The “algorithmic as if ”: Computational resurrection and the animation of the dead in Deep Nostalgia

Abstract

Contemporary artificial intelligence and algorithmic processes address deep-seated existential challenges and modes of desire. In so doing, they produce computational systems of imagination, an “algorithmic as if” that enables the expression, transformation, and seeming overcoming of existential limitations via technological means. This article elaborates the character of the “algorithmic as if” by focusing on Deep Nostalgia, an online tool that turns personal photographs of the deceased into looped animations which smile, blink, and move, promising to overcome mortality by technologically “resurrecting the dead.” Performing a close-reading of Deep Nostalgia’s technological processes and the public discourse around its 2021 launch, the article highlights its combination of computational learning, forms of visual representation (photography, video, and animation), and distinctive realignments of temporal experience. Together, these frame the “algorithmic as if” as a magical and affective space for realizing impossible longings that are also reflexive encounters with the “limit-situation” of human mortality.

Introduction

A certain thing in life is death. This certainty appears to be challenged by a new technology called Deep Nostalgia™. Launched on February 25, 2021 by the global genealogy company MyHeritage, ¹ Deep Nostalgia is an artificial intelligence (AI)-based service that animates photographs of people who have passed away: the faces in the photographs processed through its online interface are suddenly endowed with expressive movement (head tilt, eye movement) and changing expressions (e.g. smiles), producing a looped video that brings them abruptly “to life” (see Figure 1).

Figure 1.

An animation of a woman’s face by Deep Nostalgia™. (The original photograph can be seen in the small images).

The launch of Deep Nostalgia provoked a strong and immediate public reaction. It was discussed on morning programs, talk shows, social media, and the news. ² It went viral quickly, with over 106 million animations, 96 million of them created in under a year (MyHeritage, 2022a). ³ Its appeal to the public imagination seems to complement the emphasis on “liveness” and “presentness” in social media interactions and much contemporary mediated temporal experience (Coleman, 2020). Its connection to questions of identity and genealogy, and above all, its promise to return the dear departed to life, bring algorithmic processes and visual media into stark engagement with profoundly existential matters. Through facial animation and “deep learning,” Deep Nostalgia appears to offer the gift of technological resurrection.

This digital resurrection is part of a broader cultural moment: powerful institutions and corporations currently seek to intervene in the temporality, and finitude, of human life (e.g. Franklin and Lock, 2003; Levitt, 2014). This obsessive engagement includes “remedies” such as anti-aging products and surgical procedures that (allegedly) secure a more youthful appearance; AI technologies, such as Amazon’s Alexa, which simulate the voices of users’ dead relatives; and holograms of Holocaust survivors who hold “conversations” with contemporary interlocutors (Pinchevski, 2019). It also participates in a longer history, discussed below, of media’s status as technologies for bridging the ultimate frontier of human mortality, for enabling contact and interaction with those on “the other side” of death (Lagerkvist, 2022; Peters, 1999; Sconce, 2000). Finally, and no less significantly, it crystallizes the existential investments of contemporary algorithmic culture, which is especially timely given the current inflation of AI-driven technologies of cultural production such as Midjourney, Dall-E, ChatGPT, and so on. It offers a concrete visual performance of how deeply seated human longings can be addressed and expressed through systems of computational imagining, and the operations and limitations of those systems in the case of human life and death. These systems produce an “algorithmic as if”: a computational apparatus designed to envision and incarnate one’s heart’s desire—a desire that is inherently related to human life, but which cannot happen in real life.

The term “algorithmic as if” designates an overtly imaginative, subjunctive, and existential facet of algorithmic culture in general, one which also marshals and realigns the conventional temporal affordances of existing media (such as photography and film). In analyzing some of its mechanisms, this article mobilizes several areas of theory and research. First, it builds on recent scholarship that treats algorithms as agents which shape human knowledge, cultural practices, and social experience (e.g. Bucher, 2018; Gillespie, 2016; Hallinan and Striphas, 2016; Striphas, 2015), contributing to conceptualizations of contemporary computational culture that highlight the power of algorithms in everyday life, and the intimate affective entanglements of algorithmic processes with individual selves (Savolainen and Ruckenstein, 2022). Second, it draws on relevant literature concerning media’s construction of temporal experience, notably from photography and media theory. Finally, it treats the “algorithmic as if” as an apparatus of “existential media”: a way of approaching and making manifest underlying conditions of digital existence (conditions of life and death, and their digital performance), particularly as conceptualized by Amanda Lagerkvist (2022) in her reframing of Jaspers’ (1970 [1932]) “limit-situation” for the digital era.

