FTP Episode 1 - New Podcast! Sora, AI Pilot Projects, OpenAI Chat Logs, Tilly Norwood

Hello, and welcome to the Future Tech and Policy Podcast, or FTP, our first episode of a new experiment where we will discuss current tech news, trends, and possibilities with an eye for policy and social impact. We're hosted by the Institute of Government and Public Affairs, an institute housed at the University of Illinois System. Joined by faculty affiliates at all three UI system universities in Urbana-Champaign, Chicago, and Springfield, IGPA conducts nonpartisan research and brings faculty expertise to policy discussions regarding the top issues facing Illinoisans. With this podcast, we have the same goal of bringing faculty members together at all three system universities in Illinois to discuss the big tech issues of the day, tech news, tech trends, anything that's happening with an eye toward the future and a positive impact of tech on society. I'm your host, Jeremy Reel. I'm an assistant professor of educational psychology and technology at the University of Illinois, Chicago. I'm also the co-lead for the Science and Technology Working Group, and I study AI in education and how it can be used for teaching and learning, as well as helping people learn about AI literacy skills and technology fundamentals to help them think about AI in this kind of new ecosystem, new landscape of learning and teaching. I'm joined by my good friend and colleague, Kansa, today, who I will let introduce himself. Hey, good morning, Jeremy. My name is Kansa. I'm an adjunct professor at the University of Illinois, Chicago School of Law, where I teach AI law and policy. It's a relatively new course. I think we're in our second year of teaching. Prior to that, I taught the intellectual property and intellectual property comparative law courses at UIC Law. I'm in private practice. I'm a partner at a local law firm, where I practice centers around advising clients about emerging technologies, whether it's AI, data privacy, or intellectual property. Great. So thank you for joining me today. I'm hoping this becomes a regular thing. We're going to have some fun. We're going to chat tech. But for today's show, we're looking through the news. Ken and I were sifting through everything, and we're just always seeing a lot of AI, AI, AI, AI. That's dominating tech news. It's in the news every day, something new every morning when you open your feed. So we're going to hit some of the headlines that we saw, and we're going to talk about it in terms of our perspectives and implications for policy in society. So let's get started with the first article. All the links to any of the articles we reference are headlines. We're going to put in the description section below. So go ahead and check those out. And we're going to get started. So the first is Sora has released their Android version. There was a big waiting list for this. They headed out for iPhone for a while. Sora said that up to half a million installs on the first day alone of this new social network for AI-generated content. And it's all about video content for the most part. But it's kind of like, if you think about it, TikTok or Instagram for AI-generated video. And it's kind of crazy. So let's think about that just for a second. A social network dedicated to AI-generated content. There's a lot to unpack there. But also, they brought in a whole new app ecosystem this week with Android's release. And I myself, I'm a Google Pixel user, so I was able to sign up for it and try it out and check it out. So it's just lots of AI-generated videos. So what are your first thoughts? Yeah, I think the first thought I had was, you know, I spent some time in Silicon Valley earlier in my career. And my colleagues and I always had a joke every time there was like a new startup. It's like, oh, this is, you know, LinkedIn for Periketz or Facebook for puppies. You know, it's like ways to kind of create new social networks for very niche audiences. But I do think it's really fascinating because on the very positive end of things, it creates an environment only for AI-generated content. And so, you know, the big concern, of course, or one of the big concerns is that this AI-generated content, we don't know it's real anymore. And, you know, particularly with students in the younger generation, they may have more difficulty discerning what's real and not real and what the long-term side effects of that may be. Well, now we have an app. When you go in that app, it's not real. This is an assumption we make. We go into this app, we know it's fake. Exactly. We hope everybody knows it's fake. Like actually, you should be surprised if it's not generated, right? So I think it's a great idea in that sense. You know, there's obviously a lot of concerns, though. And I don't know where they're landing on how to use intellectual property rights or protect against them. But our IP laws are not designed to work well with these kinds of technologies, right? It's especially in the U.S. It's a, hey, you do things, you innovate, you push forward, and then if you broke something, you pay later. You get in trouble later. Yeah, exactly. So if you're an established business, if you're Disney, yeah, you're going to go and license