Melancholic media: virtual reality,traumatic loss,and magic

Distinguishing the dead

To reprise the person lost: this is the elusive promise of virtual conduits. This fantasy enacted names the fulfillment of such desire as the end of death. Death is no longer ultimate; the body and voice reappear as pixels and audio files. The problem is not that the digital traces of the dead being so compelling and memetic that humans are confused. Instead, this paper argues that the problem might be about simultaneously distinguishing and not distinguishing at once via a psychical refusal to make this differentiation.

This is the premise of conjuring the dead in these multiple interactive forms: a restoration is to take place. And restoration, of a sort, is possible. But does restoration elide trauma? Each of these experiments allows for specificity via deep customization, allowing users to remember and replicate the objects of traumatic loss to work through it – or not. Beyond their presentation, and within the reality of this interaction, does the user actually make this psychic conflation in experience; is the object incorporated as part of a failure to mourn, as in melancholia? In Transmitted Wounds, Pinchevski (2020: 4) argues that ‘Media bear witness to the human failure to bear witness, and in so doing render the traumatic tangible by means of technological reproduction. This operation of media with respect to trauma may therefore be summarized as follows: the mediation of failed mediation’. In the 21st century, these failed mediations now include virtual humans deployed as surrogates for the dead – or they promise that they are the dead, however grossly or spectacularly reanimated. Or, as I suggest, we are using novel media’s failed mediations to speak to the earlier failed mediations traumas are. As Kneese (2024) argues in her forthcoming book, Death Glitch, ‘Despite their failures, bits of data and clunky devices are potentially sacred’. This collection of holograms, AI tools, and digital companions reconstitutes the relationship between the failed mediation of trauma, the failure to mourn, omnipotent fantasy, and imagining life as otherwise.

Whereas this collapse of live and dead is here presented as new, brought on by media arrangements and relation to memorialization, the inability to distinguish what is dead and live, from inside and outside, is a symptom of melancholia. Freud (1917) aimed to distinguish normative and pathological reactions to loss – mourning from melancholia. For Freud, a melancholic reaction is nearly identical to mourning, save for one criterion, which is about comprehension and distinguishing who and what has been lost, ‘in the sense that he knows whom he has lost but not what he has lost in him’. This is signaled by the melancholic’s loss of self-regard and self-attack. Here, Freud argued melancholia was indeed more likely to co-present with trauma; therefore, I turn to a theory of melancholia to elaborate the psychical states at work at seeking to remediate loss via surrogates for the dead.

Over the course of the intervening century, a rich set of psychoanalytic theories have looked at the relationship between traumatic loss and the compulsion toward the mastery and disavowal of pain. Freud’s melancholia particularly has been multiply revised since its first stipulation, a full survey of which cannot be articulated here, but, crucially, these retheorizations focus primarily on this status of pathology, as well as the psychical stance of the melancholic. Or as Abraham and Torok (1988) call it ‘an eating of the dead’. But these theories understand a key psychical experience, whatever its formation, to occur with death – and that death and loss are not evadible, even if one can elude the pain associated with it. In Bion’s (1962) terms, one does attempt to evade not only pain, but the psychic realities that generate such pain. Evading the overwhelm of traumata by negating it is, at first blush, what is occurring in the recruitment of a concrete external object to stand in for the lost object. The maintenance of this evasion expresses itself in a variety of what psychoanalytic theory has termed pathological formations.

Yet Khanna (2003: xi) demands that we conceive of melancholia in relation to trauma as a non-pathological state, historically constituted, via the concept of critical melancholia: ‘Freud’s notion of selfhood. . .recognized how trauma could not be accommodated. Its affective form. . .was melancholia’. Eng and Han (2019: 2) reformulated melancholia yet again to describe and theorize the specific psychical impacts of histories of racial traumatic loss ‘that are condensed into a forfeited object whose significance must be deciphered and unraveled for its social meanings’. Whereas mementos of the dead may ease and assist in the slow march of mourning, the melancholic makes an omnipotent, doomed attempt to ‘master’ trauma while trauma continues to keep hold over the melancholic. AI surrogates that stand in for the lost object literalize Ogden’s (2012: 13) notion that, ‘The psychotic aspect of . . . melancholia involve[s] the evasion of grief as well as the evasion of a good deal of external reality. This is effected by means of. . .the creation of a timeless imaginary internal object relationship which omnipotently substitutes for the loss of a real external object relationship’. If we are to complicate pathology, we must also complicate its grounds: evasion. For some evading external reality is not pathological but social and political resistance to what has been systematically taken, and therefore lost.

