Hello and welcome to the latest episode of This Month in Responsible AI. It's part of All Tech Is Human's livestream series where we engage in conversations with responsible tech leaders, advocates, researchers, policymakers, practitioners, educators, journalists, and more. We are discussing some of the world's thorniest tech and society issues. We're so glad you're here with us for this important conversation. So, if you are joining us live, please be sure and say hi to each other in the chat. Share a little about yourself. Tell us where you're tuning in from. And if you have a question at any point, please drop that into the Q&A function. We're going to try to incorporate a few of those toward the end. So, I am your host today, Rebecca Tweed. I'm the executive director of All Tech Is Human, and we are a nonprofit focused on advancing responsible tech. We are addressing tech's societal impact through community building and collaboration, through education, and through career development. So today we are celebrating our newest initiative, which I'm especially thrilled about. So yesterday we launched a series of five responsible AI courses. They are free, they're short, they can be finished in just a few hours, and the purpose is to offer an introductory understanding of the foundations of responsible AI. So we're looking at its principles, history, the evolution of the field, plus an exploration of what the professional roles are, current industry best practices, and then of course the ever-evolving governance landscape. So you can check that out at alltechishuman.org. Please let us know what you think. Super thrilled today to be joined by Hayden Field, senior AI reporter for The Verge, where she leads coverage of the biggest names in AI, including OpenAI, Anthropic, Google, Meta, and others, while also reporting on how these tools are being used, their societal implications, and how regulators are responding. Hayden Field, welcome to This Month in Responsible AI.
Thank you so much. Yeah, it's great to be here.
So, okay, I feel like maybe we should be calling this episode This Month in OpenAI because there's just been this onslaught of releases and launches. It's been relentless. And I'm so excited to have you on today. I think you're the perfect person to walk us through this. And I want to talk about what this means for us, what kind of impacts these releases might be having on our lives. So I want to quickly give a little rundown of what all has happened in the past few weeks, all the launches. So, just very recently, we've had a ChatGPT-powered browser. We had Sora, the social network, and Sora 2, the multimodal video model that powers it. We've seen the release of the newly erotic ChatGPT. But don't worry, there's also new parental controls. We've had Pulse, so we can wake up with ChatGPT first thing in the morning. But we have now these app integrations with Canva, Zillow, Coursera, Spotify, and others. And we have infrastructure deals with Nvidia, AMD, Broadcom. And we have had more hints about this new hardware device that's in the works with Jony Ive. So, first question for you: how are you keeping up?
That is a great question. Honestly, I don't know. It's crazy. It's definitely a lot of, like, you know, thinking you're going to spend your day doing one thing and then just completely pivoting at a moment's notice. But, you know, I'm finding peace in the chaos and just embracing the unknown. There is always something going on. And I try to just give my time to what I can bring the most context to and what I can, like, analyze and, you know, give readers the best view on with my experience of, like, six years of covering AI. So if it's a small, like, incremental update or, you know, just a small piece of news or just a feature release, I may be more likely to skip it or give it to someone else. But if it's something that has a lot of societal implications or, you know, you kind of see the writing on the wall with the problems that could come from it or even just something that's going to change the industry a lot, then I try to definitely spend time on that.
Yeah. What do you think is going on with OpenAI releasing everything? You know, from, I don't know, all of these things that came through. I know DevDay happened recently. And I know you were there for that. It feels a little bit like a flood-the-zone, like there's so much happening all at once. Do you have any sense of, like, why there's so much happening now?