things because you have a lot to lose in the back end. But, you know, even like thinking about Uber, they were in dozens of cities just openly breaking the law, right? And they went on the back end and they paid their fines and legitimized themselves once they kind of proved their business model. So, you know, there's a lot of concerns. So, you know, initially, I think they were doing an opt-out model. So, yeah, yeah. And I don't know if you'd heard about, heard about the IP issues, but, you know, you can go and soar and say, hey, make me a Disney princess. And it makes you a Disney princess. Well, very similar to a lot of the diffusion models we've already seen already with, you know, photo generation and things like that, where I can become a Simpsons character from a Miyazaki movie overnight. You know, just give me 10 seconds and you got yourself a photo. So, yeah, I know that that's a concern, but I haven't got to really play with the, you know, try to break the IP stuff. I'm sorry. Yeah. Yeah. And I don't know how good it is, or how good it could be. Just, you know, just the way that we manage IP rights. Other than music, there's some mechanism for music going back to the Napster days. But, you know, for everything else, it's, there's no central catalog that would work for this function. It's like allowable IP people that have licensed their IP to be, you know, kind of thing. Yeah. If you go to the copyright office, there's a database, but if you can go and tell me what that copyright registry actually covers, then, you know, you would have an entire business model of your own on your, so. So, I mean, I think there's a lot of concern there. And I was actually on a panel last Friday with some folks who represent more of the creative industries, actress unions, stations that generate content. And they're really concerned, as they should be, right? What do we do with all the things that we've crafted as an art? And how are they actually going to police this? And who's going to police it? And who's going to have the incentive to do that? So, yeah. So, you know, again, big upside promise of this, like, app that, you know, specifically identifies generated content. But then, of course, we have to figure out how to actually make it work. Yeah. One of the things that caught my mind was my attention was on this, you can put yourself in the videos. So, you can't use other people, but the idea is that you can still put yourself in a video. And I'm wondering where the safeguards are on that, right? So, not only, you know, there's, you know, of course, creating content with yourself in a video, but how does it know it's you? Are you putting other people in videos in some way that, like, I'm just the verification process, the whole way of ensuring that others are not in these videos without, you know, consenting to it or participating in the social network aspect of Sora was just, I wouldn't say red flag, but it was an immediate just, you know, thing I was thinking about, especially as we get to minors maybe using the app. Yeah. And I think all of this kind of, it breaks our notion of privacy. So, you know, in the US, you know, the Europeans think of it slightly differently, but in the US, we have like two different notions of privacy. And one is like for criminal purposes, right? Like, what I do inside of my home with the curtains drawn and that's, that's your right to privacy, like what, what people think of. And then there's private information about me. And that's, you know, specifically protected information, regardless of if it's out there in the public. But your image and your likeness and how you walk and type and do all kinds of interact with people, that is not private in under our current system. So to your point, you know, I get tracked every day on my gaze, my walk, my pace, you know, I'm wearing a Fitbit, you know, this is all data being collected. My unique, as they would say, biometrics and just a different way may not be my face, but, you know, there's other aspects. Yeah. And how are these systems going to protect you, right? Or do we want them to? And I don't know if we've, I think, you know, afterwards we don't want it to protect us, but do we want it to, right? And so I just don't know that these questions have all been thought through necessarily. Yeah. And that's why we're bringing them up, right? Make sure we're having those conversations as we go forward with some of this stuff. All right, moving on to the next one. This is related to your parakeets, LinkedIn for parakeets and Facebook for puppies, but 95% of AI pilots and startups failing. There's an MIT article that came out recently that shows that a lot of these maybe more niche AI projects at businesses. So this doesn't just mean startups. This is also just pilots happening at organizations that are giving a try to AI and seeing if they can do it. But the estimate was up to 95% of these are failing. And so LinkedIn for parakeets doesn't seem to fly, right? Is that the point here? Yeah, I think that's right. And but, you know, at the same time, I'm sure you're seeing these articles as well, like investment in AI is still through the roof, right? So these are failing and companies are still investing in AI. Near and dear to my