Katherine Forrest: Yeah, yeah, yeah. We have to get into it because with five topics, no dilly dallying. All right. Let's do the first one. So the AI executive order that wasn't, at least hasn't been yet, shall we say. So on May 21st, everything was set for this big White House signing ceremony for this, what we thought was gonna be an AI order. And in fact, you and I were thinking of pivoting and actually recording a special episode on that as soon as it came out. And tech CEOs had been invited, the press had been invited and briefed that morning, and apparently the order was typed up and ready for signature.
Scott Caravello: Yeah, and then just hours before the ceremony, President Trump pulled the plug on the whole thing. He had told reporters that he didn't like certain aspects of it, and he was worried that it would interfere with America's competitive edge in AI, which, you know, as we've discussed many, many times, is really kind of the overriding theme of a lot of the administration's activity on AI. And so his quote was about the canceling of the order was that we're leading China, we're leading everybody. And that he didn't want to do anything that's going to get in the way of that lead.
Katherine Forrest: Yeah, so, I mean, I agree with you. I think that from the July executive order, then the December executive order, then the March executive order. We've seen sort of the signs of this. So, it's not surprising that he will at least wait until he gets it to the point that he thinks it's consistent with many of the things that he's already said. But what was in the order that almost was and still might be? And we only have reporting based on a leaked draft. And so I just want to caution people that, you know, this is not an executive order. This is just reporting on what might have been. So, first there was going to be a cybersecurity section, apparently, that would have directed agencies to shore up their defenses against AI-fueled cyber attacks. And second, there was going to be a part and this is sort of what apparently concerned the president, that had a voluntary framework for AI developers to share advanced models with the government for a kind of security review prior to public release. And there was gonna be up to a ninety day pre-launch review window. So, it was gonna be voluntary, but it was gonna be a pre-launch review.
Scott Caravello: Yeah, and that's what I think is so interesting about all of the changing positions, right? In that whole voluntary aspect, because it stated that nothing was going to be construed to authorize the creation of a mandatory licensing or permitting requirement for new AI models. So, in that way, it actually resembled the agreements that the federal government had already made with some model developers like Microsoft, Google, and XAI for early access to their models.
Katherine Forrest: Right, but even a voluntary framework isn't being approved right now, and maybe it will be, but it wasn't obviously signed last week. And according to the reporting, the main reason that the okay, I'm gonna do that according to the reporting again. According to the reporting, the main reason the signing was delayed was that Trump dislikes regulation and his AI advisor, David Sachs, also had some concerns. And there were reports that some tech executives had actually given it a thumbs down in discussions leading up to the signing date. So ultimately the president's instinct to avoid something in the AI sphere that could be perceived as government screening of commercial AI models, that actually won the day. So it's unclear when or even if the order will be issued and how it might be revised if it is issued.
Scott Caravello: All right, so lightning round topic number two. You know, we we've talked so many times about White House executive orders. One thing that we cover a lot less on this podcast is papal encyclicals. And so that's gonna be lightning round topic number two. And so what is an encyclical? Well, it's a it's a letter written by the Pope to clarify the Catholic Church's teaching on a topic. They are traditionally letters to Catholic bishops and the wider church, but recently popes have addressed them to all people of goodwill, meaning everyone. And so they can shape Catholic thought for generations and influence debates well beyond the Vatican. So on May 25th, Pope Leo the Fourteenth released his first encyclical titled Magnifica Humanitas, which is Latin for magnificent humanity. And it's not a brief statement about AI, it's a forty-eight page document with over 40,000 words, a sweeping document on, “safeguarding the human person in the time of artificial intelligence.”
Katherine Forrest: Right. And you know, I actually want to applaud you for your Latin pronunciation there. You know, that was pretty good.
Scott Caravello: Thank you.
Katherine Forrest: Did you do one of those tools that tells you how to pronounce things?
Scott Caravello: I actually did not and I will say I never took a single day of Latin. We can just chalk it up to law school, I guess. I don't know, but it just rolled right rolled right off the tongue. Yeah. Exactly.
Katherine Forrest: Just your brilliance.
Scott Caravello: Exactly.
Katherine Forrest: Just your brilliance. Because I'm about to really screw one up. And my son, I have a 24-year-old son, which also tells you how old I am. I have a 24-year-old son who was a classics major. He actually taught himself ancient Greek during COVID, okay?
Scott Caravello: Wow.
Katherine Forrest: So, some of us were learning how to do cocktail mixing…
Scott Caravello: Right.
Katherine Forrest: …and my son was learning ancient Greek. Anyway, putting that aside, okay, so this encyclical is significant for several reasons. First, Pope Leo chose his papal name as a reference to Leo the Thirteenth who published something called Rerum Novarum, and I'm afraid of about how I pronounce that. Okay, Rerum Novarum, which was it that translates into of new things, and it was published in eighteen ninety one, and that was a famous encyclical on workers' rights during the Industrial Revolution, and Leo the Fourteenth signed Magnifica Humanitas on May 15th, which is exactly one hundred and thirty five years after Rerum Novarum had been published.