Overall, these approaches are integrated into an analysis that shows how algorithmic technologies, such as Deep Nostalgia, are mobilized to articulate (and purportedly fulfill) deep-seated existential human desires. Such technologies participate in the work of more traditional funerary, mourning, and ancestor-revering rituals. They give historically and socially contingent cultural form to general aspects of human existence (e.g. birth, growth, decay, death). Hence this article contributes to the ongoing investigation of how death is constructed, negotiated, and managed by and through media technologies in the contemporary era. Furthermore, it offers a critique of how algorithmic interventions utilize visual technologies such as photography, video, and animation, and prevailing cultural perceptions of their fidelity to the real, in order to automate and even supplant human emotional response and imagination.

However, the article also insists on the ultimate incompleteness of such automation and supplantation. This incompleteness, it will argue, is particularly visible in cases of the limit-situation, where extreme existential boundaries are faced. The incompleteness of computational supplantation takes two inter-related forms: the first is that limit-situations such as death exceed calculative procedures that, for instance, might be appropriate to more quotidian domains of social and personal life. Second, human subjectivity is also fundamentally resistant—through its very volatility and opacity (Andrejevic, 2020)—to the computational automation of desire and imagination. The algorithmic expression of the limit-situation thus makes possible meaningful encounters that enhance consciousness of the very boundaries of human existence, and of the technologies devised to represent and overcome them.

Theorizing the “algorithmic as if”

The “algorithmic as if” designates the mobilization of desire, will, and imagination in computational systems. It is an energy force within “algorithmic culture,” defined as “the use of computational processes to sort, classify, and hierarchize people, places, objects, and ideas, and also the habits of thought, conduct, and expression that arise in relationship to those processes” (Hallinan and Striphas, 2016: 3; see also Striphas, 2015). Culture, which in the past was primarily constructed, performed, and defined by human beings, is today increasingly being delegated to computational systems. This delegation underpins, among other things, the emergence of new cultural forms, including those—such as technologies of reanimation and resurrection—which intervene in cultural definitions of existential categories of life and death.

A key challenge for such computational delegation is the stubbornly unpredictable character of human subjects, which the increasing automation of media through computation is ultimately designed to address and manage (Andrejevic, 2020). Human subjects are volatile, unstable, and opaque—including to themselves—and can think, feel, and behave in ways that threaten control and governance systems. In contrast, automated subjects are perfectly self-identical and can be predicted based on data. Hence, the human subject will always be somewhat resistant to control, undermining the idea of the “singularity,” a post-human future in which humans and machines are merged.

As a compound term, the “algorithmic as if” brings together the attempted computational automation and management of human subjectivity (the “algorithmic”), and questions of desire, emotion, possibility, and imagination (the “as if”). While the “as if” has long been associated with the imagination and fictionalization (Mulvin, 2021; Walton, 1990), our use of it is especially indebted to Zelizer’s (2010) conceptualization. Zelizer identifies a core tension in news and documentary photographs of people “about to die”—individuals captured photographically in the instant preceding their deaths. This tension is between the “as is,” the indexical depiction of an inexorable linear sequence in the past, and the “subjunctive voice” of the image that—through freezing the temporal flow before its fatal conclusion—enables viewers “to embellish numerous emotional, imaginary and contingent schemes” (p. 18) of what could be or might have been. The “subjunctive voice” facilitates a deeply unsettling conjectural attitude concerning the possible and the desirable (i.e. that they should not die): the subjunctive mood of the “as if.” “With photography,” says Zelizer, “subjunctivity offers a way of transforming the relationship between the possible, probable, impossible, and certain by accommodating contingency, the imagination, and the emotions” (p. 14).

The “as if” of photography invokes the unpredictability and contingency of human desires and imaginings. The algorithmic ‘‘as if,” in contrast, underpins computational attempts to predict and fulfill those desires; to automate them, in Andrejevic’s terms, to desire and imagine on our behalf. Moreover, the ‘‘algorithmic as if’’ not only describes a computational anticipation of desire, but invokes the very imaginings and drives that give rise to the algorithm in the first place as a conditional and contingent formulation: the “if . . . then” statements of algorithmic rules which “imply that certain conditions must be met in order to have particular consequences” (Bucher, 2018: 159). Since the algorithm is manifested in the relationship between “if” and “then,” a key question is always: what produces the “if” itself, the starting point of the algorithm? The “algorithmic as if” highlights the imagination of desirable and applicable “conditions” which potentially stands behind every algorithm, and that is technologically and socially realized through its operations. Yet, as argued below, its processes can nevertheless encounter and even activate the volatility of human subjects, and their own “as if” desires and imaginative capacities.

