Scott Caravello: So let's start with the facts, and then we'll get into your point, Katherine, about the judge's comments and why this was fundamental error that required the sentence to be vacated. But so back in 2021, the defendant, Gabriel Horcasitas, was stopped behind a truck at a red light at a busy intersection in Chandler, Arizona, and he repeatedly honked his horn. The truck's driver got out and raised his arms and started yelling as he walked towards the defendant's car and was sentenced. And within seconds, Horcasitas shot and killed him. He claimed self-defense, telling the police that the man charged his car and grabbed his window frame. But there were several eyewitnesses who said that the victim never got past the back of his own truck, and no one saw him touch the defendant's car.
Katherine Forrest: Right. And there were then two trials of this road rage incident, and, you know, Horcasitas shoots and kills a decedent, and they had two trials. And so the first trial ended with a split verdict because the state had failed to disclose some text messages from the victim's phone that it had actually been required to turn over under a criminal law doctrine called Brady, where you have to turn over any evidence that might be—and I'm not using the technical jargon right now that the courts would use, but—potentially exculpatory. So they had a second trial, and the second jury convicted the defendant, Horcasitas, of manslaughter. And that was actually his second conviction for manslaughter. And then on appeal, which is where this comes up, he for the first time raises a challenge. Well, not only to some additional texts that had been excluded, but also to the use of the AI-generated video at sentencing. And so I should say what happened was he was convicted of manslaughter and then sentenced. And we're going to get into a little bit of how that all came down, but the appeal was on both text messages and on the AI video aspects.
Scott Caravello: So where does this AI video come in exactly? But it's at the sentencing stage, where victims are allowed to give statements. And so some context: after guilt's been decided, victims, or, you know, in a homicide case like this, the victim's family, can give victim impact statements. And so that's their chance to actually tell the judge how the crime that was at issue affected them. And I know we'll get into a little more detail when we talk about the opinion about why exactly that's allowed and, you know, what the rules are actually around these victim impact statements. But that's just some helpful initial level setting. So here, you know, friends and family spoke about the victim's kindness and the pain that his loss caused them. His brother read a statement that the victim had written about his faith, his values, and his hopes. And then his sister spoke. And, you know, she said, "He's not even here to tell us what he would think about it, but I imagine it constantly." And then the court played this video.
Katherine Forrest: Right. So the video, it's like a deepfake, but it's using and used an avatar of the deceased victim. And so it was an AI version of the victim explaining and very clearly stating that he's a version of himself, quote, "recreated through AI." But it was really done, at that time, given that this is, you know, sort of several years ago now, it was really pretty well done. It does look, you know, the mouth and the facial features are all moving at the same time. So the victim starts to talk in this AI avatar version of himself. And then it cuts, actually the video cuts, to real footage of him that had been filmed at some time in the past when he was actually alive. So it's sort of a mixed video, real footage of him talking about his faith. And then the AI version comes back and says, with the AI avatar making this statement. So it's the decedent. It looks like the decedent, sounds like the decedent, and says, "This here is a true representation of who I was, you know, not how the courtroom portrayed me," end quote. Now, remember that after the first trial, there had been these text messages that had been turned over, and then there had been some statements about some rage issues, anger issues, that the decedent had, although there were other text messages that said more about that that were actually excluded from the trial. And the victim, in his AI avatar self, says, turns to sort of, of course, you know, Horcasitas in his avatar version at sentencing and says, quote, "I believe in forgiveness and God who forgives," end quote. And then he says to his family that, "I'll see you," quote, "on the other side," end quote. And so, you know, according to the reporting at the time, this seems to be accepted. His sister actually wrote the script, and the video was built from a single photograph. And this was known. Nobody was suggesting that the avatar was real. It's acknowledged that it's AI or that, you know, he had written a script beforehand.
Scott Caravello: But yeah, I mean, so just all those quotes that you threw out, you can sort of, you can really see the emotional impact that this sort of thing might have had. And, you know, the judge reacted to it, as you mentioned, Katherine, and that that played into the appeals court decision, which we'll get into. But he responded or said after watching it that he loved that AI, and then he felt it was genuine. And then he pointed to the victim's obvious forgiveness. He said that, you know, the family had allowed the victim to speak from his heart, and also noted that he didn't hear him asking for the maximum sentence. But even and after that, the judge then went on to impose the sentence of 10 1/2 years, which was the maximum available for the crime charged, or the crime, rather, that he was convicted of. And, you know, the family's lawyer had asked for that sentence, but the prosecutor had actually only asked for nine years, while the defense asked for seven. And the judge himself acknowledged that it was more than the state asked for.
