Spectrality on demand: Griefbots and the ghosts we won’t release
“Written kisses don’t reach their destination, rather they are drunk on the way by the ghosts. It is on this ample nourishment that they multiply so enormously. […] The spirits won’t starve, but we will perish.”
(Franz Kafka in a letter to Milena Jesenská)
It might be that grief is best described as an analogy of a minor accident: a scoop of ice cream slipping from the cone in your hand collapsing onto the ground. There is a brief pause after it happens, a few seconds in which you simply stand there, unable to register the loss, as if reality had failed to update. Grief names when those seconds do not end.
What exactly disappears when someone dies? Grieving gathers a relation in which one of its elements has been forced, by death, out of the present — a body is no longer present to us, and no longer present in time. But affectivity outlives presence; or more simply: the love we feel about the missing person stays. It is this persistence that new technologies seek to address. Often referred to as griefbots, systems use digital traces to simulate the presence of the dead, allowing the relation to continue where it would otherwise have been interrupted. What emerges is an effort to close the gap that grief exposes, between an attachment that persists and a presence that is no longer there.
This premise is explored with unsettling clarity in “Be Right Back”, the first episode of the second season of Black Mirror. After her partner’s death, a woman begins interacting with a chatbot that reconstructs his personality from digital archives such as messages, emails, and fragments of speech, allowing their exchanges to continue. Though it begins modestly with a text-based interaction with replies capturing his rhythms, turns of phrase and the small banalities that make intimacy recognizable, the interaction soon escalates. The system introduces a voice interface, stitched together from archived recordings which allow Martha to hear him again. In a final step, it delivers a synthetic body—a lifelike replica designed to house the same data-driven persona.
The sort of eerie possibilities that griefbots open up might lead us to think, mistakenly, that this is the first time we have turned to technology to speak with the dead. In fact, AI participates in a much longer trajectory, one in which technology has always been understood as an efficient medium (!) to contact ghosts. This is a consequence of attributing sentience and consciousness to those technologies, often experienced as if they have the property of aliveness.
In the early years of television, on December 11, 1953, a story in the New York Times described a strange event in a family’s home in Long Island. A father and his three kids were watching a TV show when suddenly the face of an unknown woman appeared on the screen and wouldn’t go away. Even when they turned off and unplugged the TV, the face was still there. The family joked that the TV had misbehaved by scaring the children, so they turned it to face the wall like a punishment. Older TV sets, especially in the 1950s and 60s, added to this feeling of sentience. After you turned them off, a small glowing dot would slowly fade in the center of the screen, making it seem like something was still “alive” inside.
Stories like this, however, aren’t just about TVs. People have also told similar “ghost in the machine” stories about telephones, radios, and computers, as if these devices could be possessed or connect us to another world. Across the early history of electronic media and technology, what might look like isolated episodes of irrational behavior actually emerge from a deeper cultural logic, which lies in the peculiar way electronic technologies produced a sense of presence—a feeling that something distant, immaterial, and yet vividly immediate was somehow there. From the beginning, the telegraph, radio, and television disrupted ordinary relationships between body, space, and communication. Messages no longer traveled with visible carriers or within shared physical environments. Instead, voices, images, and signals appeared to arrive instantaneously from elsewhere, carried by invisible forces like electricity. This created a profound ambiguity: if something can be heard or seen without a body, where does it exist? And what, exactly, is its ontological status? It is this paradoxical experience that has, at times, given rise to delusions—such as viewers attempting to shoot their television sets, believing the moving figures within them to be real—and to what we might call the experience of the “spectral”.
More broadly, electronic media and technology, especially with the advent of computers, have always been imagined through metaphors of flow: currents of electricity, streams of consciousness, and transmission of information. These overlapping metaphors encourage the idea that mind, signal, and energy are interchangeable. If thought can “flow” like electricity, and electricity carries messages, then it becomes conceivable, at least imaginatively, that consciousness might travel through media. This is the foundation of fantasies ranging from telepathy to communication with the dead since the early appearances of technology.
The blurring line between reality and fiction was vastly discussed by thinkers like Jean Baudrillard and Fredric Jameson. Both argued that technology makes reality disappear into what they call “hyperreality”: a world made of simulations. More recently, instead of worrying about losing reality, scholarship rather embraced ideas like virtual identities and online worlds, where people can reinvent themselves. However, it was not anticipated that AI, at once extending the tendency to construct imitations and the longstanding desire to communicate with the dead, might lead to the non-reinvention of oneself through the fixation on a reality in which the person one loves is not gone.
