On September 30, an Arizona appeals court threw out a prison sentence because of a video of a dead man.
Christopher Pelkey was killed in a road-rage shooting in Chandler, Arizona, in 2021. At the sentencing of the man convicted of killing him, Pelkey's family played an AI-generated video of Chris himself, made from an old recording and a photograph, delivering a message of forgiveness to his killer. His sister wrote the script. The judge said in open court that he "loved that AI" and called it "genuine". He then gave a sentence of ten and a half years, a year more than prosecutors had asked for.
The Arizona Court of Appeals left the manslaughter conviction in place and ordered a new sentencing hearing. Nobody suggested the family meant harm. The court's problem was simpler. The forgiveness in that video was not something Chris said. In the court's words, it came from "the imaginings of the victim's sister", delivered through his face and his voice, and a judge had treated it as his.
We have been thinking about that ruling all week, because it applies to companies like ours.
Keeping what was said, and writing what was not
These two things get blurred together, and they are opposites.
Preservation keeps what someone actually said. Their real voice telling the real story of how they met your mother. Their laugh in the middle of a sentence. Their advice, in their words, recorded while they were here to give it.
Resurrection produces what we wish they had said. It takes a voice and gives it new opinions, chosen by someone else, after the person can no longer agree or object. However good the intentions, the result is a performance. The Arizona court saw that clearly: the video was evidence of what his sister believed, not of what Chris thought.
Two more rulings in the same week
- Shanghai, September 29. An appeals court upheld damages for a voice actor whose voice turned up in an AI company's promotional audio. The company could not show where its training recordings came from, and the court put that burden on the company, not on her. Using someone's voice to train a system that produces a voice people recognize as theirs, without consent, violates their rights.
- Tokyo, September 30. Voice actor Kenjiro Tsuda had asked a court to make TikTok's operator remove at least 188 videos that imitated his voice with AI. The account had already been deleted, so his request was dismissed. But for the first time in Japan, a court said a voice can be protected the way a face is: it is "a symbol of a person's personality".
Three courts in two days, and one idea running through all of them: a voice belongs to the person it comes from.
Where that leaves Senarra
It would be easy to write "we would never" here. This is a post about honesty, so here is exactly what Senarra does.
The recording is the record. Most of what Senarra does is record people telling their own stories, on a phone call or in the app, and keep those recordings for the family.
Senarra can also answer in a preserved voice. On the higher memberships, a family can make a Voice Profile from recordings and ask questions of a loved one they have preserved. Those answers are put together by AI. They draw only on the memories the family recorded, and when there is nothing recorded to draw on, Senarra says so. That is a way of returning to what was kept. It is not the person speaking, and it is not a record of anything they said. It should never be shown to anyone as their words: not to a court, and not to a grandchild.
A voice needs consent. Before a Voice Profile is made, the family member making it has to confirm that the person has given their permission in writing. If the person has died, they have to state that they have the standing to make that choice. Either can be withdrawn at any time, and the voice stops straight away.
The Arizona ruling drew its line at treating an AI's words as a person's own. We think that line is right, and we hold ourselves to it. More in our ethics statement.
What this means for your family
The time to keep a voice is now, while the person is here to tell the story themselves and to decide what they want kept.
A recording made together is the one thing no software can produce afterwards. They knew the microphone was on. They chose what to share. Every word is theirs.
The hard part was never the technology. It is sitting down with someone you love, pressing record, and asking the questions you have been meaning to ask.
Call your parents. Press record. We will help you keep it, exactly as they said it.
Sources
- Arizona manslaughter sentencing vacated due to use of AI victim impact statement (FOX 10 Phoenix)
- Judge wrongly permitted AI-generated victim video, state appeals court says (ABA Journal)
- Tokyo court rules that a person's voice can be protected by publicity rights (Music Business Worldwide)
- Shanghai First Intermediate People's Court holds AI voice-cloning platform liable (National Law Review)