Katherine Forrest: Yeah, you know, I, let's just pause on this for a second, which is the judge saying the words, "I didn't hear him asking for the maximum sentence," right? I mean, it's like the judge for a moment sort of believed that the app, knowing that the avatar is in fact an avatar and is not the defendant. The judge had a moment of actually sort of getting caught up in thinking that it was somehow truly representative of what the decedent himself would have said. So, you know, victim impact statements are a really important part of the criminal justice system and of sentencing, and victims are always allowed and entitled to speak. It goes back to a Supreme Court case called Payne, Payne v. Tennessee. So you can have a victim impact statement, but typically, if the, always, if the victim is alive and wants to make a statement, because they don't always. In fact, I would say when I was sentencing individuals, while I was a judge, and I sentenced, you know, hundreds of people for different kinds of crimes, when there was sort of an identified victim, hardly anyone, actually hardly any of the victims, actually wanted to speak. Sometimes they would, and sometimes you'd get family members who would want to speak on behalf of the defendant. That was actually quite common. But the idea is that if you do have a victim who speaks, that the victim is able to talk to the court about the impact on his or her life circumstances, you know, what have you, and, you know, it can show sort of the circumstances in perhaps a different way, because they can also talk about the event at issue.
Scott Caravello: And so, you know, the Supreme Court has specifically held that the Constitution doesn't bar this victim impact evidence. And so then Arizona itself actually gives a lot of room to victims to give these statements. And its constitution actually has a Victims' Bill of Rights that gives victims the right to be heard at sentencing. And importantly, you know, they're not treated as witnesses. So they're not taking an oath. They're not cross-examined, and they don't have to disclose their statements to anyone in advance. You know, they can speak to the issue, as, you know, that's their circumstance. And the ordinary rules of evidence also don't apply at sentencing, which means that, you know, the statements can be heard and considered without the same sort of rules and structures that go into other evidence that's put before a court when you're, you know, deciding actual guilt. So the judge can consider, quote, "any reliable, relevant evidence, including hearsay," you know, when you say that you heard someone say something else, rather than something that's presented directly through, you know, firsthand observation. And so the thinking is that a sentencing judge should see the whole picture and then can be trusted to weigh each piece of it.
Katherine Forrest: But there are some rules that come into play, like if the judge is going to make a finding of fact about a particular, you know, piece of evidence, and that fact is going to play a role in the sentencing. Then there has to be a specific finding, typically stated on the record. It's got to be by actually only a preponderance of the evidence, doesn't have to be beyond a reasonable doubt, and there's lots of law in this area. So, you know, it's not unlimited for the judge to consider anything. And, you know, the defendant does have due process rights to be sentenced on both accurate information, but accurate information that is found in terms of the particular standards in state or federal court or by the Supreme Court. As I said, factual findings need to be by a preponderance of the evidence. And, you know, here it reminds me a little bit of our grief bot episode, because what if you're able to have a grief bot who's been all trained up come in and talk about what they would be thinking after their trial and at the sentencing of the, you know, defendant? And so it's an interesting, you know, situation that the Arizona court was faced with.
Scott Caravello: Yeah. And so just adding on to that, as we sort of get into the meat of the actual ruling, I would just note that the defense counsel didn't object to the video at the hearing. So when the appellate court reviewed it and then ultimately vacated the sentence, it was only reviewing for fundamental error, which is actually, you know, a more demanding standard than they would have actually been subject to had it been challenged at the time that it was played. And the court did in fact find that the video being sentencing was fundamental error.
Katherine Forrest: OK, so when the Arizona Supreme Court was looking at this sort of appeal that was now raised at sentencing for Horcasitas, they looked to actually a death penalty case. It was called State v. Rose. And in death penalty cases, you can have, you know, circumstances where the jury actually considers certain kinds of mitigating evidence for certain purposes, which we won't get into. But when the jury was being shown certain kinds of mitigating evidence, they were shown, in this State v. Rose case, the victims of the murder, the young sons of that victim, standing near their father's grave. So that was a different scenario. It wasn't at the sentencing of a manslaughter conviction. It was in a death penalty phase, where it was actually the jury being shown certain kinds of evidence. And the Arizona Supreme Court called that kind of evidence, of showing the victim's young sons standing there at their father's grave, as uncomfortably close to the line where talking about victim impact can become, you know, unduly prejudicial. But it didn't find error. Now here, the Court of Appeals in this case, in the Horcasitas case, said that the AI video did cross the line. You know, the photographs in Rose actually captured something that really happened. But the AI video, in the court's words, and this is now the appellate words, didn't reflect actual events. It was a depiction of the victim and his thoughts created from, and they used the word, "imaginings" of his sister. They did not mean, I'm sure they didn't mean that in an insulting way, but what they meant was it was not real. And so you can't take it to be real.