heart, of course, are general counsel's budgets for next year, because that helps me understand what my clients are thinking. And budgets are up for general counsel. So, you know, we see the articles that say, hey, clients are going to demand more efficiency because of AI, and you have to adopt AI. And I think those are correct. But we also see that client demand is going to be up next year. So yeah, it is hard to kind of square the high failure rate with the increase in investments. Unless you start thinking about what kind of projects were they looking at initially, and how are they thinking about them now? So I think that's the key thing there is the projects, you know, how is it being used? And yeah, I've seen headlines. I don't have them right now, but we could find some that a lot of speculation about the businesses jumping into AI, new business startups, things like that, just not producing a product that's really game changing, even though AI is super exciting, even though, you know, chat GBT blew us all away with how realistic it could generate stuff, still finding that what you would be killer app or that use case that really transforms business. We see, you know, coding and programming, software engineering, that's really helping transform, you see it in some writing, but then we also see that the quality of AI writing isn't always the best. So there's kind of this hump that's been hit, and now we're on the down slope there. We're looking for AI to really help us do something like the verbs that we do in our daily work, we want to find that. I think that that's where the matching needs to happen more. Yeah, and you know, to the regulator's credit, I think they're actually helping this process. There's, at least in some states, there's some regulation about how companies should think about AI, whether you're buying a product or developing your own, you have to do what's called a risk assessment. And in that risk assessment, you have to say, this is why I want it, right? Like, and it can't just be this one line, I think it's good, or I think it's fun. You have to make a business case. This is going to help our efficiency in this way for these reasons, and you have to really develop those ideas, and then eventually have them signed off at the executive level. So, you know, these kinds of regulations, of course, clients never like them. But it forces people to say, like, yeah, is this a toy, or is there an actual function that I'm looking for? And when am I expecting to see value? You know, it doesn't have to be today, right? I don't have to get all the value today. But I have to have some value somewhere, right? So you have to see the light at the end of the tunnel, too. It's like, yeah, exactly. So we put this cool toy in, and is everybody having fun? Well, is that measurable? Like, do we? Exactly. So, you know, I often bemoan regulations, but I think these are good ones that help people think about how to use this new technology. Now, you know, whether that ultimately helps us, of course, is different. But yeah, so I do think, you know, just in my day job, I'd see clients grappling with it now. Where is it that we want to invest? And someone brought up an interesting point, this conversation I was at last week, which is, you know, there's no profitable AI company right now. You know, Google is profitable, of course, but they're building off of their prior successes, though, right? Exactly. Exactly. So everything that we're seeing is running unprofitably. So when does that rubber meet the road? And what's the pricing on that? And I think all of that's going to change how we think about AI. Yeah. And I think especially, too, as we see more open source models come out, too, there might be, you know, cheaper to implement locally cases as well. So, you know, more businesses starting pilots, you know, maybe internally, without as much cost, I mean, data center costs still will exist and making sure we do that. But yeah, cool. This one's fun. I know you're going to be all over this one. Court ordering open AI to hand over, what was it, 20 million users worth of chat logs and in a piece of litigation that they are working with? Yeah. You know, let's put the actual litigation aside. And for those who are not lawyers on the audience, you know, as part of any litigation, the other side has the opportunity to ask you for certain kinds of evidence. These are called discovery requests. And in the US, those discovery requests can be very broad. So anything related to this at all in any way. So I, as the person who want that evidence, I just have to justify it in some way. And the person who has the evidence, they have the burden to explain why they shouldn't provide it. One of the reasons is that it doesn't exist. Second reason is that it's very burdensome relative to the value to the case. So, you know, we have to spend $10 million. This is a $50,000 case. We'll just give you $50,000, right? Like, they can make those kinds of arguments. So in this case, just the upshot portion of it, these logs about how users have been interacting with open AI became very important. And there's a long drawn out battle over it over the things that you and I might think, you know, there's some privacy issues. And of course, open