Materializing the “algorithmic as if”

Significantly, the case of photography places the “algorithmic as if” in a longer history of technologically realized social, economic, and political will. Photography developed into a global system of “automatic” visual representation well before today’s computational processes, tethered to scientific positivism and capitalism in the 19th and 20th centuries (Berger, 1982). The automation of perception and representation by the camera was widely materialized as a techno-cultural practice that aimed to achieve total knowledge and to satisfy seemingly universal desires. Photography participated in automating highly conventionalized and commodified “imaginings” of physical and social realities according to prevailing capitalist, patriarchal, and colonialist ideologies and interests (Tagg, 1988). The “algorithmic as if” is thus tied to the intensification of automated processes and outcomes that are facilitated by computational technologies in contemporary globalized capitalism, but whose relations of social, cultural, economic and political power have a longer history and do not originate in them.

Finally, although the “algorithmic as if” is a potential component of all algorithmic processes, its operations are likely to be most conspicuous in “limit-situations.” Here its friction with the volatility of human subjectivity is most keenly experienced, and the potential for disturbing or undermining attempts at automating desire becomes more salient. Defined by Karl Jaspers (1970 [1932]), the “limit-situation” designates a conscious awareness of the thresholds of human existence, such as struggle, suffering, and death, which offers a possibility for human authenticity and realized self-being. Limit-situations can amplify and reveal the unpredictability of the subject, evoking strong feelings and reactions the subject is not completely aware of in advance, such as crying or uncanniness, discomfort, or resistance. In contrast to many other automated algorithmic media processes, and to the technological fantasy of algorithmic technologies transcending human finitude, products of the “algorithmic as if” in limit-situations potentiality reinstate subjectivity rather than facilitate its perfect management and automation.

To outline and investigate the “algorithmic as if” in detail, this article presents a “theoretically informed close reading” (Frosh, 2018: 25) of Deep Nostalgia as a productive “object to think with.” Such an approach not only encompasses overt features of the technology but also utilizes MyHeritage’s official website, blog, and social media accounts as useful resources for interpretation and conceptualization, since these have (for reasons mentioned below) made many of Deep Nostalgia’s technical features accessible to public view, and they also address its uses and somewhat conflicted public reception. The analysis of these sources is extended through references to public discourse about Deep Nostalgia’s animation of the dead in Facebook and Twitter posts and comments, as well as in mainstream news coverage.

The “deep learning” of Deep Nostalgia

MyHeritage, an online genealogy company that provides services like creating family trees and finding lost relatives, launched Deep Nostalgia as part of the digitization phase of historical family albums. ⁴ The technology behind Deep Nostalgia is licensed by MyHeritage (2022a) from D-ID, a company specializing in video reenactment using AI “deep learning” processes. These processes involve two main stages: recognition of a person’s face in a frontally posed photograph uploaded through the service’s online interface, and the application of special video drivers that animate the face. As a result of this application, the still photograph is transformed into a short, looped video, much like a graphics interchange format (GIF).

The technological capacity to animate pictures of others’ faces is presented in the company’s marketing material through two rhetorical frames. The first seeks to distinguish Deep Nostalgia from so-called “deepfake” images. Products of manipulation software designed “to deceive or mislead viewers by using bad-faith actors for persona appropriation” (Bode, 2021: 921), deepfakes are perceived to be ethically dubious and harmful to individuals, institutions, and democratic societies. MyHeritage is therefore careful to differentiate itself from deepfake procedures. To prevent the creation of deepfakes of living people, Deep Nostalgia produces only silent animations; additionally, MyHeritage (2022a) requests that users upload only personal historical photos of people who have passed away. Deep Nostalgia thus performs manipulations on images of human faces, but for a discursively legitimated purpose—aimed at bringing the “ancestors back to life” (MyHeritage, 2022a). Furthermore, MyHeritage does not conceal the algorithmic procedures of Deep Nostalgia from users. In fact, it openly publishes explanations of how the technology works.

The second, more prominent, rhetorical framework emphasizes the quest for personal identity, focusing on historical provenance and the primacy of kinship. This ties the technology to the brand values of MyHeritage (2022b) more broadly, as expressed in the company’s slogan: “Discover. Preserve. Share. Your Family History”. The term “family history” is often repeated on the MyHeritage (2022a) site in the context of Deep Nostalgia, where, customers are told, you can “experience your family history like never before.” Founder and chief executive officer (CEO) Gilad Japhet declared that Deep Nostalgia offers “new ways to allow people to connect to their family history” (MyHeritage, 2021a).

Yet Deep Nostalgia’s technical procedures offer a revealing contrast to the company’s rhetoric of family-specific lineage. The animation of photographed subjects is underpinned by a set of facial expressions that constitute the data source for pre-recorded driver videos, which can then be applied to any facial portrait; they include a specific range of movement options, notably smiles, blinks, and head-turning (Shuman, 2021a). And these drivers have been developed from the facial movements of strangers.