Scott Caravello: Yeah. And so there's this one noteworthy line from the opinion, which I think gets at the heart of the issue and the presentation of the AI video as if it were, you know, a direct statement from the victim. And I think this gets back also to your point earlier, Katherine, about the grief bots, where something's trained on, you know, all the data about a person, because the issue here is that it was presented as if it was a statement directly from him when it wasn't, right? It was influenced or, you know, driven by the sister. But so what happens when it's actually trained on the data about a person? And it could perhaps more accurately be described as a statement directly from the victim. But anyway, back to that line. The court said the video, quote, "erases the interpretive distance between the family's belief about what the victim would have said and the victim's own voice and opinions." When the sister said that she imagined what he would say, she was speaking about him, and everyone in the room could weigh it as her belief. But this video instead was putting her words in his mouth while using his voice, his face, and his expressions.
Katherine Forrest: Right. And the appellate court made clear that no disclaimer could cure the error, that it wasn't as if there was something that should have been said that would have cured the error, that even though the video was disclosed in advance, it was. And that the sentencing judge's own words, in which he credited the video as coming directly from the victim, sort of demonstrated their point. He pointed, as we know, as we were just saying a few minutes ago, that, you know, the judge actually pointed to the victim's obvious forgiveness of the defendant, that he didn't hear him asking for the maximum sentence. And the judge was considering it all, as we've said, as if it was the victim's own statements. And so that, it's troubling, because I think that what the appellate court is saying is, even if you made a disclaimer, it's just, there's too much potential for problematic consideration here. And so in throwing the sentence out, the appeals court said that they didn't know exactly how the video had factored into the sentencing, but there was no question that it played a role. And I think that, given the judge's comments, there was no question that it played the role, played some role. You know, who knows how much? And the judge did give him the maximum sentence, even though the prosecution was asking for a year less, although, of course, the family and others in the courtroom had asked for the maximum sentence. So anyway, you know, it's tough stuff to use these avatars.
Scott Caravello: No doubt about it. But I think, and, you know, maybe we can kind of close on this point, Katherine, and just sort of bringing this to a 10,000-foot level, talking about the general debate over synthetic media and deepfakes, is that a lot of the policy conversation around deepfakes has to do with detection and labeling, on the theory that, you know, if we know something is synthetic, we're protected. But, you know, in this case, I think, as you were talking about earlier, no one was fooled by this video. We knew from the get-go that it was an AI-generated video, and it still left at least the judge with the view that the words were coming directly from the victim. And so there are studies, and, you know, I don't want to make it seem like it's, you know, total consensus, but there are at least some studies that are showing that participants rely on a deepfake video when judging something about someone, even after being explicitly warned that it's a fake, and even when they said they believe the warning. So, you know, the old saying that seeing is believing still seems to be true, even if we're presented with evidence that does suggest otherwise.
Katherine Forrest: Yeah, you know, and it's a really slippery slope, because this video that was used at the Horcasitas sentencing was made now a couple of years ago. And the capabilities of the video generation technology have gotten to be so much greater now that you can only imagine how much better one of these videos could be. And, you know, the slippery slope is that this was just one where the decedent was saying something that sounded like he was, you know, talking about forgiveness and things of that nature. But you could imagine that there could be one that would be created where they're talking about, "Why did you take me away from my family? Why did you cut my life off, you know, at such an early stage? I had so much more, you know, living to do." And you could have really powerful, impactful, sort of deepfake videos that could then be used at a sentencing that could show sort of the pain in the eyes, if you will, of the decedent. So I don't know how the appellate court really could have done anything but say no to this, because it wasn't the victim. And if his sister wanted to talk about, as she did, her brother's faith and about the impact on her, that's one thing. And she could say, you know, "Knowing my brother as I knew him, I think that he would be forgiving, and he'd want us to all know that he's going to see us on the other side." That's one thing. But it's very different when you've got the decedent actually talking to you, because we could turn that around, and there could be very different circumstances. So I have to say, I think this was the right decision by the appellate court. I have really no doubt about that.
Scott Caravello: Totally agree. Like you said, can't really imagine them coming out any other way.
Katherine Forrest: OK, well, we're going to let Scott go now and go off to the Caravello offerings throughout the, apparently the...
Scott Caravello: Family reunion.
Katherine Forrest: The family amily reunion that people don't even know is about to occur all over. What town are you in?
Scott Caravello: Palermo.
Katherine Forrest: Palermo. All over Palermo. Right. So you're going to be wandering all over Palermo, running into Caravellos, you know, looking at, you know, you're going to have to get something at this window frame shop. And hopefully you don't have to go to the gastroenterologist who's the Caravello or whatever. Right?
Scott Caravello: Exactly.
Katherine Forrest: All right. Well, enjoy, and all right, folks. That's it. Signing off. I'm Katherine Forrest.
Scott Caravello: And I'm Scott Caravello. Don't forget to like and subscribe.