Scott Caravello: I just want to say you definitely got the more difficult one to pronounce. So. .
Katherine Forrest: Yeah, that's another way of saying I didn't get it right.
Scott Caravello: I have no idea. I have no idea. Anyway, but yeah, he's explicitly drawing a parallel between AI and the Industrial Revolution
Katherine Forrest: Right. And his core message in this encyclical is that technology is not inherently evil or antagonistic to humanity, but that it's never neutral either. The Pope calls in this encyclical, he calls for a robust set of legal frameworks that have independent oversight, inform users, and a political system that does not abdicate its responsibility. He warns that without stronger safeguards, AI could deepen inequality, weaken human agency, and shift critical decisions out of human hands. So there's a lot to be discussed in all of that and a lot of debate to be had on all sides of those issues, but that is the content of what he was writing.
Scott Caravello: Yeah, and then there's also a significant focus on statements about the use of AI in warfare.
Katherine Forrest: Right. He broadly rejects the just war, in quotes, theory. You know, the idea that there's a justification for why and how wars are fought in favor of dialogue and diplomacy, and we've seen that actually in other actions that he's taken. And he goes on to explain that the growing ease with which autonomous weapon systems can be deployed makes war more feasible and less subject to human control. Which conflicts with the idea that armed forces should only be used as a last resort and for legitimate self-defense. And then following that the encyclical instructs that AI that is used in warfare should be subject to the most rigorous ethical constraints.
Scott Caravello: And you know, one sort of thing that was kind of remarkable about the whole thing and adjacent to the actual encyclical itself was that the Pope presented it alongside Christopher Olah, who is a co-founder of Anthropic. And Olah acknowledged that AI labs face strong commercial pressures, but said, “We need informed critics who will tell the labs when we are failing. We need moral voices that the incentives cannot bend.”
Katherine Forrest: Right. So whether or not you're a Catholic or religious at all, you know, we've got a leader that is a leader of a one point four billion member global community that is fully endorsing a position on AI development and regulation. So that's a big deal. And we'll see how it impacts the conversation. But let's keep on moving. Lightning round, let's go to number three.
Scott Caravello: Okay. So, Gavinner Newsom in California's AI Executive Order. So, this one was issued–
Katherine Forrest: Wait, you know what that was? You know what you said? Don't even erase this. Don't even like edit this out…you said Gavinner Newsom.
Scott Caravello: No I didn't. No I didn't.
Katherine Forrest: And that's like a beautiful… yes, you did! You did say that.
Scott Caravello: Oh my gosh.
Katherine Forrest: Okay, the listeners have to listen to that themselves. Don't take it out. Robert, don't take it out—our producer. Okay.
Scott Caravello: Wow.
Katherine Forrest: So, you put Gavin Newsom and Governor Newsom together and came out with Gavinner. And I–I just think that that's beautiful, okay? After you got Magnifica Humanitas all correct…
Scott Caravello: Thank you. Yeah.
Katherine Forrest: …Gavinner Newsom came out of your mouth.
Scott Caravello: It's all about balance.
Katherine Forrest: All right. Okay. Go for number three again. Here we go. Gavinner Newsom.
Scott Caravello: Gavinner Newsom in California's recent AI executive order. On May 21st, it is the first executive order by any US governor, let alone a Gavinner, specifically aimed at confronting the economic impacts of AI on workers and small businesses.
Katherine Forrest: Right, you know, what does this do? It directs the state to expand job training programs, very important. And it has a particular focus for white collar workers like customer service reps, software developers, marketing professionals whose roles may be eliminated by AI. What's going on with our pronunciation today? All right, all right. And I find the the focus on the white collar workers to be really interesting. And it calls for a review of labor policies and safety nets and notably orders an examination of what's called universal basic capital, which is the idea of giving residents stakes in assets like corporate stocks, bonds, or wealth funds.
Scott Caravello: Yeah, and this is this is significant because of course if California was its own country, it would have the fourth largest economy in the world and it's obviously the epicenter of the AI boom. It's home to thirty three of the top fifty privately held AI companies throughout the world. So California moving on AI policy in this way causes others in and outside the industry to pay attention.
Katherine Forrest: Right, so Gavinner Newsom. Yeah, all right, so here we go. Topic number four, and this is a good one for our lawyer listeners. A Connecticut federal court just ruled that an expert witness's AI prompts are discoverable.
Scott Caravello: And so let me quickly cover some basics on expert witnesses to set the stage. So in federal and state litigation, parties can retain these experts, people with specialized knowledge, training, or experience, who are permitted to offer opinion testimony on technical or complex subjects. And so, unlike fact witnesses who can only testify as to their personal knowledge, like what they saw or experienced during the events at issue in the case, experts are actually able to give opinions on the ultimate issues in the case. Like whether a product was unreasonably dangerous or by putting together a complex model to demonstrate damages that a party did or did not suffer.