I think it's honestly because all of these companies are doing this at the same time. It's not just OpenAI. They're all, it feels like they're all kind of climbing ladders and advancing one rung, and then the other one advances one rung, and then the other one advances one rung, and they're all trying to just keep up with each other and stay in the news and stay at the top of mind of consumers. I think they're all trying to corner the market on different things right now and really just be part of consumers' everyday lives and enterprise and governments, obviously. But for consumers, they're thinking, okay, like, how can we get in their routines first thing in the morning, and how can we stay with them at work, and how can we be with them, you know, in their personal life after work? So they're all trying to kind of desperately, you know, grab user attention besides their work in enterprise and government, obviously. But yeah, none of these companies are really generating profit yet. So they're really trying to figure out how to do that, especially as compute becomes more expensive and, you know, harder to get. So it's definitely, I think, just an arms race where everyone's trying to do the same thing at the same time. And also, you know, a lot of these companies, including OpenAI, have had a lot of bad press lately. So, you know, whenever they can flood the zone with feature releases and updates and products, you know, it helps drown out some of that noise, probably they're thinking.
Yeah, that sounds exactly right to me. So I want to get into some of the societal impacts and, you know, what the potential implications are of some of these new developments across, like, what you mentioned. You know, we have products and features and new model capabilities. So the first one I want to talk about is Sora. I feel like that one maybe made the biggest splash. And you had, like, this social network, but also this new model that's powering it or a new version of the model. So yeah, can you share a little bit about, like, what that's about and what OpenAI is doing there?
Yes. So it's interesting, you know, testing Sora. I remember when they showed me a demo of it before it was out. It was the most terrifying thing I had seen in terms of, like, multimodal AI because it was the first time that I couldn't tell the difference between AI-generated video or images. Obviously, this is video, but it was the first time in general that I could not tell the difference between AI and reality ever. So, you know, there were a bunch of, obviously, if there's something, you know, that's fantastical, like, you know, Sam Altman drinking out of a huge juice box or something, then, you know, you're going to know it's not real. But there were a lot of videos they showed that were very run-of-the-mill, like two people talking in a room or someone opening a refrigerator, you know, someone pitching to an investor, things like that. And they all seemed completely real, and there was zero sign that it was AI-generated besides if you were to download it, a watermark. Even they were very also proud of, like, their physics research that had led to this, you know, that we weren't seeing any more six-fingered hands or if someone dove into a, you know, lake in a video, the ripples looked accurate. You know, they had done a lot of research on, you know, world-building and how to make these, like, consequences of an action in a video seem real. But, you know, the thing I was wondering about first was, of course, you know, there were a lot of people depicted in these videos, and you can give your own permission on whether people can depict you in cameos or who can do that. But still, you know, there's always ways to get around things like that, like describing someone to generate that looks like someone else. Maybe even if you're not specifically saying, oh, generate, you know, a cameo of this person, you can describe the person enough that they will generate someone like that. The other thing that bothered me was the fact that immediately, you know, we saw a ton of watermark removal tools proliferating everywhere, and it's very easy to remove these watermarks. Honestly, a lot of people are coding their own tools for doing that. A lot of existing tools offer that. So yeah, it's just, it was just kind of the first time I saw it, I was like, you know what, this is going to be a misinformation nightmare more than we already have. And I think it has, you know, started on the path towards that. You know, I've seen a bunch of, like, non-watermarked videos out there. You know, we're seeing, I'm sure our government may start using some of these, you know, if history is any guide. So yeah, it's definitely a, it's, some people think it's so fun. I've had a lot of friends love, you know, making videos of their pets, and, you know, a lot of OpenAI employees are having fun, like, making inside-joke videos with their friends and stuff. But, you know, besides the copyright stuff and, you know, the potential for misinformation, there's also a lot of other potential pitfalls here. So it was definitely a big moment when I saw it for the first time.
Yeah, it sounds like it's, like, that ordinariness of it makes it even more concerning. It does feel like, you know, I think this one had a lot of buzz. A lot of people had strong reactions to it. Some people I heard were saying, like, "Oh, this is just a flash in the pan. It's going to get boring for people." But then other people, I think, and what I tend to think is that it will have staying power because of those cameos, and it has this real, like, personalization and that ordinariness and the potential for memes. It's like they worked really hard to make this fun, at least for a little while. So, yeah, I wonder if they're doing it because of that stickiness. And of course, like you mentioned, they need to bring in revenue in whatever way possible. They're trying to pull in more data for the next generation of models. But yeah, what's most interesting to me about it is, you know, this normalization and this attempted, like, memeification of deepfakes almost as a genre, like as desirable content, when of course the responsible AI community, we have long perceived this as very risky, leading to many potential harms, undermining the information ecosystem itself, having implications for, like, what a shared basis for truth is, and ultimately democracy itself. So, yeah, I think that one's definitely something to keep an eye on, and curious if you particularly think it is going to have staying power or if you think this is something that might, like, get boring quickly for people.