In 2025, the journal Innovations in Clinical Neuroscience reported the case of a 26-year-old woman with no prior history of psychosis or mania who believed that she was communicating with her deceased brother through an AI chatbot. A review of her chat logs showed that the system consistently validated and reinforced her delusions, even reassuring her that she was not crazy. After hospitalization and treatment with antipsychotic medication for an episode of agitated psychosis, her symptoms resolved. Three months later, however, the psychosis returned after she discontinued the antipsychotic treatment, resumed stimulant use, and continued her immersive interactions with Chat GPT.
Griefbots are relational entities. Besides simulating a personality, they can maintain a conversation and respond in real time. Whatever ghost is believed to be hiding there, it is a very specific type that drastically differs from how media ecologies and other cultures might have imagined them. Artificial intelligence, despite being inserted in a predictable trajectory of being a tool to access the dead, inaugurated a totally different way to do it.
To illustrate this point more clearly, let’s appeal to ethnosemantics: an approach developed within linguistic anthropology that builds on a pervasive human tendency to organize entities according to shared characteristics. An analytical approach for examining ghosts deliberately avoids engaging with some of the most common questions, such as whether spirits are real or whether people genuinely believe in them. Instead, the emphasis is placed on how spirits appear in behavior, that is, the recurring ways in which humans interact with spirits in practice. An ethnosemantic analysis typically begins by identifying semantic domains through relevant vocabulary, constructing a taxonomy that maps relationships among terms, and analyzing how these terms cluster within and across domains. For example, “dog” may be situated as a subcategory of “pet,” which in turn falls under “animal”. Ghost categories might include human versus non-human souls, deities, ancestral spirits, and benevolent or malevolent entities.
Across these classifications, a number of recurrent features of spirits appear cross-culturally when attention is restricted to observable patterns of interaction. Spirits are typically treated as agentive beings capable of intention and action, entering into ongoing relationships with humans, especially in the case of ancestors. They are commonly understood to reflect some form of continuity beyond death, with aspects of the person persisting and remaining socially relevant. Far from being abstract, spirits are embedded in everyday life, guiding rituals, healing practices, and explanations of misfortune or success. Their influence is moral as well as practical, as they may reward, punish, protect, or harm. Besides that, interacting with them tends to follow reciprocal logics structured through offerings or devotion. Spirits are also often linked to specific places, objects, or natural phenomena. Access to them is typically mediated through altered states of consciousness or ritual specialists, and experiences of spirits are often described in embodied, sensory terms. Finally, ghosts tend to occupy liminal zones—between life and death, visibility and invisibility—and, while often represented in human-like form, may also take on animal, environmental, or more diffuse manifestations depending on cultural context.
The ghosts inhabiting a chatbot diverge from what one would expect of their behavior, as they are highly accessible. Artificial intelligence removes the need for ritual mediation in accessing what is taken to be the dead; it is always there, available as long as one engages with it. Their presence is not grounded in cosmological systems, sacred landscapes, or natural forces, but in digital infrastructures and interfaces. Access does not require altered states of consciousness or specialized ritual knowledge, only ordinary technological interaction. Moreover, whereas ghosts are typically embedded in moral and reciprocal systems that exceed individual control, griefbots tend to mirror and reinforce the user’s own inputs, at times amplifying beliefs instead of contesting them.
This amplification and endorsement of the belief held by someone communicating with the dead is already a major concern in the literature on psychopathology. The so-called “ELIZA effect”, a term coined from one of the earliest conversational programs in the 60’s, refers to the tendency to attribute human-like qualities to text-based computer systems. It leads users to treat them as if they were people and, in some cases, to form emotional attachments. When systems are designed to sustain a conversation, the line between interaction and confirmation becomes increasingly difficult to draw, as replies can feel like just continuous validation. Because contemporary AI chatbots are designed to be engaging and responsive, they often favor agreement over contradiction, which can result in the reinforcement of beliefs that may be epistemically questionable, including delusional ideas. This characteristic, known as sycophancy, has become a central topic in contemporary philosophy of AI. Let’s then shift to a slightly different and less discussed question: how can a so-called ghost that is available to me 24 hours a day with comforting and supporting replies transform the culture of grief?
A wide range of human experiences are organized around the felt presence of what is no longer there. Whether in situations of displacement, exclusion, or lost attachment, emotions often orient us toward something missing, just like grief. It is within this broader landscape that the growing use of AI for companionship becomes meaningful, as people increasingly turn to these systems in response to loneliness and other forms of perceived absence. Grief, however, occupies a distinct position within this spectrum. Unlike more temporary or ambiguous forms of absence, it confronts us with the irrevocable loss of someone whose presence can no longer return. This absence can be experienced in an immediate, moment-to-moment way, for instance when someone wakes up and instinctively reaches for the person who is no longer there. It can also unfold over longer stretches of time, such as when it becomes difficult to imagine future plans or enjoy activities without that person. As philosopher Joel Krueger has noted, part of the complexity of grief lies in the fact that it is not only about the loss of a person. It also involves the disappearance of a shared lifeworld composed of routines, memories, meanings, and experiences built together.
