How to Die When the Machine Wears Your Face
In May 2025, road rage murder victim Chris Pelkey “appeared” in court to offer an impact statement on his own death and extend forgiveness to the defendant, Gabriel Paul Horcasitas, accused of killing him.
This, at least, was the enticing hook with which numerous news outlets presented the case. In reality, the victim’s sister, Stacey Wales, presented to the court a simulacrum of her brother that “spoke” to the court via an AI-generated image and voice. The avatar discloses at the top of its statement that it is
a version of Chris Pelkey re-created through AI . . . I was able to be digitally regenerated to share with you today. [pause, breath sounds] But here is insight into who I actually was in real life.
The script, which was written by Pelkey’s sister, simultaneously establishes and effaces its own representational authority. Is it a “version” of Pelkey, or is it Pelkey himself “regenerated”? The phrase “who I actually was in real life” further seems to signal Wales’ own ambiguity about the representational value of the simulacrum, but nonetheless, the script she pens is written in the first person, suggesting that she ultimately trusts the simulacrum (and her own speculations about her deceased brother’s thoughts and feelings about his death) to “give voice” to Pelkey at the trial.
In a quote given to local news outlet AZ Family, Wales recounts:
I said, ‘I have to let him speak,’ and I wrote what he would have said, and I said, ‘That’s pretty good, I’d like to hear that if I was the judge.’
The second half of this quote adds an important yet implicit addendum to Wales’ felt mandate to “let him speak”: she also needed him to speak well. What is valued in her statement is not just the posthumous restoration of Pelkey’s voice in the courtroom, but also the perceived rhetorical efficacy of her script as delivered by the AI simulacrum.
Wales’ tactic was successful, in any case. Judge Todd Lang was moved by the AI’s performance, and proceeded to deliver a harsher sentence for the defendant than the prosecution had initially asked for. Indeed, in court he told Wales that he
loved that AI. . . I heard the forgiveness. I feel that that was genuine.
Genuine to what, we must ask? To whom?
Despite Wales’ and Judge Lang’s espoused faith in the authenticity of the AI’s representation of the victim, defending attorney Jason Lamm argued that the AI represented a “kinder, more gentle” Pelkey than the one who had actually initiated the road rage conflict that led to his death when he approached Horcasitas’ car. Unsurprisingly, the use of the AI simulacrum became the grounds for Horcasitas’ legal representation to successfully argue for an overturning of the sentence (though not of the conviction). The appellate court’s statement to overturn the sentencing offered this line of reasoning:
It [the AI] erases the interpretive distance between the family’s belief about what the victim would have said and the victim’s own voice and opinions. . . Indeed, the AI victim stated, ‘This here is a true representation of who I was, you know, not how the courtroom portrayed me,’ [and] No disclaimer could cure that error.
Notice that even as the statement disavows similitude between the real Pelkey and the AI simulacrum, it nonetheless refers to the simulacrum as “the AI victim.”
These interlocutors’ uncertainty — Wales’ ambiguity about the relation of the AI simulacrum to her actual brother, and the appellate court’s semiotic collapse of Pelkey and the AI under the single identifier, “victim” — points to the disruptive power of prosopopoeia, a rhetorical figure that confers a voice upon the voiceless. We see it at work, for instance, in children’s media that feature talking animals, and in the eulogies of the bereft when they make claims on how the deceased would want mourners to interpret the meaning of their death.
However, in her Zoetropes and the Politics of Humanhood, rhetoric scholar Alison Rowland emphasizes “the importance of face for prosopopoeia,” the etymology of which derives from prosopon poien: in Paul de Man’s translation, literally, “to confer a mask or a face” (97). Rowland argues that prosopopoeia is a critical trope in the rhetorical making of existents as legible participants in the bio-ontological hierarchies that structure societies.
Put more straightforwardly: beings matter more to us when they have a face, or when we can at least imagine them to have a face. This simple insight, Rowland suggests, animates much of our public discourse about what sorts of beings get to “count” as human. (Think billboards by pro-life advocacy groups that represent 11-week-old fetuses as infants or even toddlers.)
Through Rowland’s lens, the true trouble with AI Pelkey begins to emerge with more clarity. Giving voice to the dead is not a new practice. Neither, to some extent, is giving face to the dead — there is a reason, after all, that we mount photos of our loved ones at funerals, and why some choose to hold open-casket services. Centuries if not millennia of human material culture archive the ways in which we have tried to en/face the dead (death masks, sarcophagi, and photos being but a few examples).
What is new then about this technology is its ability to enact both dimensions of prosopopoeia, to en/voice and en/face. We know that there are at least two interpretive layers between Pelkey’s “real” self and the AI version of him that appeared in court. We know, in other words, that the AI simulacrum extending forgiveness to Horcasitas is not Pelkey. And yet, for Pelkey’s mourning loved ones and for the sentencing judge, it came close enough that it might as well have been.
A very real concern as AI technologies progress is that they will be used to create simulacra of people that are indistinguishable from real photos and videos. Indeed, we’re arguably already upon this moment and already living inside the consequences of what that kind of technology can do to a social world whose coherence hinges on things like truth, accuracy, and evidence.
As real as the fear of indistinguishable similitude between the human countenance and the AI simulacrum is, though, I almost fear the possibility anticipated in Pelkey’s case more — that we may reach a point where we decide that even where dissimilitude is acknowledged, we proceed as if a deceased person and an AI version of them (no matter how creepily uncanny) are functionally the same being.
In such a world, how do we proceed with the project of justice, if a murder cannot be said to erase a person from the world? How do we deliberate about end-of-life decisions in healthcare if comatose patients can appear as AI simulacra before an ethics board? How do we peacefully pass from the world if a multi-billion dollar venture tech industry stands ready to digitally recall our voices from beyond our graves, with or without our consent? If we know that the machine wearing a face is not the human it claims to represent, yet proceed as if they are in fact one and the same — or close enough that the ontological difference doesn’t matter — how do we live?
In such a world, what meaning, then, do we make of death?
Linked articles
Richard Pollina. Arizona man’s sentence tossed after AI video of dead road rage victim used in court to deliver message to killer. The New York Post, 2 Oct 2026.
Tom Powell. Chandler road rage shooting victim speaks using artificial intelligence. AZ Family, 6 May 2025.
Ivan Pereira. Murder victim ‘speaks’ beyond the grave in AI generated video at sentencing. ABC News, 9 May 2025.
Wendy H. Wong. The Creepy New Digital Afterlife Industry. IEEE Spectrum, 10 Oct 2023.
Featured image courtesy of Denis Doukhan from Pixabay.