AI doesn't want it known that they're handing over interaction histories, prompt histories, whatever you want to describe the mass. But I thought it was interesting, one of the arguments they did not really hone in on was the fact they don't have it. So they're handing over, presumably, anything you and I put into open AI into this litigation. So from a privacy perspective, of course, consumers ought to be somewhat concerned. But also, I think, just from a liability perspective, we might be seeing another Napster type situation. And this is, again, not a law school course, but maybe more law than you wanted. But there's a type of copyright liability where if you're the one asking for the Disney princess to be generated, you could be liable for indirect copyright infringement. And so these logs are a way to find you. And so there's a secondary concern there. So yeah, that's definitely something I didn't consider before. Do you know if when you saw the article, did you see if it was all kind of the whole scope of all of their logs? Or is it just the free accounts? Or is it, do you know Open AI loves to make that distinction between the free accounts and the enterprise accounts or the paid accounts? Yeah, so the article, I don't think differentiated that. And it's also clear that they didn't want every blog of every interaction of every user. So there was some way to narrow this down. I think what's going to be unfortunate is that it's going to, the way it's narrowed down is the most relevant portions for the case. So it's a copyright case and it's going to ask for interactions involving generating characters related to movies, related to movies that the plaintiff owns the rights to. And I think not in this case, but in other cases, what we've seen is that these companies can identify you fairly quickly. So whether it's by IP address or a combination of other things, geolocation, the type of words you use, cookies that may be sitting on your computer, their ability to actually identify you is fairly high. Now, you know, with 20 million users logs, they're going to have to use AI to review the AI logs. Inevitably. Find me all instances. That's right. So yeah, I think this is really interesting. And I'm curious about your take on it, kind of more related to your teaching. Yeah, I think students feel like their interaction with AI is just like sacred thing. So you know, that's 101 when I'm working with AI in my classes with students, we're talking about IP. And I've been talking about IP even before AI came out. You can't just go lift images off the web and throw them into your products, your designs, things like that. Everybody creates. And even you, as you create something, you're generating your own IP. So we have pretty robust discussions because, you know, in academia, fair use gets thrown around a lot. And I don't know the degree to which we all really understand the depth of that if that were actually, you know, taken to court and like how it would hold up. So just being aware of all of the different types of IP that we use in education. And if you yourself or a student, what you're using in your projects, that kind of stuff is something that I've definitely honed in on quite a bit. When it comes to AI, though, we had some great conversations in my classes and over the last two, three years about the kind of these implications that, like I said, you know, making like you said, making a character, I make a Simpsons character of myself, you know, just like, oh, it's funny, it's great, but it's still based on somebody's intellectual property and design. And so we talk a lot about our originality also and just the value that academia kind of places in art, you know, even, but, you know, art, people imitate people all the time. So there's gray areas there even in the, I would say the legal space, but also just kind of the intellectual space of, you know, if I'm a musician and I really like a style, you know, is jazz a copyrightable style or is rock a copyright, everybody's kind of taking kind of similar themes. And then there's the uniqueness of each musician. So all styles come from somewhere, right? So you're going to imitate the original rock stars if you're a rock musician. But yeah, we have lots of conversations about this. I'd say students are actually pretty receptive to it too. But, you know, when it comes down to it, I think anybody's going to still do what they want to do. And these tools make it super easy to do. So when it comes to the data, you know, that's another thing is we have conversations a lot about what you put into the AI, it's being saved. It gets to actually, that's by design in ChatGPT, it gets to know you better, you start to see results that are tailored to you with more and more input and the more you use it. So, you know, it is in there. And you always have to ask the question if it's saved, somebody can get it. We don't know who it could be hacker, it could be the company, it could be a court. In the discovery process, like you mentioned, but if it exists, it exists, right? And it could theoretically be used. So it's just thinking about what we put into AI. So I do add research and we have to be very careful about our how we collect data of people that we work with in