This past May, a judge in Arizona viewed AI-generated “testimony” from murder victim Chris Pelkey during the sentencing of his killer. If you watch the video above, you’ll see that the technology is pretty basic. It’s Mr. Pelkey’s likeness and apparent voice, which is confirmed with included footage recorded prior to his death. But the special effects aren’t seamless – it’s pretty obvious it’s a deepfake, which makes sense since Mr. Pelkey had been killed four years prior.
Interestingly, Mr. Pelkey’s likeness speaks of mercy and forgiveness. He says that he and his killer might have been friends in another life. And at the end, there’s footage of him fishing, suggesting a peaceful afterlife. Rather than his avatar’s script being generated by AI, it was written by his sister who had been thinking about what she’d say in her post-trial sentencing impact statement for at least two years. In crafting the script, she interviewed Mr. Pelkey’s elementary schoolteachers, his prom date, fellow Army servicemembers, friends, and other family.
My team discussed this case briefly, and one worry was that judges might put more stock in testimony of this nature than is deserved. We can’t know for sure what murder victims might have wanted. Judges realize this, but might be inappropriately swayed to grant more weight to speculation from an AI-generated video figure as opposed to a family member. In this case, Mr. Pelkey’s avatar suggested a possible desire for leniency. But there’s no reason to think that would always be the case were this to become common practice. And perhaps driven by the Mr. Pelkey’s avatar’s warmth and personability, the judge ultimately sentenced his killer to 10.5 years in prison, which was 1 year more than prosecutors requested.
Further, as broached by the case writers, usage of speculative testimony from a deceased victim opens the door to using the same from deceased, uncooperative, or hard-to-locate witnesses. A prosecutor or defense attorney could imagine what a witness might have said were they able to testify (and probably imagine testimony that would be most helpful for their case), and pass off hearsay as firsthand testimony using a similar AI-avatar technique. The jury could be reminded that such a deepfake witness wasn’t real. But they’d likely still be more emotionally swayed than warranted.
As you continue to think about the stakeholders, benefits, drawbacks and risks of allowing deepfakes in the courtroom, work through coach Michael Andersen’s excellent study guide below, which includes a link to a Court TV interview with Mr. Pelkey’s sister.