Mind uploading is a technical project. AI resurrection is a cultural one. The two are frequently conflated in public discourse, but they represent fundamentally different undertakings — and in 2026, only one of them is actually happening.

A peer-reviewed paper published in the OMEGA — Journal of Death and Dying in February 2026 by Sherman Xie provides the most comprehensive sociological analysis to date of AI resurrection as a social practice. The paper documents the shift from AI resurrection as an experimental technology explored by early adopters to a service used by grieving families globally, offered by commercial companies, and discussed without stigma in popular media. Xie’s analysis examines what people want when they choose AI resurrection, why adoption is accelerating, and what the gap between this practice and genuine mind uploading reveals about the human relationship with death and identity.

What AI Resurrection Is, and Is Not

AI resurrection in its current form works by training a language model — typically a fine-tuned version of a large foundation model — on a corpus of data from the deceased: text messages, emails, social media posts, voice recordings, video content, and any other data that captures how the person communicated. The resulting system can respond in a style that mimics the deceased’s communication patterns, uses similar vocabulary, makes references consistent with the person’s documented interests, and in some systems can produce synthetic voice output trained on existing recordings.

What this produces is a behavioral simulacrum. It reproduces the surface patterns of how a person communicated, not the underlying cognitive processes that generated those patterns. The system has no access to memories, reasoning processes, or beliefs that were not encoded in the training data. It does not update from conversation to conversation — it does not learn that you argued last Tuesday or that circumstances have changed since the last recorded data was collected. It cannot experience anything, form new preferences, or be hurt by neglect.

The distinction from whole brain emulation is absolute. WBE aims to capture and run the actual computational process that constitutes a person’s mind. AI resurrection builds a statistical model of how a person expressed themselves in recorded communication. The first is a technical program of dubious feasibility at its outer limits; the second is a deployed commercial service available today.

Xie’s paper does not argue that AI resurrection is whole brain emulation. It argues that understanding why people adopt AI resurrection is essential for understanding what they actually want from digital immortality — and that this understanding has been largely absent from the technical debate.

Drivers of Adoption

The paper identifies four primary drivers of the shift from fringe to mainstream.

Technological accessibility. The cost of training a custom AI model on personal data has fallen by orders of magnitude since 2020. What required specialized AI research infrastructure five years ago now runs on consumer cloud services for a monthly subscription fee. Companies including HereAfter AI, StoryFile, and several Chinese-market services have built consumer-facing products that require no technical expertise to use.

Social media as memorial archive. The average person who died in 2025 after age 40 left behind years of recorded digital communication: social media posts, group chat histories, video calls, voice messages. This data is, in many cases, sufficient to train a recognizable behavioral model. The deceased person created the training data unknowingly and without consent, a point Xie treats as ethically significant.

Grief and social need. Xie documents extensive interview data showing that bereaved individuals who use AI resurrection services are primarily seeking continued social connection, not technical immortality. They want to ask questions they forgot to ask while the person was alive. They want to hear a familiar voice. They want something to talk to in the first days after a loss when the absence is most acute. These are grief processing functions, not consciousness continuation functions.

Commercial normalization. The appearance of AI resurrection in mainstream media — television dramas, news features, celebrity discussions — has reduced the social stigma that would have attached to the practice five years earlier. Xie documents the shift from media coverage framing AI resurrection as disturbing or uncanny to coverage framing it as a personal choice and a new form of memorial.

The Consent Problem

Xie’s paper gives sustained attention to the consent issue. In nearly all current AI resurrection cases, the deceased person did not explicitly consent to being modeled after death. Their data was collected for other purposes — social connection while alive — and repurposed for resurrection modeling by survivors.

This creates a category of posthumous representation that existing legal frameworks do not address adequately. The deceased cannot object. Their digital estate — if one is legally recognized — may or may not cover behavioral modeling. The survivors who commission the resurrection may believe they are honoring the deceased; the deceased, if asked, might have found the practice disturbing.

Xie argues that the consent problem is more serious for AI resurrection than it initially appears because the systems produce novel outputs, not just archives. An AI resurrection system will generate responses that the deceased person never actually said. In long conversations, it will make claims, express preferences, and tell stories that have no basis in the training data but are generated probabilistically. The bereaved person is interacting with a model, not with a record. The risk of misattribution — believing that a generated response reflects the deceased person’s actual view — is significant.

This issue does not arise in the same form for digital memorial technologies that are explicitly archival rather than generative. An interactive archive that surfaces actual recorded messages is different from a system that generates new ones.

