Sharon Goldman (Ground Level AI): AI as It Meets the Real World

0:08Kevin Werbach: Hi, I'm Kevin Werbach, Professor of Legal Studies and Business Ethics at the Wharton School of the University of Pennsylvania. For decades, I've studied emerging technologies from broadband to blockchain. Today, AI is promising to transform our world. But AI needs accountability, mechanisms to ensure it's developed and deployed in responsible, safe, and trustworthy ways. On this podcast, I speak with the experts leading the charge for accountable AI.

0:39Kevin Werbach: Sharon Goldman has been a journalist on the AI beat since 2022 and focusing on other areas of tech for a decade before then. She recently left Fortune to start her own outlet on, as she describes it, how AI meets the real world. We talk about the recent incidents of AI and cybersecurity, data centers, AI and trust, and how to tell stories to people so that they understand what's really going on in the AI world today.

1:11Kevin Werbach: Sharon, thanks for joining me and welcome to The Road to Accountable AI. Oh, thank you so much, Kevin, for having me. Looking forward. You have covered the tech industry for a long time. But if I read this right, you started full time on the AI beat in 2022, just before the whole chat GPT explosion, adventure beat and then Fortune. Looking back, did you anticipate what the crazy ride the last several years has been? Absolutely not. It's funny to think about because in 2022, I had covered AI a little bit. I had been covering really more enterprise tech. It was really about everything from software to cloud to mobile.

2:03Digital transformation was one of the big buzzwords at the time. Big data had been on the scene for several years. And AI, you know, old school machine learning had been part of that. So when I started AdventureBeat and I started covering AI as a daily beat, I sort of figured, you know, I've been there, done that. I couldn't have been more wrong. I immediately, probably two weeks in, I discovered that I was like a babe in the woods. There was this whole storyline going on that I felt like I had been plopped into the middle of. I really didn't know anything, for example, about OpenAI and Anthropic had just started the year before.

2:45I really didn't have any insight into these sort of frontier AI firms and how quickly things were already accelerating at that time, even before ChatGPT came out. So it really was a huge surprise. In addition, I must say that when ChatGPT first came out and I received the email about it, I didn't really appreciate what it was. There had been a couple of other chatbots coming out at the time, and so I perceived this sort of in the same vein. I didn't really see it as a momentous occasion. And two weeks later, it was everywhere. And suddenly, you know, my mother knew what ChatGPT was, and that was a huge change.

3:28So I definitely did not see that coming. But once the roller coaster ride got going, I feel like I kind of got my got my legs under me.

3:38Kevin Werbach: Yeah. What was it like covering this technology at major business publications over the past several years? It's been a real roller coaster ride, I think, partially because people didn't immediately see, as I say, like how quickly it was going to move. And right from the beginning, there was a lot of challenges and controversy, similar to what's being covered today. But it was sort of the beginning of that. People saying that there were real dangers and different open letters about pausing AI and all of that. That was pretty early on. And just how quickly agents came into the game. I probably started covering AI agents back in 2023, these very early days of that.

4:29But, you know, there were just various milestones along the way. And for a business publication, you know, you're not just covering the latest research and technology. You're also covering how it's being adopted, what companies are coming up. So it became such a vast and broad beat that, you know, I'm certain everyone can appreciate that as a reader, seeing the headlines, you know, just more and more every day to the point that you're like, I would really rather watch sports than see another AI headline. But as a reporter, it's just completely overwhelming. And many times over the past few years, I've had to make decisions about, you know, what am I going to cover?

5:12What am I not? What's important? What's not important? How do I see different sides of an ongoing debate? And it's never boring. I can definitely say that. Yeah.

5:24Kevin Werbach: And I want to get on to what you're doing now, but I just want to follow up on that a little. How do you make those decisions? How do you think about what stories to tell? I think you have to certainly think about the audience that you're dealing with. So Adventure Beat, for example, this was a little bit more of a technical audience. At least that's how I perceived it. So I did feel a little bit freer to go into the weeds and not have to explain every technical term. At Fortune, that was quite different. You know, I was constantly being told by my editors, Now, you need to explain what inference is or, you know, just other jargony words that I was already taking for granted.

6:05You know, people are not going to understand what you're talking about. And I appreciated that because it didn't really occur to me that a business audience, you know, a broad business audience, really quite different from, you know, a little bit more technical audience. And so, you know, I had to make my decisions there. And, you know, at Fortune, I was really tasked much of the time to be covering the biggest players. You know, they really wanted to cover what's the latest at OpenAI, what's the latest in Anthropic, is Google in the lead, is Meta in the lead, who's leading, who's losing, what's the race?