Animating people’s faces from the past is, of course, not unknown in media industries such as cinema. For instance, practices of “de-aging” and resurrecting have recently become familiar when film actors need to play a younger version of themselves or their characters, or when actors have passed away (Golding, 2021). However, in these cases, digital de-aging uses computer-generated imagery (CGI) that draws on photographs and filmed recordings of the same actors in the past. In stark contrast, and despite the rhetoric of identity, family, and kinship, Deep Nostalgia’s driver videos are based on a set of filmed sequences from a sample of MyHeritage employees (see Figure 2).

Figure 2.

An example of how the source data of the driver video is applied: On the left, a driver video based on the facial expressions of Sagih Keinan, of the MyHeritage marketing team; on the right, the application of the driver to a still photo.

The final animation created by Deep Nostalgia, therefore, constitutes a new indexical-generic hybrid, which is itself produced by combining two different indexicalities—the historical photograph (which has indexically registered the faces of those in the pre-photographic scene) and the video (which has indexically registered the facial expressions and movements of MyHeritage employees). While the photograph is visible and significant as the (digitized) material foundation that the animation works upon, the driver video functions as the invisible operative procedure that creates a sense of presence—presence through movement—in the final animation (Niemelä-Nyrhinen and Seppänen, 2020). However, while the still photograph indexes a specific, identifiable, historical subject, the final animation does not, since the facial movements attached by the driver video originate from someone else.

The claim that the application of expressions of contemporary individuals to 19th or 20th-century faces can reflect history and, precisely, family history is, of course, deeply suspect. The dataset of driver videos is based on particular assumptions and worldviews (e.g. Barlas et al., 2021); for instance, there is no sensitivity to gender, as shown in Figure 2, where a man’s expressions are applied to the portrait of a woman. Furthermore, in contrast to many AI facial recognition and image-production technologies which are based on millions of images (Bucher, 2022), Deep Nostalgia creates universalism from a small dataset; only 20 driver videos from a company with 520 employees (MyHeritage, 2022c).

By applying driver videos of MyHeritage employees, Deep Nostalgia reproduces genealogy not only by canceling the individual’s “visual DNA”—which can be expressed through dimples, facial mimicry, and other facial expressions—but also by naturalizing its own models of facial movement which it presents as “universal.” Historically, this production of “new” facial expressions by combining images of different people resonates with the “composite portraits” (Galton, 1879) of the 19th century. With these portraits, Galton attempted to legitimate and operationalize social Darwinist beliefs, arguing that the generic facial structures supposedly revealed by his composites demonstrated the existence of universally valid psychological, social, economic, and racial types. While the ideological context is very different, Deep Nostalgia’s composites similarly construct generic resemblance; not, however, as generic faces, but as generic facial expressions. Thanks to this procedure, one’s photographed ancestors now move and smile in the same way everywhere. Deep Nostalgia’s intervention thus diminishes individual and group difference, animating one’s family with an algorithmic processing system that reduces possible expressions to a repetitive set of predictable facial movements.

This procedure evokes long-standing controversies over the purported universality of facial expressions and their “underlying” emotions, and the discriminatory biases attending such theories (Ekman et al., 1987; see Ratner (1989) and Leys (2011) for critiques). In the contemporary context, it echoes concerns among researchers regarding the politics of computational and AI systems, and their ability to naturalize and universalize deeply biased epistemologies and culturally entrenched inequalities. Chun (2021) has critiqued the biases of datasets and stereotypes in facial recognition and similar technologies, which produce forms of “recognition as generalization” that discriminate against individuals in marginalized and stigmatized populations. Building on Chun, one could argue that Deep Nostalgia employs processes of “animation as generalization,” a generative version of the procedure of black-boxing a dataset of expressions that—by being applied to a wide range of diverse faces on a range of different photographs—are framed as “ideal types.” Bucher’s (2022) concept of “fAIce communication” also develops this critique. Deep Nostalgia’s bias is not in the recognition of faces, however, but in their construction: not in the way machines “read” human features, but in how humans are positioned to read faces that are assembled and composited by machines. It gives new weight to Bucher’s (2022) claim that “with regard to machine learning and relational faces, today, there might even be more signifying power emanating from the amassing and assembly of faces than from the individual face” (p. 651).

Animation, temporality, and the mediated presence of the dead

The portrayed ancestors are almost always dead. While Deep Nostalgia realigns the generic and the particular in a hybrid system of computational imagining, the technology’s “as if” dimension also draws, as mentioned earlier, on a longer history of media’s relation to life and death. Nineteenth-century recording media (e.g. phonography and photography) enabled communication between the living and the dead by preserving traces of individual presence, whether through sight or sound, filling the world with “‘phantasms of the living’ for playback after bodily death” (Peters, 1999: 142). In the 20th century, electronic media like radio and television circulated new spectral forms. “. . . Sound and image without material substance, the electronically mediated worlds of telecommunications often evoke the supernatural by creating virtual beings that appear to have no physical form” (Sconce, 2000: 4). Bringing this history to bear on contemporary digital contexts, Lagerkvist (2022) proposes the term “transcendent media” to designate “media that promise to transcend the ultimate boundaries, by allowing for relating across the threshold” (p. 172).

