . . and Beyond (New York: Columbia University Press, 1986): 41.
34. Wood, 42.
35. Wood, 48.
36. Those essays that attempt post-critical readings of texts in Elizabeth Anker and Rita Felski’s collection, Critique and Postcritique, are exemplary of this tendency.
37. While there are certainly many texts valued for their incoherence, from James Joyce’s Finnegans Wake to Michael Snow’s Wavelength (1967), many others, such as The Room (Wiseau, 2003) or Transformers (Bay, 2007) are considered in/glorious dreck. The vast majority are not remarked upon at all and are simply forgotten. There are also texts such as certain forms of graffiti, static noise, or even some muzak that register as so meaningless that they may not be experienced at all.
38. One could easily write a book on the various efforts made by journalists and scholars alike to interpret Donald Trump’s tweets. They range from those paranoid and deep readings that suggest he is playing “three dimensional chess” to the surface readings that argue we must instead take his words “at face value.” Nguyen, Tina, “Nancy Pelosi Still Worried Trump Is Playing Three-Dimensional Chess.” Vanity Fair. Vanity Fair, May 7, 2019. https://www.vanityfair.com/news/2019/05/nancy-pelosi-still-worried-trump-playing-three-dimensional-chess-impeachment; “Anderson Cooper: Can We Take What Trump Says at Face Value? -CNN Video,” September 20, 2019. https://www.cnn.com/videos/politics/2019/09/20/kth-whistleblower-foreign-collusion-ac360-vpx.cnn.
39. Wendy Brown, “American Nightmare: Neoliberalism, Neoconservatism, and De-Democratization,” Political Theory 34, no. 6 (2006): 691. Brown follows this by also arguing that we just learn to “avow identification and affinity with some of what we excoriate” if we are ever to move forward.
40. Dorothy J. Hale, “Aesthetics and the New Ethics: Theorizing the Novel in the Twenty-First Century,”PMLA 124, no. 3 (2009): 899.
41. Butler paraphrased in Hale, 901.
42. Butler paraphrased in Hale, 901.
43. Edward W. Said, Orientalism: Western Conceptions of the Orient (London: Routledge & Kegan Paul Ltd, 1978).
44. See Martin Heidegger, Hegel’s Phenomenology of Spirit, trans. Parvis Emad and Kenneth Maly (Bloomington: Indiana University Press, 1988); and Edmund Husserl, Cartesian Meditations: An Introduction to Phenomenology, trans. Dorion Cairns (Dordrecht, The Netherlands: Springer Science and Business Media, 1973).
45. See Emmanuel Levinas, Totality and Infinity: An Essay on Exteriority (Dordrecht, The Netherlands: Springer Science and Business Media, 1991); and Emmanuel Levinas, Otherwise Than Being, or Beyond Essence, trans. Alphonso Lingis (Pittsburgh, PA: Duquesne, 1991).
46. Cornelius Castoriadis, The Imaginary Institution of Society (Cambridge, MA: MIT Press, 1997).
47. As Baudrillard writes, “With artificial intelligence and the alterity of machines, we are still faced with the same issue: alterity is in danger.” Baudrillard in Baudrillard and Guillaume, Radical Alterity, 113. For arguments directed at a general audience, see Rachael Rettner, “Are Today’s Youth Less Creative & Imaginative?,” msnbc.com, August 12, 2011, http://www.nbcnews.com/id/44121819/ns/technology_and_science-science/t/are-todays-youth-less-creative-imaginative/; Lee Rainie, John Horrigan, and Michael Cornfield, “The Internet and Campaign 2004,” Pew Research Center’s Internet & American Life Project (blog), March 6, 2005, http://www.pewinternet.org/2005/03/06/the-internet-and-campaign-2004/; David DiSalvo, “Study: The More Stuff We Have, The Less Creative We Are,” Forbes, November 19, 2015, https://www.forbes.com/sites/daviddisalvo/2015/11/19/study-the-more-stuff-we-have-the-less-creative-we-are/; and Rosa Inocencio Smith, “The Internet’s Impact on Creativity: Your Thoughts,” The Atlantic, March 3, 2017, https://www.theatlantic.com/notes/2017/03/internet-creativity-responses/518514/.
