Discussion

AI recreations of the dead raise a question grief cannot answer for us

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#13580

AI can now give a deceased person a face and voice, but the harder questions are who consents, who controls the simulation and who gets to switch it off. In an Irish Examiner opinion piece, Colin Sheridan examines the real services behind that uneasy idea, including the difference between a scripted memorial video and a chatbot that can keep replying.

Watch Desk analysis

What happened

Sheridan describes a South Korean man commissioning a scripted video of his grandfather to comfort his father. The apology in the video was invented, he writes, but the family’s reaction was real. The format has a clear ending. A conversational simulation is a different proposition: it can answer again, and a company can change, charge for or close the service.

The piece says South Korean company Vaice offers short, scripted films. Its chief executive told the Associated Press that a basic three-to-five-minute video cost 600,000 won, or about €407, and that the company served around 300 customers a month. Sheridan also recounts a chatbot simulation that ran on credits, with its memory wiped when the credits ran out. That example came from a 2021 San Francisco Chronicle report.

Why it matters

A generated voice can put words in someone’s mouth that they never said. Sheridan asks what happens when a family disagrees about a simulation, or when a grieving person wants to keep talking after another relative wants it switched off. Those questions concern both the deceased person’s wishes and the living people whose messages or recordings may be used to build the imitation.

The article notes that Ireland’s Data Protection Commission says the GDPR does not apply to the personal data of someone who has died, while records may still contain information about living people. It also points to EU AI Act disclosure requirements for covered deepfakes and interactions with AI. A disclosure can identify a simulation; it does not settle who had the right to make it or what the person would have wanted.

Our read

The useful distinction here is not simply “AI memorials: good or bad”. A family watching a finite tribute is not the same as a subscription chatbot offering another conversation at any hour. Sheridan’s account makes a strong case for treating consent, control and a dignified way to end the service as core design questions, not small print after the emotional sell.

What to watch

  • Whether services seek consent from the person represented, where possible, and from the people who will interact with the simulation.
  • Whether users can preserve genuine recordings while deleting generated material and chat histories.
  • Who can end a simulation when family members disagree, and what happens to its data afterwards.

Discussion spark: Should families be able to keep a deceased relative’s AI simulation running if other relatives object, or should the person represented have the only meaningful say?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.