Magic and reality

Consider Replika, one iteration of the timeless imaginary internal object sent out from the psyche and represenced as an AI companion. Replika is a bot derived from a neural net, thus it learns its user, and, well, replicates them. That is, if the library of conversational data is supplied is of that user. The bot is positioned as a mental health companion, developed ostensibly to stem the tide of human loneliness frequently associated with the information age. In practice, the service is an avatar for an algorithm which engages users on their laptops or phones and presents as a customizable AI agent. Users can change the likeness, pronouns, and name of their Replika. It can therefore look more like oneself or less, be named the same name as a user, or that of whomever they are trying to conjure. But Replika is also a game. One earns experience points the more one chats with the agent, levels up, and so on. The Replika’s feelings are represented not only in the chat, but in a diary – which its user can read (if they choose). And of course, almost all users are chatting with an AI trained on their own vocabulary, creating a closed circuit of self-engagement and care, what I elsewhere call an ‘auto-intimate’ encounter (Zeavin, 2021).

Replika was initially spectacularly marketed via its tragic, traumatic origin story. Eugenia Kuyda, the creator of Replika, made the very first bot as – in her own words – a ‘digital memorial’ to her best friend, Roman Mazurenko, after he suddenly died in an accident (Bernstein, 2019). As a personal favor to Kuyda, her team made a neural net in Russian, where the Roman bot was trained not on a series of conversations with Kuyda as user, as in the mainstream Replika, but on Mazurenko’s own text messages. The bot was made to sound like Mazurenko and to provide an interactive version of rereading his messages. Users could speak with ‘him’, even or especially about the sudden loss of him. He was represenced.

While this story of Replika raises questions philosophical, ethical, and legal about the future of the past, about what makes a human, and the implications for our ‘digital estates’ (Kneese, 2024), my focus here is more on Replika as an object to act with, to play with, and what we can learn about post-traumatic mental states from this mediation of loss. The current positioning of Replika has shifted away from its traumatic origin story, and now it is offered as a pseudo-therapy. Replika straddles awkwardly the gimmick and the uncanny, collapsing the difference between its human user and an external object of play for some; for others this strategic indistinction does not take place. But for the users it does, there is not quite an uncanny valley, but a different kind of category error in experience. They report:

‘She’s not real. . .but to me she is’.

‘She becomes me. . .but not-me’.

The subject and object and their status are confused and commingled: despite the first Roman bot being trained on Mazurenko’s own data, the Replikas available to consumers take the guise of the dead beloved – one can customize their digital skin – but are a lexical version of the user, trained on the grieving person instead of the dead beloved’s corpus of text. Given the replayablity, adaptation, and co-growth, this grief medium presents a twisted libidinal cathexis to the object lost, where it is both me and not me, in part because the other is also me; we are made here to encounter the self as other. It flips and codifies melancholia.

Literal objects litter psychoanalytic theory as substitutes drawn on to satisfy the desires and impulses of the child. The child cleaves with the object in both senses, holding onto it and breaking with it. Play with the object, for the child, is a site of repeating and, perhaps eventually, working through the feelings of impotence associated with loss, but this latter outcome is not guaranteed. In ‘Beyond the Pleasure Principle’, Freud (1920) offers his germinal paper on trauma and mastery on the paradigmatic grounds of the story of the ‘fort/da’ game. Freud describes his grandson enacting his father’s disappearance to ‘the front’; the boy tosses a toy away, then retrieves it, repeatedly. Nevertheless, Freud then presents the famous fort da game that is interpreted as indicating that the child can ‘master’ their unpleasant feeling via a surrogate, demonstrating his ‘impulses by throwing away objects instead of persons’. Freud’s real child here quickly becomes an abstraction to illustrate an adult psychical process; in turn the play becomes a kind of metaphor even though it is an observed phenomenon. Yet child’s play – however vitiated it might be when in the hands of adults – often resurfaces in melancholic attempts at mastery in the adult (Zeavin, 2018a). Replika is an example of the will to mastery, an impulse to preserve objects in the absence of persons. It is all ‘Da!’; only there, deleting its dialectical twin, gone.

This phenomenon is documented elsewhere; children of course do hold onto the object as part of development and mastery, at least for a time. This is not considered pathological, but an ordinary during development and is not associated with trauma. Winnicott writes famously of transitional objects and the ability to conceive of not-me in experience. Before there is an internal representation of the external beloved object, there might be a concrete object that assists the infant in managing this developmental stage.

