When the dead become avatars for the living they are always present, but as on-demand services rather than messy intrusions, as applications to open or close.
Last year your mother died–now, she’s smiling blankly on your screen, reminiscing about the night of your birth. It’s her voice but also not–vacant somehow, more pristine. Or: weeks after losing your partner, you send a text into the void: “Are you there? Do you miss me?” A message from the afterlife comes through: “of course i am babe,” in your partner’s quintessential lowercase. From popular to academic literature, meditations on Digital Afterlife Services (DAS) often begin in the speculative tense, projecting readers into a future realm where lost loved ones are reborn as chatbots or uncanny avatars. Others draw on speculative media that already exists, as in the oft-referenced Black Mirror episode “Be Right Back,” which foretells DAS’ capabilities to remain in touch with the dead. The question of “what if?” haunts DAS for good reason, reconciling promises of healing hyperrealism with the potentially dystopian exploitation of both living and dead.
These imagined futures are governed by real corporations such as Project December and HereAfter AI, two recent start-ups in a longer string of DAS that use Generative AI to create post-mortem avatars that talk back to the living in convincingly intimate ways. Project December, a text-based service founded by Jason Rohrer in 2020, charges customers $10 a month to create post-mortem chatbots of loved ones. Soliciting details about the dead from living users, the platform asks subscribers “for remarkably little information”—only “one small paragraph…and one quote”—to train a chatbot that simulates the dead. Other DAS cater to mortals-in-waiting rather than mourners-to-be. HereAfter, founded in 2019 and spotlighted in The New York Times and The Wall Street Journal, sells customers the curation of their own legacy while they’re still living, vowing to preserve “Your stories and voice. Forever.” By completing interviews with a virtual biographer, recording memories in response to story prompts, and uploading precious photos, HereAfter users train their own post-mortem avatars and optimize commemoration for prospective mourners, who can “instantly access stories and photos just by asking. “Hello, what part of my life would you like to hear about now?” an avatar of “Grandpa” asks their grandson in an eerily upbeat marketing video.
DAS invite speculation-as-necessity; we don’t know exactly how (or how well) these products work, nor the extent to which our own viscera will be stowed, bought, and sold on the digital marketplace. These questions are also temporal, as “time will tell” how these products unfold and also who may be rendered more or less precarious by DAS technologies. DAS are distinctly in media res, which means the speculative is the only tense we have, one with paranoid and reparative potential. In one way or another, attempts to make sense of DAS indulge a version of writer Kim Stanley Robinson’s maxim: “We’re in the science fiction novel now, which we are all co-writing together.”
Amidst these crucial but still-unfolding projections, we approach DAS as inherently “mid” technologies: mid as temporally in-between, and also for how they fall short both affectively and logistically. DAS function in a liminal space between the dead and the living, assuming the role of pre-mortem archivist and post-mortem medium. In this way, they situate users in a variety of purgatory-like states, from the control over our own mortality to the ongoing suspension of lost loved ones in the now. The development of DAS is unfolding in the middle of a present polycrisis of collective lamentation that renders their promises particularly alluring. In a time of mass grief characterized by financial precarity, ecological collapse, pandemics and multiple ongoing genocides, we are always experiencing, processing, and–for those with the privilege to do so, compartmentalizing–an ongoing state of mourning that feels both ambient and urgent.
DAS largely sell themselves as a tool for managing loss at an individual rather than collective scale and, therefore, appears to have distinct scopes and stakes from such polycrises. What does talking to Grandpa’s avatar have to do with climate change? As we suggest—from the middle of things, ourselves—DAS’ promises of individual agency over our own legacy and those of our loved ones pacify a helplessness with respect to collective crises. DAS sells the compartmentalization of personal grief amidst mounting collective grief, postponing catastrophe by building middle spaces that are in between acceptance and denial. The mass tendency to compartmentalize is bounded in new ways within DAS applications, which offer optimized, reliable pockets where users can go to grieve. When the dead become avatars for the living they are always present, but as on-demand services rather than messy intrusions, as applications to open or close.