One key connection of this history is to photography, since photographs are part of Deep Nostalgia’s technical infrastructure. “If the plastic arts were put under psycho-analysis,” notes Bazin (1960) in “The Ontology of the Photographic Image,” “the practice of embalming the dead might turn out to be a fundamental factor in their creation” (p. 4). However, in Deep Nostalgia, the combination of photography with algorithmic processes does not aim to embalm the dead, but to animate them. In fact, Deep Nostalgia can be seen as a new kind of postmortem photography. While 19th-century postmortem photography denied death by creating the illusion that the deceased in the images were merely sleeping (Ruby, 1999), Deep Nostalgia denies death through the illusion of wakeful people.

This animation of the dead is an inherently temporal operation: “Historical photos provide us with a tangible link to our past. Seeing our ancestors’ faces come to life through video reenactment deepens our connection to our family history and is simply breathtaking” (D-ID, 2022). This statement by MyHeritage’s CEO, as well as the term “nostalgia” in the technology’s name, harness history to an affective matrix in the present that activates desire and loss. But the relations between present and past are also manifested in Deep Nostalgia’s “component” media as well. As a hybrid of photography and moving images, Deep Nostalgia mobilizes the temporal differences between these technologies, especially regarding the cultural perception of movement as a representation of life. While the photograph was associated with an absolute past, an instantaneous “that-has-been” (Barthes, 1981), frozen and detached from its scene of occurrence, moving images (i.e. film) became associated with continuous lifelike unfolding and temporal progression (Bazin, 1960; Deleuze, 1986). Temporal experience is understood to be structured differently by these media, and the absence of movement in photography is associated with the absence of life. ⁵

Deep Nostalgia reaffirms this dichotomy between stillness-death versus motion-life. Its animations offer a visible performance, through the body of the transformed face, of the transition from stasis and death to the life of the moving image. The very term “animation,” from the Latin anima meaning “soul” and “life,” expresses this transformation, as do the antique origins (in, for instance, the Greek myth of Pygmalion and Galatea) of the movement-as-life motif. Recently, the contemporary salience of animation has become an important theme in cultural research. Animation has been promoted as a key phenomenon in digital culture, not only because of its centrality in computer-based imaging techniques, but also as a dominant medium of the 20th and 21st centuries, designed for the purpose of “making-live” (Levitt, 2014). Moreover, animation has been mobilized as a concept to designate new kinds of performative action for computer-mediated self-identity (Silvio, 2010), through which humanity is not measured by biological and physical appearance but by how the subject is technologically dispersed, controlled, and encountered in virtual worlds and digital networks (Ehrlich, 2021; Gershon, 2015). Yet animation, as the self-activating bestowal of movement, is also relevant to shifts in social media interfaces, notably the increasing ubiquity of short-form and looped videos as formats in a dynamic social media environment that is perpetually being put into motion. Perhaps emblematic of this shift are the advent of “live photos”—pictures that move because they are really two-second videos—as the default mode of “photography” on certain smartphones, and the rise of the “autoplay” function for embedded videos.

Deep Nostalgia’s performance of animation resonates with these tendencies toward heightened movement in platform formats. It also mobilizes shifting expressions, such as smiles and especially the illusion of eye contact with the viewer, to establish the impression of responsive, non-verbal communication (Goffman, 1967), a performance that simulates interaction (Silvio, 2010) between co-present individuals. This simulation of interaction is fostered by a technical characteristic of the system: photographs uploaded to Deep Nostalgia are automatically cropped, producing “passport” proportion close-ups around the faces in the image to be animated (a photograph containing several faces can therefore form the basis for multiple animations). This combination of expressive facial movement, interactive visual cues (such as eye contact), and the intimacy of the close-up turns the viewing of one’s ancestors into an intimate responsive encounter (Bateman et al., 2017). In the words of a user on MyHeritage’s (2021b) Facebook page (see Figure 3):

I look at them and as I see them looking “around” it’s as if they are looking at you and your surroundings and seeing how much things have changed, with the ones I’ve done I want to introduce myself and our relationships. (Shuman, 2021b)

Figure 3.

Comments from Facebook.

Deep Nostalgia’s harnessing of these media conventions—movement as liveness and “para-social interaction” (Horton and Richard Wohl, 1956)—produces a shared framework of space-time with the portrayed person.