I think it's hard because I feel like, you know, I've made predictions like this before and I've been very wrong. But I will say, I think that the technology will have staying power and that, you know, this is a huge game-changing moment for the AI industry. Misinformation, deepfakes, for sure. I don't know if the app, like as a social media app, will have staying power. And I think that's because I've just seen so many, like, really buzzy social media apps come and go, especially during the pandemic, you know. We've seen a bunch try to compete and then just fall flat, or they're a novelty for a little while and they fall flat. Like BeReal is a great example. You know, there's, like, David Dobrik's photo app. There's been a bunch like that. But I will say, you know, I think this will probably keep, like, a loyal set of users that really love it. But something that Sam Altman said during DevDay, like in a Q&A with the reporters, which I thought was really interesting, was that he did not anticipate that a lot of people would just be using this kind of privately for videos to download and send to a group chat of, like, two or three people. He thought everyone was going to be either, like, just lurking it or posting videos for everyone to see. He didn't seem to anticipate that it would be, like, you know, a lot of people weren't really comfortable having their likeness out there for everyone, but they did want to generate videos and just send it to a couple of trusted people off the platform. And so I think that's what we might see this end up becoming. And I think they might be building features to that end as well, because, you know, it's just something they didn't anticipate about how people were really going to use it in real life. And that seems to be kind of like a leading way that people are doing that.
Yeah, that's really interesting. So yeah, it might become, like, a fixture of the group chat. But like, I mean, I can't imagine, you know, there might be a gendered aspect to this. I don't know. But like, I would never allow my likeness out there.
I really wouldn't either. It's like, I don't know. I already get weird about, you know, which social media platform I'm really professional on versus my personal life. So it's like, you know, maybe it's just because we have, you know, seen it all these days. But yeah, I definitely would not be like, everyone generate images with my likeness anytime, anywhere. Exactly.
Okay. So, yeah, kind of like following that thread a little bit, I'd like to talk a little bit next about, you know, ChatGPT, like, veering into the lane of, like, erotic ChatGPT on the heels of adding parental controls. And just, yeah, can you talk us through, like, what's happening there?
Totally. So, yeah, it was really interesting. I remember a couple of months ago, Sam Altman was on an interview with a YouTuber and he said that he was really proud that OpenAI hadn't juiced numbers for short-term gain with something like a sexbot avatar, and he was appearing to maybe take a dig at Elon Musk at the time. And now it's been a few months and things have changed. You know, he's gone full-fledged into the treat-adults-like-adults view. And I do know that that's partly because, you know, people have been, some people have been asking OpenAI for this. You know, it's been, there's been a small but vocal group that have asked for them to do this. So I'm sure it's been on their mind, but it is interesting that they would, timing-wise, announce something like this right after the parental controls. I'm not sure why the timing was what it is, but, you know, yeah, it was interesting to see the, you know, reaction on social media. Some people were, like, memeifying it to no end, mocking the company for, like, pivoting from its AGI mission seemingly to erotica. And Altman actually had to respond to those, like, memes and jokes. A few days later, he said, "Oh, I didn't anticipate the erotica news blowing up as much as it did. It was just one example of what we're doing to, like, give people more freedom." So, yeah, it was interesting for sure. But, you know, I think it might also be tied to just the fact that the company's concerned about profit and compute to fund its larger mission. It may think that this is something it needs to do along the way to reach its eventual goals. You know, and when I was in a Q&A with executives at DevDay, they could not stop talking about how much they would eventually need to turn profit and they need an ever-increasing amount of compute to, like, reach their goals. So, yeah, it's going to be an interesting rollout to watch. I also thought it was interesting that Altman wasn't super specific about how the company was going to aim to protect users in mental health crises. Like, you know, what happens when, you know, the persona that you've been talking to has its memory reset