Simultaneously, when we buy into the transformation of ourselves into post-mortem products, we streamline the stuff of our lives—images, lexicons, memories—into digital compartments. This phenomenon is distinct from death prevention; in a future where DAS is normalized, we know we will die, but we also know that a digital avatar can survive us. Among both pre-mortem archivists and post-mortem mediums, this proposed reality operates more clearly as a deferral of both lost life and lost agency. Services such as Hereafter borrow from existing modes of death preparation; guided by loved ones and end of life doulas, mortals-in-waiting are encouraged to sign wills and cite funeral wishes, share their passwords, and arrange future homes for pets. They differ by optimizing these tasks into collapsible interfaces shaped by user interaction, ushering death preparation into the era of the “for you” page. In an increasingly uncertain present tense, one which breeds powerlessness and speculation-as-necessity, DAS monetize the dream of suspending death’s finality while enabling its ongoing curation. They vow to nurture comfortable, reliable compartments for ourselves and others even in the afterlife.
Amidst ongoing precarity, the commercialization and commodification of DAS induces competition over grief as a theoretically finite but constantly compounding resource. We call this macabre variation of the attention economy lamentation economics, an already-unfolding network of production and distribution that we begin to trace briefly here, from the middle of things. As we see it, success in the lamentation economy market necessitates timely capture and exploitation of customer grief. In order for a DAS provider to efficiently capitalize on loss, it must be able to detect or anticipate emotionally vulnerable states of grief. Strategies for doing so include: scraping obituaries and linking these records to the social media networks of the deceased to identify potential customers, optimizing ad presentation timing through analysis of grieving patterns, and modeling when grieving customers are most likely to make emotionally motivated purchases (e.g., aligning advertisements with affectively potent events such as such as anniversaries, birthdays or death dates).
Potential usage of DAS as a transitional tool to help users navigate the grieving process and “move on” (delete the Grandpa avatar) is in direct conflict with the present growth model in commercial software. Within this system of lamentation economics, individual DAS providers are more likely to cultivate habitual usage by delivering experiences that capture and hold the users’ attention. Competition between DAS services may incentivize common patterns for vendor lock-in: keeping customers dependent on their own services by making the transfer to another DAS provider inconvenient and/or costly. For providers using a subscription model, the threat of losing a loved one once again – this time virtually – compounds DAS’ already insidious forms of coercion. The lamentation economy exploits feelings of guilt, shame, or regret by suggesting that consumer unwillingness or hesitation with respect to digital preservation is irresponsible, cruel, or a form of abandonment.
Such consumer manipulation is not limited to mourners, but also extends to legacy curators. Particularly vulnerable customers—such as those planning their estates, faced with terminal illness, or grappling with their own mortality—will increasingly be met with well-timed advertisements to “preserve yourself for the ones you leave behind.” At their most corrupt, DAS transform post-mortem preservation into a mode of social responsibility with the primary intention of preserving their consumer base. Users are called to share more data detailing the intimacies of their lives: more photos, more stories, more letters, more journal entries, more home video recordings in service of optimizing the avatar they leave the loved ones who survive them.
DAS’ promises belie the fact that these technologies currently do – and may continue to – remain kind of mid themselves. Despite claims of hyperrealism and premature praise featured on websites like Hereafter’s—“the next step in the human quest for immortality,” “the best tool to preserve…legacy,” “the best way to keep the stories of our loved ones alive”—DAS continue to fall short in its technological capabilities and their emotional promises. Competition to create the best quality models and overall user experience is a data-hungry enterprise with a potentially boundless appetite. Addressing data limitations in the lamentation economy could play out similarly to how it has proceeded in other machine learning applications: harvesting and pooling data from user interactions to further improve the “collective memory” of generalized foundation models for cross-application to other users. Presenting users with interactions derived from the data provided by a loved one mixed with data from myriad complete strangers raises a labyrinth of ethical and privacy concerns for both the living and the dead.