48. Guillaume in Baudrillard and Guillaume, Radical Alterity, 100.
49. Framing AI art as “dreamlike” or “surreal” is common. For instance, Google’s Deep Dream program illustrates what an AI sees in an image as it tries to find faces and other patterns. The generated images, which often look like piles of fractalized dogs, are variously described as “dream-like hallucinogenic” and “psychedelic and abstract art.” These creations are exemplary of this effort to make sense of nonsensical pieces by arguing that they are a product of the technology’s subconscious and thus purposefully irrational. Deep Dream Generator. http://www.deepdreamgenerator.com/(Accessed December 11, 2020); Wikipedia contributors, “DeepDream,” Wikipedia, The Free Encyclopedia, https://en.wikipedia.org/w/index.php?title=DeepDream&oldid=992436270 (accessed December 11, 2020).
50. Goodwin quoted in Annalee Newitz, “Movie written by algorithm turns out to be hilarious and intense,” Ars Technica, June 9, 2016, https://arstechnica.com/gaming/2016/06/an-ai-wrote-this-movie-and-its-strangely-moving/.
51. Newitz. Articles and comments online largely considered whether the film is a mirror and what the film reveals about science fiction, if anything. See Kathryn Lawrence, “Sunspring and It’s No Game: Sci-Fi by AI,” MONTAG, August 7, 2017, https://www.montag.wtf/sunspring-sci-fi-by-ai/; and HAL90210, “This Is What Happens When an AI-Written Screenplay Is Made into a Film,” The Guardian (US), June 10, 2016, https://www.theguardian.com/technology/2016/jun/10/artificial-intelligence-screenplay-sunspring-silicon-valley-thomas-middleditch-ai.
52. See John McCarthy, “From Here to Human-Level AI,” Artificial Intelligence 171, no. 18 (2007): 1174–1182, https://doi.org/10.1016/j.artint.2007.10.009; Hans Moravec, “When Will Computer Hardware Match the Human Brain?,” Journal of Evolution and Technology 1 (1998): 12; Tristan Greene, “Researchers Developed Algorithms That Mimic the Human Brain (and the Results Don’t Suck),” The Next Web, April 4, 2019, https://thenextweb.com/artificial-intelligence/2019/04/05/researchers-developed-algorithms-that-mimic-the-human-brain-and-the-results-dont-suck/; and “The Brain Inspires a New Type of Artificial Intelligence,” Science Daily, August 9, 2019, https://www.sciencedaily.com/releases/2019/08/190809085729.htm.
53. Marx quoted in Guy Debord’s Sur le passage de quelques personnes à travers une assez courte unité de temps (1959).
54. Safiya Umoja Noble, Algorithms of Oppression: How Search Engines Reinforce Racism (New York: New York University Press, 2018); John Cheney-Lippold, We Are Data: Algorithms and The Making of Our Digital Selves (New York: New York University Press, 2017); and Taina Bucher, If . . . Then: Algorithmic Power and Politics (New York: Oxford University Press, 2018).
55. See Ted Striphas, “Algorithmic Culture,” European Journal of Cultural Studies 18, no. 4–5 (2015): 395–412, https://doi.org/10.1177/1367549415577392; Eli Pariser, The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think (London: Penguin Books, 2012); and Mark Andrejevic, iSpy: Surveillance and Power in the Interactive Era (Lawrence: University Press of Kansas, 2007).
56. For one example, see Laura Hurley, “How Netflix Knows Its Subscribers Better Than We Know Ourselves,” Cinema Blend, June 2, 2017, https://www.cinemablend.com/television/1666330/how-netflix-knows-its-subscribers-better-than-we-know-ourselves.
57. Newitz’s article, in which the film originally “premiered” online, is the most obvious example of this trend, but it is pervasive in every article that followed it. Newitz, “Movie Written by Algorithm.”
58. Quoted in Newitz.
59. Karen Armstrong, A History of God (New York: Ballantine Books, 1993): 306.