“As if”: magic, imagination, and the subjunctive algorithmic image

Materializing the heart’s desire to resurrect those no longer alive requires cognitive flexibility and imagination. More specifically, the desire to see the impossible, and the technological wish-fulfillment of animating the dead, imbues Deep Nostalgia with magic. The analogy to magic is apposite to algorithmic processes in general: Gillespie (2016: 19), citing MacCormick (2013), refers to algorithms as “tricks of the trade,” comparing them to the idea of “tricks” as magicians understand the term, since algorithms are a set of instructions in programmable steps that help engineers (“magicians”) achieve a desired outcome quickly, but which are concealed from non-specialist viewers. Deep Nostalgia performs magic in several ways, beginning with MyHeritage’s textual and visual rhetoric. According to MyHeritage (2022a), “the technology that animates faces in photos looks like magic” and “some people love the Deep Nostalgia™ feature and consider it magical”; furthermore, the visual interface of the system itself incorporates a motif directly evoking magic shows: after the photograph is uploaded to Deep Nostalgia, an icon of a magic wand hovers over the image as the system processes it.

This textual and visual-operational rhetoric not only reinforces technology’s historical association with magic (Gell, 1988) but emphasizes its imbrication with visual media. During (2004) notes that in the 19th-century magic became a business-driven entertainment industry that drew energy from popular science and spiritualism, and whose cultural significance extended through technologies of optical illusion. More specifically, Barthes (1981) described photography as an “emanation of past reality: a magic, not an art” (p. 88, italics in original), while a key technological device underpinning animation and cinema is the magic lantern (Lanterna Magica). Through Deep Nostalgia, the historical connection of magic to media technologies is extended to AI processes. Indeed, Deep Nostalgia’s animations resemble the moving pictures and photographs in the wizarding world of Harry Potter.

Unlike magic tricks, however, Deep Nostalgia does not require that the “backstage” operations producing magical effects are hidden from view. It is different, therefore, to the orientation of users that Bucher (2018) calls the “algorithmic imaginary”, since its operative processes are—as noted earlier—clearly explained, and there is no mystery about how the algorithmic “magic” is created. In fact, the framework of magical belief established by Deep Nostalgia even seems to overcome obvious technical failures such as glitches (e.g. Figure 4). Even though such glitches can draw attention to the medium’s imperfection (Menkman, 2011), users remain fascinated and moved by the technological resurrection (see Figure 3).

Figure 4.

An example of a “glitch” in a Deep Nostalgia animation: On the left, a “successful” application of the driver video; on the right, distortion of the face.

Deep Nostalgia’s ability to elicit enchanted acceptance of the impossible as “real” is an achievement of imagination. Slater’s (1995) observations regarding photography are apposite:

In the form of magical representation, the technical accomplishment of realism is the basis not of knowledge of the world but of the production of simulated worlds, worlds which we can pleasurably inhabit through the very opposite of the modern attitude—by suspending our disbelief. (p. 232, italics in original)

Again, Deep Nostalgia extends this suspension to AI processes: “Takes my breath away” as one user wrote on Twitter, “this is my grandfather who died when I was eight. @MyHeritage brought him back to life. Absolutely crazy” (Hawran, 2021). ⁶

Deep Nostalgia’s imaginative and emotional power brings to the fore the conditional “tense” of image-creation and the viewing process. It invites a suspension of disbelief that prioritizes the subjunctive as a visual force; what could be takes priority over what is. This subjunctive visual power is explicitly expressed with reference to media themselves, allowing Deep Nostalgia to simulate how “the person in your photo would have moved and looked if they were captured on video” (MyHeritage, 2022a, italics added). It expresses a desire for a resurrection that is also a technologically anchored counterfactual hypothesis.

Zelizer’s (2010) analysis of “about to die” images is relevant here, as discussed earlier. In stark contrast to “about to die” images, however, Deep Nostalgia technologically performs the desire to see individuals who have died “as if” they were alive, “as if they’re right in front of us” (MyHeritage, 2021c). It turns every old photographic portrait into a potential “about to live” image. Unlike the photographs Zelizer discusses, Deep Nostalgia actively fleshes out the possibility of life through technological animation—movement, simulated interaction, and presence. This is precisely the significance of the “algorithmic as if”: the conjectural work of imagination is visually materialized through a system of algorithmic imagining, a human-computational matrix of simulation and desire.