As early as 2007, the now defunct Alabama-based startup Intellitar promised to “give users the gift of immortality,” charging $25 a month for the chance to shape a post-mortem doppelganger. Intellitar’s short-lived database for afterlives employed a combination of self-reports, voice samples, and other digital remains to promise “10,000 people eternal digital life—then it died,” folding in 2012 due to an intellectual property dispute. Three years later, entrepreneur Marius Ursache announced the founding of Eternime, another DAS employing self-reports and an AI biographer. Despite an influx of personal emails and news outlet inquiries, Ursache’s small team paused operations early on for pragmatic reasons surrounding time, resources, and family obligations. After an attempted relaunch with new partners and revamped AI, Eternime closed its doors for good in 2020 without a formal announcement, mirroring the boarded-up fate of Intellitar.
Although not explicitly a form of DAS, the successful AI chatbot app Replika was famously trained on a chatbot that founder Eugenia Kuyda built to replicate close friend Roman Mazurenko, who died tragically in a car accident in 2015. The Replika app materialized around “an AI like Roman but one you build yourself” by “telling the story of your life every day in text format.” Now, Replika sells empathetic AI companions, including friends, partners, and mentors, who “see the world through your eyes.” A platform built on the stuff of grief has since expanded and narrowed radically in scope; recently, Replika became best known for its erotic roleplay features, fulfilling users’ sexual desires, as well as the subsequent and controversial removal of this feature. This decision, publicized in late 2023, prompted angry and bereft subscribers to deem the app mostly dead, further highlighting the potential transience of DAS within volatile markets and user preferences.
Assuming present popular trends continue, the mid-ness of machine learning applied in digital afterlife services will only become more apparent. Necessarily indulging the speculative, we ask: Whose agency will DAS prioritize? Will curated post-mortem personas be handled as sacred memorial artifacts or will the living be allowed to finetune, tweak and modify their behavior? How will they filter disclosures of information private to a specific user, or manage the presentation of differentiated affect or user-conditional personalities? Should post-mortem “memories” persist across users? If I tell my dead aunt I split ways with my lover, can she “remember” this and talk about it with my living sister? Can I add a dead friend bot to the group chat? What happens when DAS language models “hallucinate” or bullshit in potentially traumatizing ways? Might they confabulate false memories of intimate events that never occurred or contrive impossible plans with living users to “be together again”? In our increasingly terminally online lives, how will the lines between living and “ghost” users continue to blur? Will DAS change the way we treat the dying or infirmed? These in-media-res questions haunt discourse on DAS for good reason–because they are not yet well understood.
We remain unconvinced by the emotional promises of vendors such as HereAfter and Project December. DAS are inadequate archivists and reproducers of the improbable. As much as DAS services attempt to reproduce and predict the dead’s idiosyncrasies through generative text and multimodal repositories, their underlying technologies are inherently statistical in nature: they are pattern recognizers and generalization machines. What we miss is a person’s surprises, the incalculably small and statistically insignificant details that make up a life: a loved one’s evolving notes in the margins of a novel, the small effects of hunger and frustration and fatigue, the experience of watching them change their mind.
At present, DAS are distinctly in media res; their own speculative promise, however, is to one day become a given fixture of living and dying. But what happens to our grief compartments when DAS themselves are privy to decay? Amidst the urgent and ambient stakes of our polycrises, will DAS go wildly wrong or wildly right? Or, wielding stakes that feel impossibly urgent in a time of mass precarity, might they end up meaning very little at all? Clawing intentionally at these speculations, we insist that our vulnerability and fascination with DAS is crucial to consider in its own right, refiguring our own sense of mid-ness in an uncertain, grief-steeped, and always inadequate present.
Sarah Sgro is a Visiting Assistant Professor of Literature and Creative Writing at Hamilton College and holds a PhD in English Literature from SUNY, Buffalo. She studies digital waste and is writing a novel about digital mourning, Instagram reels, and conspirituality.
Mike Walton is an Emerging Technologies Fellow with xD and holds an MS in Computer Science from Georgia Tech. He studies Machine Learning and the social impacts of Artificial Intelligence.