or its personality changes with the latest update and the connection is broken? You know, we've seen a bunch of, like, subreddits and other conversations online of people being really worried when, you know, a companion of sorts changes and they're desperately trying to get it back. So, yeah, it's been an interesting thing to watch. I get a bunch of, you know, outreach from people going through, you know, problems with emotional connection to AI systems, whether they know it or not. You may get those too. A lot of journalists I've spoken with also get them. So, yeah, I do think it'll be interesting to see if this makes things worse in terms of, like, you know, emotional reliance on AI. People that are already in vulnerable positions, you know, falling into more vulnerable positions because now there's, like, a sexual aspect involved. You know, even, like, a 14-year-old boy last February died by suicide after romantically engaging with a Character.AI chatbot. And so, yeah, I don't know. It's interesting to see how people might use this. And then the last thing I'll say is, there have been, something I saw recently in a report was that there was an account of, like, jailbroken chatbots being used by pedophiles to roleplay sexually assaulting minors, and there were 100,000 such chatbots found online. And that's not an OpenAI thing, but I do think that anytime you introduce, like, erotica into these systems, there's just a lot of potential for misuse. So, yeah, we'll definitely have to be watching that.
Yeah. No, that's really insightful. I feel like there's, when you're kind of thinking through the relationships that people have with these tools, there's this real, like, trust bifurcation where I feel like there are people who are, like, deeply trusting these tools, and then, you know, there's, in general, like, if you zoom out a little bit, there's definitely, like, a larger trust issue that people have with AI. They don't trust it in general, and there are a lot of, like, negative connotations about it. And so I think this, this is kind of what struck me about this was, was, like, pairing it with the parental controls is that, like, potentially you have all these parents who are maybe likely not to be trusting, like, OpenAI to be having, you know, to really be, like, putting all the safeguards in place necessary for their kids who use ChatGPT. And I can see people maybe leaning more toward, or parents leaning more toward, you know, Claude, for instance, or Gemini instead of, like, taking, like, rolling the dice on ChatGPT or Meta AI or, God forbid, Grok. So, okay, next I want to talk a little about Pulse. So this one did not make quite as big of a splash. So I don't know if everyone has maybe heard of this one. So can you talk us through, like, what Pulse is and what OpenAI might be trying to do with this?
Sure. So, yeah, Pulse is a really interesting feature, and I think it did fly a little bit under the radar, but it was one of the features that I actually thought was the biggest deal in the past year or so, maybe just as an average consumer, you know. For me, it was interesting because it's a use of agentic AI that is actually probably pretty useful for the average person in a way that some of the other updates have not been. You know, maybe those are sometimes updates, like, you know, marketed towards power users or people that are, you know, really into AI, whereas this I thought could, like, reach a wider swath, for better or worse. So it's basically a way for ChatGPT to research things for you overnight. So, you know, you say every morning you start your day with Chat. That's what they're aiming for here. So, you know, maybe you say, "Hey, every morning I want to have, like, a vocab lesson for this language I'm learning." Or, "Hey, you know, every day can you send me a new Pilates workout that targets, like, a different area of the body?" Sometimes, if you connect it to your calendar, it can say, "Hey, like, I see you have this run and then you have a dinner later. You know, that's going to be a little bit tight. Here's the best way to walk to get to the dinner, and here's what you might want to order since I know you're gluten-free." So it's a little bit scary because it does have a ton of personal information about you if you allow it to. Again, it's if you allow it, but I do, I can see a lot of people relying on this because if you start your day on some, if you start your day with something and it's kind of assistant-like in this way, I mean, it is pretty helpful. But, you know, there's also the tendency that it's going to learn a lot about you from your calendar, from your preferences, again, if you connect it, your email, your contacts. So, yeah, I was definitely interested to see how wide-ranging this could be and, like, you know, how, like, just many people would probably find it useful.