Endlessness liveness: emotional and uncanny intersections

For all its emotional and simulative power, the subjunctive image of Deep Nostalgia is fundamentally artificial and anachronistic. It is not a static photograph of a dead individual from the past, nor a chronological moving image such as film, but rather a looped sequence. While both photography and film are compatible with chronological time, Deep Nostalgia disrupts traditional temporal linearity. The mechanical artificiality of Deep Nostalgia is emphasized by its looped movement, making the image seem closer to a photographed GIF. The animations operate under the principle of recursive temporality, promoting a technologized “desire for endlessness” (Hoelzl and Marie, 2015) in our visualizations of life. Duration is disconnected from development; the portrait moves only to repeat: the facial expressions recur over and over again in an incessant cycle. The looped movement does not enliven the dead but rather makes them undead, stuck in Limbo. “There is no death in animation,” Levitt claims, “because there is no being—no existence—to begin with. There are no necessary limiting features, no essential finitude—everything is shadowed by its possible metamorphosis, erasure, and resurrection—and there is thus no ontology” (Levitt, 2014: 128).

This unsettling temporality partially informs the unease expressed by some users about Deep Nostalgia, despite its popularity. A day after the service was launched, critiques appeared in various news articles, blogs, and comments on social media, identifying Deep Nostalgia as creepy, uncanny, disturbing, and even eerie. In the first week after the launch, discussions of Deep Nostalgia’s uncanniness spread from the niche spaces of technology blogs and forums to mainstream news platforms such as BBC News (Wakefield, 2021), The Guardian (Hern, 2021), and The Independent (Cuthbertson, 2021) (Figure 5). Even MyHeritage (2022a) noted the potentially uncomfortable feelings that the technology could elicit, observing that “some people love the Deep Nostalgia™ feature and consider it magical, while others find it creepy and dislike it”.

Figure 5.

Headlines from news articles: (a) BBC News, 26.2.2021, (b) The Guardian, 1.3.2021 and (c) The Independent, 1.3.2021.

This dichotomy in the public reaction between enthusiasm and creepiness regarding Deep Nostalgia’s animations invokes Freud’s (1971 [1919]) uncanny, which constitutes something that is frightening precisely because it is not known, and yet familiar. In this case, the portrait and depicted face are familiar, but the repetitive looped movement is not. As Fetveit (2018) notes of the GIF format generally, Freud’s (1971 [1919]) list of examples of uncanny phenomena are strikingly relevant since they “excite in the spectator the feeling that automatic, mechanical processes are at work, concealed beneath the ordinary appearance of animation” (p. 226, cited in Fetveit, 2018: 49). The mechanical recurrence of facial expressions creates, in Fetveit’s (2018) words, an “uncanny mediality”—an uncanniness specifically arising from the combination of photographic indexicality and looped movement. Deep Nostalgia gives presence to the restless dead; or rather, to new creatures, neither dead nor alive but perpetually in motion.

The uncanny feelings that stem from this combination of the mechanical and the natural are also explicitly related to the “uncanny valley,” conceptualized by Masahiro Mori (2012 [1970]) in the 1970s. A robotics researcher, Mori proposed a ratio between an object’s degree of similarity to a human being and emotional responses to the object: when the object not only looks like a human but even moves like one, it produces an aversive reaction, uneasiness, or even fear, among some human observers. As a headline in The Independent noted: “Deep Nostalgia AI brings old photos to life with ‘creepy’ accuracy” (Cuthbertson, 2021). Deep Nostalgia’s hyper-realistic level of “accuracy” materializes the “uncanny valley.”

This uncanniness does not necessarily cancel out the power of the emotional connection with those who have been reanimated. Both can be felt simultaneously. As one user put it: “I’m crying MyHeritage app, #DeepNostalgia feature restores old photos and can animate them to look so live like. This is kind of creepy, but i needed to see. #RIPDaddy” (the rich auntie, 2021). The uncanny feelings constitute a sort of “emotional glitch,” an affective signal that the boundaries between the living and the dead have been artificially crossed, and yet remain. However, both the positive emotional connections and the feelings of uncanniness are based on the same crucial semiotic operation: the photographed image becomes transparent; the image and the dead individual’s face are treated as one; the photograph becomes a proxy (Mulvin, 2021), standing in for the actual person depicted. To cite one typical response from a user: “this is my grandfather” (Hawran, 2021; italics added). Deep Nostalgia’s hybrid indexicalities are thereby structured hierarchically; the photograph’s status as a representation made “infrastructurally” invisible by the driver video that is applied to it, enabling the dead (and not merely their images) to penetrate the world of the living.

Conclusion: Deep Nostalgia and the limit-situation

Algorithmic technologies will continue to evolve in addressing and enhancing human requirements and desires, alongside endeavors to create an “algorithmic as if” aimed at transcending existential constraints due to their profitability: the perpetual cycle of life and death remains unaltered, guaranteeing commercial “demand.” Nevertheless, we have argued, such technological initiatives unveil limit-situations, undermining the aspiration for a perfectly seamless convergence between humans and machines.