What People Want from Digital Immortality

The most analytically important section of Xie’s paper is the examination of user motivations through interview data. When Xie asks bereaved users of AI resurrection services what they hoped to get from the technology, their answers cluster around several themes that are distinct from the technical goals of mind uploading research.

Continued relationship, not continued person. Most users describe the AI resurrection as a relationship object rather than a person. They are aware it is not the deceased. They use it because the interaction provides something — familiar language patterns, a sense of presence — that pure memorialization does not. This is closer to how people relate to photographs or journals of the deceased than to how they would relate to a resurrected person.

Processing, not preservation. Many users describe the service as something they used in the acute grief period and then stopped. The function was not to maintain an ongoing relationship with a resurrected person indefinitely, but to have access to something during a difficult transition. The services are used most intensively in the weeks and months immediately after bereavement.

Answering specific questions. A significant proportion of users describe specific practical or emotional questions they wanted to pose to the deceased: where a document was filed, whether a specific regret was shared, what the deceased would have thought of a family decision. The AI resurrection was expected to address these questions, which it often could not do reliably. User dissatisfaction tracked closely with cases where the system hallucinated responses inconsistent with what the deceased would actually have said.

The overall picture is of a technology being used for emotional support functions that are qualitatively different from consciousness continuation. Users are not primarily motivated by a belief that they are interacting with the deceased person’s ongoing consciousness. They are using the system as an extended memorial with interactive properties.

The Gap This Reveals

Xie’s analysis reveals a gap between what AI resurrection users want and what the mind uploading research community is working toward. The bereaved users in the paper are not demanding consciousness continuation. They are demanding recognizable presence, accessible communication style, and emotional availability.

These are achievable with current technology. Consciousness continuation is not. The fact that AI resurrection is succeeding commercially — meeting genuine human needs — does not mean that the philosophical and technical questions around genuine mind uploading are resolved. It means that a different, simpler function has found a market.

This has an important implication for how the ethics of digital identity should be framed. The risks that require governance now are those of behavioral simulation: consent, misrepresentation, emotional dependency on systems that hallucinate responses, commercial exploitation of grief. The risks of actual consciousness transfer — questions about rights of uploaded minds, identity continuity across substrates, the copy problem — remain speculative.

The cryonics and brain preservation community draws a sharp distinction between preservation as a path to genuine future revival and digital memorialization as a comfort for survivors. Xie’s paper suggests that the populations interested in these two things overlap less than the research community sometimes assumes. The people who commission AI resurrections are largely not the people who sign up for cryopreservation. They have different theories of what death means and different intuitions about what continuation would require.

Cultural Normalization and Its Consequences

Xie concludes by noting a consequence of AI resurrection’s normalization that may affect the long-term trajectory of the field. As interactive AI models of the deceased become a standard part of bereavement culture, public understanding of the distinction between behavioral simulation and genuine mind uploading may erode. If the dominant cultural reference for “digital afterlife” is the AI resurrection chatbot, the conceptual space for understanding what genuine whole brain emulation would require becomes harder to maintain.

The Chinese youth digital immortality phenomenon documented in earlier research showed that younger cohorts often frame digital afterlife primarily in terms of AI persona preservation rather than biological consciousness continuation. Xie’s paper suggests this framing is becoming global, not merely a regional cultural pattern.

For the research community, this argues for clearer public communication about the distinction between behavioral simulation and cognitive emulation. Not because AI resurrection is wrong or harmful — Xie’s paper suggests it serves real human needs — but because conflating it with consciousness uploading obscures what the technical challenges actually are.

Official Sources

• Xie, S. (February 2026) — Why AI Resurrection Is Becoming Increasingly Popular. OMEGA — Journal of Death and Dying. DOI: 10.1177/00302228261427860. https://journals.sagepub.com/doi/10.1177/00302228261427860
• Kasket, E. (2019) — All the Ghosts in the Machine: Illusions of Immortality in the Digital Age. Robinson. ISBN: 9781472138361
• Savin-Baden, M. & Burden, D. (2019) — Digital immortality and virtual humans. Postdigital Science and Education, 1(1):87–103. DOI: 10.1007/s42438-018-0007-6
• Mori, M. (1970) — The uncanny valley. Energy, 7(4):33–35. (Classic reference on human responses to simulated presence.)
• van den Hoven, M. et al. (2023) — The Ethics of Digital Afterlife Technologies. Ethics and Information Technology, 25(4):50. DOI: 10.1007/s10676-023-09726-0