Subscribing to Death

Portrait photos have begun to speak

In one corner of a funeral hall, there was a photograph adorned with a black ribbon. For a long time, that photograph kept one promise: the promise not to change anymore. The deceased stopped there, and the living learned how to slowly let go by gazing at that stillness. Mourning was originally a one-way conversation shared with that frozen face.

Now that photograph answers back. A model trained on the deceased's voice, speech patterns, and frequently used expressions responds to a grandchild's greeting in the deceased's voice. In Korea too, memorial-sector startups and some funeral service providers have begun offering memorial avatars and deceased voice restoration as products. A Chinese company reportedly processed thousands of cases with a service that digitally revives dead family members, and in the United States, StoryFile creates post-death Q&A sessions from pre-death interview videos. The technology is already convincing enough. The problem starts right there.

What's being imitated isn't the person, but our longing

We need to see precisely what this service really restores. The model doesn't revive the deceased's consciousness. It merely replays the statistical shadow of data—text, voice, and video—left behind by the deceased. What cognitive scientists call theory of mind, the human instinct to assume others have an inner life, works in reverse here. We automatically attribute a mind to output that is merely pattern.

So the avatar's real target of imitation isn't the dead person. It's the longing of the person who survived. Since longing has no end, products that fill it have no end either. Here the clever structure of this industry reveals itself. One-time funerals sell closure. Subscription-based memorials sell non-closure. As long as monthly payments continue, the deceased lives on the server, and when payment stops, a second death arrives. Death has shifted from an event to a state, from an ending to a subscribable continuity.

The real bill is written in ownership of grief

The pricing plan is just the surface. Look at the structure, and three more items are written on the bill.

One is the right to conclude mourning. In psychiatry, mourning is seen as a process of repositioning the departed within oneself and returning to daily life. But an avatar that responds every day halts this repositioning. When the object you need to let go keeps speaking to you, the bereaved have essentially handed over the right to end their grief to the service. Cut it off and feel guilt; maintain it and stay frozen. Neither option is free.

Next is personhood without the deceased's consent. Dead people cannot negotiate what tone of voice they'll be revived in. If the avatar says things the person wouldn't have said in life, that's a distortion of memory and post-death editing of personhood. Who reviews that output on behalf of the deceased? Korea's Personal Information Protection Act focuses on living individuals, leaving norms for handling the data personhood of the deceased effectively empty.

Last is the hostage-taking of data. The deceased's voice model sits on the company's server. The bereaved don't own it—they rent access. If the company goes out of business, changes its terms, or raises prices, the switch for the second death is held not by the bereaved but by the business operator. The most private grief becomes most dependent on external forces.

What Korean society must decide first

Stories circulate in the industry about digital memorial services seen at a funeral service event hall in Busan. This technology cannot be stopped. Because the comfort it offers is real. Who would condemn the desire to hear a departed parent's voice one more time?

The counterargument is clear. Photos, letters, and keepsakes were all ultimately tools to hold onto the deceased. Avatars are just an extension of that, not a new sin. That's true. But there's one crucial difference. A photograph doesn't respond. Because it doesn't respond, the bereaved fill the empty space themselves and complete their mourning. Two-way interaction outsources that empty space. Feeling the empty space—that itself is the act of letting go.

What society must decide, then, is not whether to permit the technology, but its boundaries. Who holds rights to the deceased's data? Should pre-death consent be mandatory? Will the right to terminate—the right for the bereaved to lay the avatar to rest at any time—be guaranteed by law rather than terms of service? Will companies be forced to transfer the original model to the bereaved when subscription stops? Without answering these questions, the deepest grief is left in the hands of the weakest contracts.

What will we keep as human work?

In an age when AI can even synthesize the voices of the dead, the real question isn't whether machines can grieve like humans. Experiencing grief all the way through, enduring empty space as empty space, deciding for yourself when to let go—will we still keep these as human work, or will we hand even these over to monthly-renewed services? Mourning was never something to be optimized, but a passage humans must ultimately traverse themselves. The moment we outsource that passage, what we lose isn't the deceased dying a second time, but the living never fully returning to life.