Spirits in the Algo

YYour friend has been dead for two years when Instagram decides you might enjoy seeing her. Maybe an old photograph appears because somebody commented on it, her account turns up in a suggestion, or a video she once liked wanders back into circulation. There is no reason to think the machinery underneath this is doing anything unusual. The system has noticed a relationship between pieces of data and done what it was built to do with relationships between pieces of data, except one of the people involved is no longer alive.

Death has always left objects behind. Photographs, letters, clothes, recipes, answering-machine recordings kept long past the answering machine. What is relatively new is having those objects participate in systems designed to decide when you should see them. A photograph in a drawer waits until somebody opens the drawer; a photograph inside a platform can be sorted, ranked, resurfaced and placed in front of you while you are doing something else entirely.

A 2026 study of digitally mediated grief interviewed 24 bereaved people about the ways social media and other digital traces shaped their continuing relationships with people who had died. Participants described preserved conversations, images, voice notes and profiles as places where the relationship remained accessible, but the researchers also identified algorithmic return: platforms did not merely store traces of the deceased but could surface them again, repeatedly inserting those traces into the present.

Suggested for you

There is a meaningful difference between choosing to remember someone and being reminded. You can open the old messages because you miss her, search for the photograph from that weekend or play the voicemail you've kept through four phones and three increasingly aggressive storage warnings. The encounter may still hurt, but you initiated it, which means you had at least some idea what room you were walking into.

Algorithmic memory removes that small preparation because the systems deciding what to surface are not reconstructing the emotional significance of what they contain. They are sorting enormous amounts of information according to signals such as prior interaction, similarity, relevance and engagement. From that perspective, a dead person can remain an unusually strong data point. You exchanged hundreds of messages, appeared in photographs together, visited the same places, knew the same people and liked many of the same things. The relationship may have ended in the most absolute way available to human beings while remaining impressively intact as a pattern.

That creates a mismatch between social reality and computational reality. In one, the person is permanently absent. In the other, years of interaction may continue to identify her as unusually relevant to you. Death changes almost everything about the human relationship while leaving many of the signals by which a platform understood that relationship exactly where they were.

People you may know

Grief already has terrible timing without technological assistance. You can be fine in the grocery store until you see the cereal he bought, hear four seconds of a song in a dentist's office or notice somebody walking ahead of you in the same coat. Human memory has always been capable of turning harmless objects into sudden encounters with people who are no longer there.

Digital traces change the number and behavior of those objects. A sweater usually stays in the closet. A photograph remains in the box. The birthday card does not periodically leave the drawer and position itself between an advertisement for running shoes and somebody's engagement announcement. Digital objects can be copied, reordered, resurfaced and attached to contexts that have nothing to do with the circumstances in which you originally knew them.

Researchers studying digital mourning increasingly describe platforms not simply as archives but as environments that shape how relationships with the dead continue. Preserved messages and profiles allow people to return deliberately, while algorithmic systems introduce another form of encounter in which the timing is partly determined elsewhere. The photograph itself may be identical in both situations, but finding it because you went looking and finding it because a system placed it there are different experiences of the same memory.

Still active

There is another complication in leaving a digital life behind: much of the data itself has no reason to behave differently after death. The message thread still accepts text. Old comments remain written in the present tense. Photographs continue to be tagged with a name. Depending on the platform and what happens to the account afterward, the traces of a person can retain much of the structure they had when ordinary participation was still possible.

We usually understand death as creating a decisive distinction between someone who exists and someone who existed, but digital systems were built around different distinctions. A photograph does not become a different kind of file because one of the people in it died. A history of messages does not cease to record a strong relationship. Unless information about the death enters the system and changes how those signals are handled, the behavioral residue can remain computationally useful in many of the same ways it was before.

This produces something that is neither ordinary preservation nor anything resembling actual presence. Your friend is not alive inside the phone, but her traces have not necessarily become static either. They can still move through systems, enter a feed you opened for an unrelated reason and appear beside material created that morning. The dead have always remained part of the lives of people who loved them; some of the objects through which they remain are now capable of circulating without the living deliberately retrieving them.

The wrong day

Calling this simply cruel would misunderstand the problem because the return is not always unwanted. Bereaved people in the 2026 study described digital spaces as supporting an ongoing sense of connection with the deceased, including through preserved interactions and profiles. A photograph that feels unbearable at one moment may be precious at another, and the same person who closes an app after an unexpected reminder may later search deliberately for exactly the material she did not want to see that morning.

That instability makes grief an unusually difficult personalization problem. You may want every photograph and then none of them. You may spend an evening reading old messages and resent seeing a memory the next morning. Looking at photographs of someone who died may mean you want to see more, that today is their birthday, that you couldn't stop looking at the first one or simply that you are having the worst night you've had in months. From the perspective of a system learning from behavior, very different emotional states can produce remarkably similar evidence.

Recommendation systems ordinarily become useful by treating past attention as information about future relevance. Grief interferes with the assumption underneath that process. The system may correctly infer that this person remains important to you while having almost no useful way to determine what importance should mean today.

No séance required

Much of the newer conversation about technology and death focuses on more dramatic possibilities: griefbots, voice clones, generated avatars and services built specifically to simulate interactions with people who have died. Those technologies raise obvious psychological and ethical questions because they can generate new material from a dead person's digital traces rather than simply preserve what the person actually left behind.

The quieter change happened earlier. Ordinary platforms accumulated enough of our relationships that death no longer guaranteed digital stillness. People leave profiles connected to profiles, photographs connected to dates, messages connected to names and years of behavior connected to other behavior. None of this needs to simulate a person convincingly to alter the experience of remembering them. It only needs to remain legible to systems that continuously decide which pieces of the past belong in the present.

That leaves grief in a peculiar relationship with technologies built to predict relevance. The person may be absent in every ordinary sense while remaining one of the strongest associations in the record of your life. The system can be completely correct that she matters and completely unequipped to know what should follow from that fact.

For a long time after someone dies, you may not know either.

Sources

Malik, S. (2026). "Digitally Mediated Continuing Bonds: Social Media, Grief, and Relational Continuity after Death." OMEGA---Journal of Death and Dying. https://doi.org/10.1177/00302228261479913

Thommandru, A., & Mone, V. (2026). "Datafication of mourning: Emotional labour, memory politics, and ethical innovation in digital grief technologies." Journal of Responsible Technology, 25, 100159. https://doi.org/10.1016/j.jrt.2026.100159

Kraaijeveld, S. R., & Schuurbiers, D. (2026). "Never say goodbye: assumptions of 'normal' grief in framings and portrayals of AI grief technologies." Journal of Responsible Technology, 25, 100149. https://doi.org/10.1016/j.jrt.2026.100149

Chen, X., Sun, X., & Yip, A. (2026). "Digital grief technology to support bereavement: A systematic review of potential benefits and risks." Computers in Human Behavior Reports, 22, 101148. https://doi.org/10.1016/j.chbr.2026.101148