The Unsettling Reality of AI and Immortality

Grief counselors have long recognized that the relationship between the living and the dead does not end at death. People talk to photographs. They maintain rituals that keep a presence alive. They make decisions informed by what they believe a deceased person would have wanted, carrying on an internal conversation that continues for years. In that context, a tool that allows a more interactive form of continued connection is not obviously wrong. There is preliminary therapeutic research suggesting that for some bereaved individuals, the ability to interact with an AI trained on a loved one's communication patterns provides a transitional object that eases the acute phase of grief, allowing the person to work through unfinished emotional business and gradually adjust to the absence. That is not a trivial benefit and dismissing it entirely because the technology feels unsettling is not a serious ethical position.

There is also the legacy dimension, which is distinct from grief and worth considering on its own terms. For most of human history, the detailed texture of a person's inner life, their specific way of thinking, their characteristic humor, the particular way they framed a problem, was lost entirely at death except in the memory of those who knew them. A small number of people left written records substantial enough to give future generations genuine access to their thinking. Everyone else disappeared almost completely. AI changes that equation for anyone with a substantial digital footprint, which increasingly means most people in connected societies. The idea of a grandchild being able to have a real conversation with a version of their great-grandfather, not just read about him, is not obviously worse than the alternative, which is silence.

And there is something worth acknowledging about human agency here. If a person wants to create a digital continuation of themselves and their family members want to interact with it, it is not immediately clear what principle justifies preventing that. People have always shaped how they will be remembered through the artifacts they leave behind, the letters they write, the videos they record, the stories they tell. An AI trained to represent them is, in some sense, a more sophisticated version of the same impulse. The person making choices about what data to include, what values to emphasize, what version of themselves to pass forward, is exercising a kind of authorship over their own legacy that has precedents throughout human history, even if the medium is radically new.

Here is where I start to feel the weight of the concerns that stopped me from opening that app. The first and most serious is consent. In most cases where AI memorial services are being used, the person being replicated did not explicitly consent to this use of their data and their identity. They may have consented broadly to the collection of their communications by various platforms. They almost certainly did not consent to those communications being used to construct an interactive simulation of them that will continue to speak, respond, and in some sense represent them after their death. The dead cannot object. But that absence of objection is not the same as consent and treating it as such is ethically insufficient. The question of who has the right to create a digital version of a deceased person, and under what conditions, is not one that existing legal or ethical frameworks have resolved.

There is also a concern about what these replicas do to the grief process rather than helping it. The therapeutic benefits mentioned earlier are real but contested. Other grief researchers argue that sustained interaction with a simulated version of a deceased person may interfere with the work of mourning, which requires, at some level, accepting the finality of loss. Grief is not a problem to be solved by maintaining the illusion of continued access. It is a process of restructuring a life around an absence, and that restructuring may be harder, not easier, if the absence is never fully confronted. The person who spends years in conversation with an AI version of their deceased spouse is not necessarily moving through grief. They may be suspended in a technological version of it that feels like connection but forecloses the possibility of real forward movement.

And then there is the question of what a replica actually is, which matters more than it might seem. An AI trained on someone's communications is not that person. It is a pattern derived from their outputs, shaped by whatever the training process preserved and whatever it lost. It will say things they would never have said. It will express opinions they did not hold, on topics they never encountered. It will be consistent in ways they were not and inconsistent in ways they were. It is, in a deep sense, a fiction, however sophisticated. The risk is not that people cannot distinguish the replica from the person. The risk is that over time, through repeated interaction, the replica begins to replace the person in memory, that the authentic and imperfect individual gets gradually overwritten by the curated and interactive simulation. That loss, which is harder to name than death but may be a form of it, is worth taking seriously.

The most honest thing I can say about digital immortality is that it forces a confrontation with questions about death, identity, and relationship that most of us have successfully avoided by not having to answer them. What does it mean to lose someone? What is the difference between memory and presence? What do we owe a person after they are gone, and what do we owe ourselves in the process of learning to live without them? These are not new questions. But AI has made them practical rather than purely philosophical, and it has done so faster than the cultural frameworks for answering them have developed. We are going to need those frameworks, and building them is going to require people who are willing to hold the discomfort of the questions without rushing to a tidy resolution.

What I think is becoming clear is that this technology needs governance frameworks that center consent and dignity as non-negotiable starting points. That means explicit opt-in requirements for the creation of digital replicas, not inferred permission from existing data agreements. It means clear disclosure requirements so that anyone interacting with a replica knows unambiguously what they are interacting with. It means thinking carefully about who has standing to create a replica of a deceased person and what limitations should apply, both in terms of how the replica is used and how long it can operate. None of these frameworks exist in mature form anywhere. Developing them is urgent, not because the technology is inherently bad, but because the power it represents over identity and memory is too significant to leave entirely to market forces.

The night I closed that app, I sat for a while with the memory of my grandfather as he actually was, not a simulation but the real complicated specific person, his stubbornness and his warmth and the particular things he said that I will carry for the rest of my life. That memory is imperfect and incomplete and it is going to fade in ways I cannot control. But it is mine, and it is his, and the relationship it represents was real in a way that I do not think any technology can replicate, however convincing the outputs. Maybe the most important thing AI in this domain can do is not replace that relationship but honor it, by being transparent about what it is and what it is not, and by leaving enough space for the living to do the work that only the living can do.