our research and how that data then gets transmitted to any kind of third party. And right now, we're very protective about data being sent to third party AI providers in any form of human subjects research, not just in education, but the whole of social sciences. Because, you know, that data leaves our custody. And we're those, you know, the custodians of that data as researchers. So these are also, you know, pretty big concerns thinking about that data living somewhere else every time you make a copy of it, it gets saved somewhere. Even if they say they don't save it, you don't know for sure. Yeah, and there's, I think there's probably not a lot of test cases for this, but there certainly are cases that we're seeing now where the guardrails just didn't work. And there's always some explanation for it. And I'm not, you know, I'm not technical enough to actually understand the explanations. But what I do know is that there are guardrails and they don't always work. And we can always drill down as those and analyze them and make sure we're being sophisticated in how we look at the risks. But the risks aren't zero. Right. So it is concerning. And of course, you know, the idea that your chat logs can be handed over. This is, you know, a generation ago, you know, your Google searches could be handed over. And we've all kind of gotten somewhat comfortable with the notion that's true. And now the idea that our AI chat logs can be handed over, I think is a little more personal than just a quick three or four string, you know, three-word string Google searches. Exactly. You know, food in downtown Chicago, you know, I love food in downtown Chicago. So, hey, you know, everybody can see that. I don't care. But, you know, another wrinkle to this is OpenAI's Terms of Service prohibits the use of the app by anybody under the age of 13 and then anybody 13 to 18 need parental permission. And so, you know, you can tell me how many chat logs are on there of 18-year-olds and up only. And it's a wrinkle in this because that data will be moved out of there. And you presumably, and I mean, we'd be foolish not to think that there aren't minors data in this as well. Yeah. And that's, I think that's another area where we're just, you know, again, early on, and I think lawmakers are doing what they can. And courts are doing what they can. But, you know, the regulations that exist basically says if you just put on their decision for you, then that's it, right? So, you know, for my banking clients, the website said this website is not intended for you to use by minors or, you know, anyone under the age of 16 or 13, whatever the age in that state might be. That works, right? Because what are the, what are the odds that some 13-year-olds are actually going to use it? You know, there's someone. But when you start looking at something like this, this is like, these are almost geared towards 13-year-olds, right? They're probably better at using AI than many of us are. And so, you know, well, what good is that terms of use paragraph that you and I probably don't read either? I mean, yeah, I'm a lawyer and I don't read those. I want to use the service. Oh, yeah, I'll do it later when the litigation comes. Exactly. Like, I want to use, I want to use the app. I can't be reading this, right? So, yeah. And then, you know, like you said, open AI is now going to lose control over that chat. So, whatever the control they had over it, the logs of a 13-year-old are now going to be out and shared under the protection of the court and all that, of course. But it's just one additional person that has a copy of it. Like you said, it's a non-zero risk. Like, even just transferring it, there's a possible risk, right? So, yeah, it's not, I don't think we have the answers for this yet, but it's definitely something we should be talking about. Okay, one last thing related is, I find it funny because we both picked this one on our list of, like, hey, which news articles you want to talk about this week. Tilly Norwood doesn't exist, but is the next AI Hollywood star, and it's causing quite a stir in the movie industry. I can't remember the company's name that produces Tilly, but it's completely AI-generated movie star that can be plugged right into scripts and plugged right into sets. Yeah, and mind blowing, like, you know, mind blowing in one sense, but you know, at first I thought, well, you know, it's just like some publicity stunt by a tech company, but if you go look at the, like, release, press release, and they have pictures of this character, like, on the red carpet, it looks like real, it looks like she's attending the Oscars, right? She went on Graham Norton, like, this fake Graham Norton episode, and she's sitting on the couch on Graham Norton. It's just, it's wild, right? And of course, the company, I can't remember either, but they don't, they're not doing anyone favors by just kind of coming out guns blazing. I think there's some quote now from one of the executives saying, oh, she's like a combination of, like, Natalie Portman, Scarlett Johansson, right? All these big name actresses, like, well, we combined the best of those people and made them into one person, which only creates more panic, I think, and more fear. You know, maybe rightfully