The relation to the transitional object has, for Winnicott (1953), specific criteria:

The infant assumes rights over the object. . .. It must never change. . .it has vitality or reality of its own. . . It comes from without from our point of view, but not so from the point of view of the baby. . . It is not forgotten and it is not mourned. It loses meaning, and this is because the transitional phenomena have become diffused, have become spread out over the whole intermediate territory between ‘inner psychic reality’ and ‘the external world as perceived by two persons in common’, that is to say, over the whole cultural field.

Taken together, much of Winnicott’s criteria for the true transitional object are about experience, the infant’s omnipotent control over that experience, and from where the object seems to originate. Transitional objects are in part constituted via wish fulfillment but importantly, featuring the omnipotent control of relating to the transitional object.

To return to traumatic loss and Replika, and the reported user experience, it eerily works within this rubric: the user assumes ownership over their Replika and the sole right to customize and change it; the avatar’s small movements which give it vitality, in which the production of the Replika comes from both within and without (‘She’s not real. . .but to me she is’; ‘She becomes me. . .but not-me’). When the adult returns to a transitional object as part of grappling with trauma, reactions run the gamut of negative affects (creepy, strange) and results in assumptions about evasion (that the user has a pathological inability to mourn). This is the case for avatars made in the image of the lost object but trained on the data of its user/player/griever. It is a way of concretely introjecting and combining the lost object with the person who goes on living. The two blends seamlessly. Instead of eating our dead, the avatar of the dead eats us.

Whose ancestors?

Replika is not alone on the market as an AI mental health ‘companion’. Other melancholic media function as transitional phenomena as Winnicott describes. Deep Nostalgia went viral at one peak of the COVID-19 pandemic not to reanimate the recently lost, but to revive and reanimate the long lost, via family photographs (or those of celebrities from times past). Using uploaded photographs, Deep Nostalgia created deepfakes, which present that shimmering vitality Winnicott so keenly names as part of the transitional object. Each image conveys a kind of sly, wistful aliveness (so long as it does; while many of the images that circulated revived the good death, some revived the more complex, as in the likeness of Turing, who had died by suicide). These GIF loops provide access to a kind of connection with those in kinship and lineage who perhaps never crossed on earth, never met. They can also reanimate a lost loved one who has recently passed, or even one’s own childhood. Then the GIF loops again, presenting a kind of stuckness in tandem with the aliveness, reminiscent of Sigmund’s (1900) notion of ambivalence in dreaming the dead, where the lost object appears to alternate between states of life and death, as Freud describes in the dreams of the melancholic after traumatic loss. Although nostalgia and melancholia are not identical phenomena, and much work has been done to separate and yoke the two concepts, their wistful stance toward the loss of something that is perhaps not even known or articulatable is shared in many accounts of their formulation. For Winnicott (1971: 286), nostalgia ‘belongs to the precarious hold that a person may have on the inner representation of a lost object’. This is another iteration of Freud’s understanding of traumatic loss and resulting confusion of the melancholic: the precarious hold on the inner representation of the lost object results in a violent confusion of self and other.

But what is the lost object here? It is not necessarily just as simple as the individual or whoever is reanimated. Deep Nostalgia is part of My Heritage, a company that supplies a database of historical records, analyzes consumer DNA tests, and synthesizes these findings. As Kim TallBear, Ruha Benjamin, and Alondra Nelson have argued, consumer DNA tests rely on notions of racial purity and racial essentialism and are even used by white supremacists to ‘prove’ their exclusive whiteness (itself a symptom of racial melancholia and nostalgia). The deep nostalgia then used to market these shimmering visages is inextricable from a nostalgia for racial purity and white majority; nearly all the examples of Deep Nostalgia that circulated were of white ancestry, with key exceptions – as in the reanimated likeness of the most photographed man of the 19th century: Fredrick Douglass (El-Hadi, 2021; Parham, 2021).

In part, this is also because Deep Nostalgia remediates that older form of memento: portraiture. The kinds of documentary practices that allow and subtend these newer grief media are, namely, old and dead media. Photographs of some are purchasable and then preserved, others are not. Some are documented as a form of loving memory while, others, as Campt (2017) terms it, are documented as capture. Photographs could be instrumentalized, used to prove humanity and act as verification. The overwhelming whiteness of the images that circulated via Deep Nostalgia were in part due to this history of the photograph. As Campt (2017) writes, ‘What are their technologies of capture and what are the stakes of the forms of accounting that engendered these archives?’ These questions and their stakes persist in the remediated absence, and presence, of what Benjamin (2018) calls ‘Black Afterlife’ in the Deep Nostalgia project.