Katherine Forrest: And under the federal rules of civil procedure, which governs civil suits in federal court, there are disclosure requirements for experts. They have to produce a written report containing opinions, the basis for their opinions, the data or information that they relied upon, methodology, and the opposing party gets to look at all of this and then depose the expert, put in their own responsive report and challenge it. And sometimes the challenge can also go to the point of motion practice if there's a basis to try and exclude the expert, and that can be called a Daubert motion. And a Daubert motion, if successful, can actually either preclude an expert's testimony entirely or part of the expert's testimony.
Scott Caravello: And so here's what happened in this case. It was an action under the Clean Water Act, and an environmental advocacy group had retained an expert witness, and then it came to light that the expert had used an AI tool in crafting her expert report. So, the opposing party moved to compel disclosure of the prompts that she used.
Katherine Forrest: Right. And the court ruled that the AI prompts used by the expert are discoverable as part of her methodology and reasoned that because methodology of an expert is discoverable, then the fact that she relied on certain information from these AI tools and reaching her conclusions meant that her prompts were fair grounds for discovery.
Scott Caravello: And so this is a really important development for litigators. And actually we should probably also mention the case caption, which is Conservation Law Foundation v. Shell. And so if you're retaining experts who use AI in any part of their analysis, those AI interactions are potentially discoverable. But again, this is, you know, one case from one district court and we'll see if others follow suit. But it means that experts do need to be thoughtful and deliberate about their AI usage. They need to document it, they need to be prepared to defend it, and they need to understand that opposing counsel might see it because future courts may order similar discovery.
Katherine Forrest: Yeah, you know, I'm sure there are going to be courts that are going to come out different ways on this question and so there'll be people and litigants taking different positions on whether or not there's an automatic hold on prompts or not, because people can, as we all know, you know, set their settings so to speak for their LLMs to do not retain or retain for short periods of time and things of that nature. But it's gonna be interesting to see how this really starts to develop. But let's go on to the last topic… have you take us home, Scott. It's the Gemini Omni model.
Scott Caravello: All right, I'm on it. So on May 19that Google IO, which is Google's massive annual developer conference held every spring, Google unveiled Gemini Omni, which is its natively multimodal model for AI video generation, which Google says can create anything from any input.
Katherine Forrest: Right, and so you can feed it text, images, audio, video, and you can do that in a single prompt and it reasons across all of them to produce a coherent video output. So, this is not just stitching inputs together. It's combining Gemini's reasoning intelligence with video generation capabilities and what it describes as an intuitive understanding of physics. And you know I love physics.
Scott Caravello: Sure do. And like we've talked about so many times with video generation models and world models, it's that understanding of the physical world that's so important. The model needs to know what happens next so you don't get objects appearing and disappearing or flying across the screen, or even Humpty Dumpty just, you know, automatically putting himself back together again. But then what I think is really interesting about Omni is the conversational editing. Once you have a generated clip, you can describe changes you want to it in plain English, like shift the camera angle X-way or change the background to a city street. And Omni reworks that specific element while keeping everything else intact. So, the characters and physics can stay consistent and the scene remembers what came before. And you know, I think that adds a lot of value compared to having to re-prompt and regenerate output until you get it right each time, especially because the model might, you know, change up a lot of things in between the different prompts.
Katherine Forrest: Yeah, you know, when things are consistent across frames and across scenes, you know, it's called continuity. But one thing that all of this makes me think is because you can do it with audio through the microphone, you can imagine that there are gonna be people literally sitting in those directors' chairs. You know the kind of director's chairs I'm talking about? And they're gonna be sitting there and they're gonna be watching their monitor and they're gonna be sitting there literally you know, able to sort of click on their microphone, move the camera back, let's roll, and it's gonna almost be like they're directing a movie right there with the camera using some of those audio features. And there's an avatar feature also where you can create a digital version of yourself through a verification process and then generate videos that look and sound like you. So you can be the star of your own movie. And all videos carry Google's SynthID digital watermark, which is spelled s-y-n-t-h-capital “I”-capital “D”, digital watermarks so that they can be verified as AI generated. But here's the thing. Google deliberately withheld the ability to edit speech and audio in existing videos. They said they're still working to test this and to better understand how we can bring this capability to users responsibly.
Scott Caravello: And so right now the model is available in the Gemini app for paid subscribers and then for free on YouTube Shorts. And the current cap on the videos you can create is actually just ten seconds right now. But Google has made clear that that's a decision it made about deploying the model rather than some built-in limitation of the model and the technology.
Katherine Forrest: Really interesting stuff. And we're starting to see short form videos actually coming out with some of the other video generation models now where they're putting together, you know, entire storylines, fascinating stuff. So that's our lightning round, folks. Five major developments in one week and we had almost an executive order. A papal encyclical, a state level focus on AI and workers, particularly also now bringing in white collar workers, and a court ruling on AI discoverability with regard to experts, and then a new frontier in multimodal AI… it's a lot. And that's all we've got time for today, folks. I'm Katherine Forrest.
Scott Caravello: And I'm Scott Caravello. Don't forget to like and subscribe.