so. So this isn't an IP issue directly where there's content being generated, you know? We hear that a lot out of the movie industry right now, but the concern of content being created from existing stuff. But this is, this is a whole other kind of can of worms, right? This is, yeah, I think they're still trying to, you know, the creative industry, the actors, actresses, and they're trying to still trying to figure out what this is. And I don't, I won't pretend to understand. Yeah. No, I saw the guild got involved right away. Yeah. They're like, well, let's talk about this. And it's, and it's tough. I think it's, you know, I, I don't know what you tell your students, but I tell my students, you will have jobs. This isn't going to, AI's not going to take away your job, but I think what's valuable about your job may be refined or narrowed or different than when I was younger. So I wonder if that's what this is going to do to actors, actresses. I am wondering about any of the precedent that might exist on this, because, you know, movies like Star Wars and Marvel movies that threw a lot of CG generated characters in or Gollum from Lord of the Rings, even though there was a human behind it, there was still a lot of computer generation. And that doesn't preclude any animations where something may have been computer generated completely. So, you know, there is kind of a world of entertainment that has been done in that space. And I just, it's now encroaching on Hollywood's blockbusters. Yeah. To your point earlier, it's easier, right? I'm not saying it's easy, but it's certainly easier than you don't have to own Lucas Films to create one of these. You just get some funding and, you know, a little bit later, you're worth a billion dollars. But yeah, I think, you know, there's some protection for, like, you know, name, image, likeness. And some states have their own versions of that, which are a little more protective. But, you know, if you look at the actual character, the Tilly Norwood character, she doesn't actually look like a specific one person. You know, they, I think they try to put together a bunch of different people into one. So, I don't know that our laws would respond to that. Yeah. You know, who does this remind you of? Well, it reminds me of 10 different people, right? And that's, that actually. My doppelganger somewhere in the world, you know, would remind me of myself if I found them, right? Exactly. So, it's, yeah, it's, it's, I think it's a difficult question. And, of course, now they're, I don't know if you saw this, they're, they're planning, like, 40 more AI characters. And who knows what those are going to look like and what they'll do. Yeah. No, this is, this is fascinating to me. I was like, yo, I'm a big Star Wars Marvel fan. I'm all about CG when it's done right. You know, I also had to live through the Star Wars era when it was questionable. But I would be interested in seeing it. But at the same time, you know, I have this kind of just nagging question of, you know, as we see more and more and more and more AI generated content, it just starts to flood everywhere. I'm wondering if people are going to crave the real more, you know, like the actors actually will have this like resurgence, like anti AI generated anything in films. Yeah, I mean, I, I think that's, I think that might be all right, too. You know, I started doing my holiday shopping and, you know, some of the biggest things are, of course, retro from, from our childhood, your tape set players are back. Or I think tape players are back, but you can also get an mp3 player that looks like a, like a Walkman, right? So it's actually like, you know, physical. And the idea is that we're all so tired of, you know, everything just looping around. So, you know, people kind of crave, you know, doesn't matter what generation you are, you kind of create things that are real. And so very much in book person. Yeah, exactly. I gotta have physical books. I could never get into Kindle. I could never. Yeah, I've got a printer. Same thing. I just returned another Kindle, you know, the scribe, they keep saying better. I print, I write on it, and then I take a picture, and then that's my scanning. So that's great. So, you know, for all the tech that we do, it's like, we still got to have some tangible paper here. Yeah. All right, Kent, thank you for coming on today and hanging out. It's been fun. We'll do this again and get some more of our colleagues on here and have some more chats about just what's happening in tech and AI and quantum, hopefully at some point and energy. We're thinking a little bit more broadly here that hopefully we can, you know, think science and technology a little more broadly and how it's impacting kind of society and the future policy discussions that should be happening and, you know, those big issues that we should be all talking about. So thanks for coming, Kent. I look forward to hanging out with you again next time. So thank you all for listening today. Transcripts to the program and links to the articles that we mentioned will be provided in the comment section below or on the IGPA web page for this episode. So check for that. And until next time, thanks for joining us. All right, thank you. Yeah, thanks.