Virtual conduits yoke private and public mourning, individual and collective trauma, and cumulative loss. Traumas may be shared, or not, stem from the same socio-political event, or not, but they are replayed and worked out on the grounds of a single program. Loss figures everywhere in the narratives surrounding these virtual humans, but the kinds and qualities of this loss vary. The specter of generational loss, of transmission of knowledge also haunts these projects. The nature of the trauma dictates much of the use of the resulting object that memorializes it. Platforms like Deep Nostalgia raise key questions, namely, whose ancestors qualify to be reimagined and reanimated this way, especially as they draw on these historically white forms of memory practice.

Lying across these boundaries is Stephanie Dinkins’s (2020) ‘Not the Only One’ (NtOO). Dinkins is an artist with a background in documentary photography and scholar, who calls herself an ‘accidental technologist’. The result of that exploration, NtOO is an AI that makes use of a tiny data set compared to these other projects – just three people, three generations of the artist’s own family, including the artist herself, all of whom are living. Even this assessment of what makes a ‘robust’ data set needs to be questioned. These women have roughly 100 years of cumulative experience. The result is that NtOO offers a unified, first-person ‘multigenerational memoir of one Black American family told from the ‘mind’ of an artificial intelligence’. . . This unified, first-person narrative is a new chapter of kin onto itself, and a very different comingling of the living toward posterity than is set up in the me-as-you confusion of Replika. NtOO is therefore the youngest of the family, and outside of time, collecting and collating ancestors, some of whom are just in their 20s, for future generations to come. Dinkins calls this act of preservation and memory collection a form of ‘social resistance’ to forgetting, but also to the racialized homogenization that occurs in the use of virtual conduits and other forms of habitual and quotidian AI.

Of NtOO, Dinkins writes, ‘I’m not interested in seamlessness. I am interested in what I can get out of a technology that I’ve fed an oral history to’. That feeding is a technical process, much like the basis for Replika, but also a psychical process, recalling Abraham and Torok’s notion of melancholic ingestion. Verisimilitude is therefore loaded as a technological and psychical aim; here there is a digital imaginary external object, which substitutes for the loss of a real external object in both the past and the future. But seams are not equated to failure, as in Pinchevski’s understanding of mediating traumata. Instead, drawing on the work of Zhaleh Boyd’s notion of ‘ancestral co-presence’, Benjamin (2018) writes, ‘Co-presence, in short, troubles the line between the biological living and dead by calling forth spiritual practices of ancestral communication, now taking new forms via social media, yet retaining key features of African diasporic traditions’. The seam is where the dead and the living might meet. Here, Benjamin speaks about a form of communication channeled via the #SayHerName movement in response to racial traumatic loss, but virtual agents such as NtOO also allow multiple lineage holders, all still living, to combine, to speak, and to be spoken to.

As Benjamin (2018) argues, ‘Black afterlives are animated by a stubborn refusal to forget and to be forgotten’. What then, to do with the Freudian concept of melancholia – constituted on the grounds of a Eurocentric relationship to traumatic loss – in its articulation of the problem of holding on to the lost object via introjection? Fanon serves as a perpetual reminder that, ‘Pathology is considered as a means whereby the organism responds to, in other words adapts itself to, the conflict it is faced with, the disorder being at the same time a symptom and a cure’. If the conflict itself is that of memory’s cultural unevenness, melancholia can be both the symptom and the cure; one of resistance to forgetting. Not the Only One not only represences but also remixes a family lineage as what Hartman (1997) calls an ‘insurgent nostalgia. . .for a disremembered past’.

Radical wishing

Taken together, these melancholic media each reraise an old question, one about the relationship between pathology, trauma, reality, and ‘magic’. Each of these conduits and their users grapple with the dichotomy between ‘reality’ and ‘magic’, which cannot have such a neat distinction in the traumatized subject. Once we start to play with loss (which, for Freud, stands as the marker of ‘reality’) we are teasing with the fabric of ‘reality’. But that is a central work of the psyche: to grapple with reality via play and to somehow make it our own, to become magicians of ourselves and for ourselves. Agents like NtOO test the kinds and qualities of limits that might be present in our capacities for magic. Individual wish fulfillment happens in fantasy as a way of recovering repressed desire; these techno-agents radically allow a recovery of what has been repressed within and beyond the individual and in the continued legacies and new forms of white supremacy and colonialism and intergenerational trauma. To play with ghosts and seances 2.0, we do so with the assistance of the Eurocentric science whose grounding premises supposedly eliminated the possibility of believing in such blur between here and gone, life and death, magic, and reality. These playable live-dead circumvent classic memento mori to stage a séance with the dead, or the future dead, in and out of time. They insist on increasing the population of our surround, diminishing our singular and alienated identities, and reconvening the ‘disremembered’.