Our case study, Deep Nostalgia, offers digital resurrection of the depicted dead. It is an explicit manifestation of automated media’s “ability to fulfill desire in advance, before it manifests itself: that is, to shrink to nothing the temporality of any unfulfilled want” (Andrejevic, 2020: 8). Deep Nostalgia lets us see family members come back to life without us knowing that we want it. Deep Nostalgia materializes what this article has called the “algorithmic as if,” an aspect of algorithmic culture that designates the constrained, biased, historically constructed forces of imagination embedded at the core of computational systems, and the economic, political, and cultural power structures they reproduce. Yet for all its potential commercial applications (such as marketing MyHeritage’s other products) and its ideological power, it also addresses and manifests deep human longings, with potentially profound subjective resonances for viewers. As a recently widowed user explained in a Facebook comment: “I did the picture of my husband who died 4 years ago. It makes me so happy to see him smile again” (MyHeritage, 2021b). Deep Nostalgia can thus serve to articulate processes of mourning through the ‘‘algorithmic as if,” manifesting for grieving individuals the (impossible) return to presence—and to the present tense—of those whose existence is definitively past.

Deep Nostalgia does more than act as a vehicle of memory that constructs relationships between the present and the past (Pickering and Keightley, 2006). Rather, it emphasizes how digital media can expand human being and, accordingly, the definition and experience of limit-situations—situations, such as suffering and death, that provoke awareness of the boundaries of human life (Jaspers, 1970 [1932]). Human beings attempt to ignore their own death, that future which must occur, in order to survive mentally and to prevent themselves from suffering paralyzing fear that blocks their existence. Jaspers defined death as a limit-situation, emphasizing not the fact of extinction per se, but rather the consciousness of it as an utter negation that threatens life, as the complete non-existence of the self. Crucially, however, this consciousness is not a denial or escape from death: it makes possible authentic human existence. In the words of Lagerkvist and Andersson (2017), “the grand interruption or limit-situation is what actually generates ‘becomings.’ From the vantage point of the hurting, dying, and vulnerable existers, after the grand interruption, the salvific, mundane, haptic, and processual lifelines of the Internet are life-sustaining as well as life-changing, exemplifying a being-in-and-emerging-with digital technology” (p. 560).

Some AI technologies are difficult to accept, especially when they require us to imagine a different future from the familiar present (Markham, 2020). This lack of consensus is even more conspicuous when it concerns life and death issues, and when the materialization of our hearts’ desires provokes ethical dilemmas and controversies. ⁷ However, “people and algorithms will continue to become ever more entangled in cultural production, both on and offline” (Gillespie, 2014: 183–184), and thus cultural aspects also need to be explored in existential terms. According to Lagerkvist (2022), limit-situations have intensified in the context of contemporary global crises to the extent that is possible to characterize our era itself as a collective limit-situation (instantiated recently, at the time of Deep Nostalgia’s launch, in the construction and experience of Covid-19 as a “global crisis” [Frosh and Georgiou, 2022]). Moreover, digital media extensively and profoundly “co-shape the material and symbolic worlds we inhabit” (Lagerkvist, 2022: 20), creating new digital limit-situations: “while existential media pertain to aspects of being human that have been of concern for humans for centuries, the digital limit situation also encompasses new and binding realities” (p. 20).

Hence a technology such as Deep Nostalgia exemplifies not only how algorithmic culture is being produced, but how photography-based AI software is employed as an “existential medium” (Lagerkvist, 2017, 2022). The unsettling, composite temporality of Deep Nostalgia, combining stillness and movement to materialize the subjunctive “as if” of loss-laden desire, can intensify the limit-situation as a moment of (self) encounter. It evokes a limit-situation by letting us watch our ancestors come technologically to life, providing self-reference for one’s own ephemeral human existence, and becoming a reflection of our presence and our limited lifetime. In addition, however, Deep Nostalgia provides access to specifically computational interventions and constructions of limit-situations, such as the mediation of life and the digital-visual desire to overcome death. Experienced as uncanny or “creepy,” such computational interventions can themselves be perceived as inherently limited and incomplete, as inevitably insufficient “transcendent media” (Lagerkvist, 2022) by virtue of their very mediality. No medium provides perfect transcendence of the “ultimate boundaries”; the feeling of uncanniness is a “relating across the threshold” (p. 172) between the living and the dead that can reinforce consciousness of the threshold itself. The extension of our existential engagement through new visual and computational technologies that seek to alter human chronology thus reminds us of the very limitations that constitute our being per se, and our digital existence in particular. Deep Nostalgia shows us the “algorithmic as if”: newly acquired algorithmic techniques that manifest the socially and technically orchestrated desire to overcome death, and our consciousness of its utter impossibility.