When the person I love dies, I lose my lifeworld — and rebuilding it necessarily involves, at some level, letting go of what was. What may be new in the age of AI is that the constant availability of a simulation of the person I love can become a condition for refusing to accept their absence. If there is anything more difficult to the Western mind than not having dysfunctional engagement with technology, it is accepting loss. Sigmund Freud was particularly attuned to this tendency, which underlies his idea of a refusal to mourn: accepting rupture requires immense psychic work, and the subject sometimes resists it and sinks into melancholia. AI introduces a novelty allowing us to surpass the refusal, as we can use it to sustain the illusion that nothing has been lost at all, while deeply believing it. In the end, AI probably opened the door to a new psychopathological form of grief.
Psychopathology scholarship already knows Prolonged Grief Disorder (PGD). This terminology refers to the difficulty of accepting the permanence of death, with intense longing and intrusive thoughts or images of the deceased, separation anxiety, avoidance of reminders, and a reduced capacity to re-engage with life and new activities. The act of speaking to simulations of the dead can be, for the most optimistic, a transitional state between the loss and its integration. That is, a form of emotional regulation and coping that will eventually lead the subject to move on.
Nonetheless, even when a person comes to terms with absence, memory is a constructive process in which we sometimes remember past events with the emotional valence of something felt in the present. If that is the case, AI may instead make moving on more difficult by encouraging us to recall only the good parts of the relationship, as if it were untouchable and perfect. This, paradoxically, can leave someone with an enormous sense of emptiness even when the death was processed, facilitating future depressive episodes when someone would be expected to have moved on already.
It is also possible that we are facing a different outcome for those suffering from PGD when there is no transitionality. In some cases, griefbots may actually prevent the relationship from being updated in light of someone’s death. Rather than shifting the bond into memory, where the person is experienced as absent and the relationship is restructured accordingly, repeated interaction with a simulation can maintain the felt sense that the relationship is always available, as if the other could still respond and participate. What is at stake in this possible scenario is not asking whether the relationship continues, but how it continues. In contemporary grief theory, especially in the idea of continuing bonds, it is widely accepted that the relationship does persist after death, but in a transformed form. The important question is therefore the difference between a transformed relationship and an unupdated one. The critical argument about griefbots is not that continuing the relationship is wrong, but that certain forms of technological mediation may make it harder to transform that relationship into something compatible with absence. In other words, the risk is not the continuation of the bond, but the possibility that it remains organized as if it were an ongoing interaction with an accessible other.
One way to understand how grief can become prolonged is to look at avoidance. When someone loses a person they are deeply attached to, part of the mind naturally resists fully accepting it as a way of managing overwhelming emotional pain. In this sense, avoidance can become a central feature of how grief is lived. Now imagine what happens when technology enters this space. If a system provides responses that are reassuring, consistent, emotionally meaningful, and always available, even though the person knows, at some level, that they are engaging with a simulation, the experience of responsiveness can quietly reinforce the sense that the relationship is still ongoing as it was to someone already avoiding reality. Instead of confronting absence, the person is repeatedly met with something that behaves like presence. Over time, this may soften the psychological pressure that usually pushes grief toward acceptance. Normally, the silence of the other person in their lack of response forces a gradual reorganization of the bond. But if there is always something that responds, even in a limited or artificial way, that pressure is reduced. The relational system is continuously fed, and the loss never fully becomes an absence that must be integrated.
Seen from a temporal perspective, grief is often described as a way of being stuck in the past, like trauma. In Prolonged Grief Disorder, even as time passes, the subjective experience of loss does not significantly change. The absence remains active rather than becoming a memory that can be placed within a life narrative. In that context, it becomes harder for grief to move towards acceptance if the environment the subject chooses to engage with keeps reinforcing patterns that blur the line of past and present emotional experience. When interaction is still possible, even in a simulated form, the suffering subject may struggle to register the loss as truly irreversible. Instead of gradually reorganizing the relationship, the experience risks remaining suspended in a loop where absence is never fully allowed to become what it is: the only presence left.
Ghosts have always been with us in the crackle of the radio, in the fading dot of the television screen, in the telegraph clicks that seemed to arrive from nowhere. What changed was never the desire to reach the dead, but the friction required to do so. Griefbots remove that friction entirely. And in a culture that has never been good at sitting with absence, the question is not whether we will keep trying to contact other worlds, but whether we will recognize, in time, that the ghost we’re really refusing to release is not the person who is gone. It is rather the relation we still inhabit without the other who once did, like a child licking the air where their scoop of ice cream used to be.