6:40You know, and in a little bit of a way, I feel like I realized that there were other stories that I wanted to tell that didn't fit quite neatly into the Fortune space. sometimes they did like I did I was very interested in covering the AI data center boom last year you know even before it started becoming such a discussion point I felt like it was so fascinating and I was lucky enough that Fortune was very supportive of that and sent me to several data center sites in different states to cover that but you know that wouldn't necessarily have traditionally been a Fortune story. So I felt, you know, gratified that I could get in there.

7:25But, you know, VentureBeat had a very different angle. So, you know, partially it's what I'm personally interested in, but always thinking about the publication's audience as well.

7:36Kevin Werbach: And you recently left Fortune to go out on your own. What are the stories that you want to tell at Ground Level AI? Yeah. So Ground Level AI is a publication and platform I launched on Substack It's a newsletter and a podcast. I launched in mid-June, right after I left Fortune. And the idea is, in my mind, I think of it as like AI as it meets the real world. That can mean a lot of different things. But to me, it's a real focus in going beyond some of the headlines that are really just about open AI or Anthropic and the models themselves. You know, the latest capabilities, the latest gossip. You know, I'm like, there's so much more like on the ground type of things underneath the models that are so important, you know, like the infrastructure, both digital and physical, the security, cybersecurity and governance and safety and security, policy issues that, you know, are a wide swath of angles around that.

8:43And even the geopolitics, which can kind of cross into all of those lanes. And then, too, some of the consequences of AI around workplace adoption and communities affected and stuff like that. So that still is quite a wide, broad swath. But still, it was a way for me to at least wrap my mind around the bead a little bit more because it's changed. AI is now, any reporter covering anything in a way is covering the AI bead. So if I'm going to do that, I'd like to kind of like put a little bit of boundaries around it.

9:19Kevin Werbach: Are there common themes among those areas that you're interested in? Yeah, I mean, I do think that so much of the discourse is around trust. So I feel like the various issues of trust and security are issues that sort of cross all of those lines. If you're talking about enterprise adoption, for example, you're talking about things that other people might think are not very sexy. The governance, the risk, the compliance, all of those things to me are really interesting because that's how it's going to get adopted and that's how it's going to be used securely in your firm. I also really do enjoy covering AI and cybersecurity.

10:08And funnily enough, it's become headline news over the past few months. I never could have imagined that. I thought it was like this, you know, side thing that most people aren't very interested in. But, you know, I go out every year to the Black Hat Security Conference. And that's when OpenAI made its big briefing announcement about Hugging Face. And I posted immediately about that. And, you know, it kind of went viral. And that's just an example of how, you know, these parts of the AI story and evolution as it really affects people is something that everyone wants to hear about. It's not just a, you know, a small audience anymore.

10:54But yeah, so I feel like that's some of I do sort of circle a lot of themes. Like, I really do enjoy getting into some of the nitty gritty stuff around infrastructure. Like, how are these data centers built? Or like, just today I was posting about like, how do we understand even what makes up the whole technology stack? Like people think of just the models, but there's software and applications, there's energy, there's data centers, there's the racks themselves, there's the servers, there's the networking, there's so much. And on the policy side, I must say I'm very interested in following what's happening around this year's election and how AI issues are, you know, everywhere around the election.

11:47It's become incredibly political. Absolutely.

11:51Kevin Werbach: How do you translate those technical elements for a broader audience? As you talked about, cybersecurity, everyone has known it's important for a long time, but most people's eyes glaze over when you get into the details of the mechanics. And, you know, the hugging face incident was fascinating, but you need to have some understanding about the technology to get what happened. So, you know, how do you think about presenting that in a way that doesn't lose the accuracy? I think, you know, that's where I feel like for me, you know, there are a lot of sub stacks out there covering tech. There's a lot of people that do things in different ways.

12:31But I'm not, you know, I do write my own analysis or essay type pieces, but I'm also speaking to a lot of people and I'm interviewing experts like I'm not a cybersecurity expert. I think the most important thing in covering a story like that is asking the expert people what they think about it. So I think if you hear from people on that wall about whether or not what they really are concerned about and what they aren't and sort of in a basic way what they would compare it to or how they could explain it to you, that's a way of imparting. you know you don't need to get that deep in the weeds to understand that that was a serious incident but you also don't need to get that deep in the weeds to understand that you know people have different opinions about it and and there isn't one you know one opinion that you know that you know has everything in it you know you you you there's different ways to explain it it's a nuanced issue is what i should say i feel like that's what i'm covering at ground level AI is really nuanced issues.

13:41And that isn't always as viral or popular as, you know, headline making scary news about AI. I think that's very popular. It definitely gets a lot of clicks. And not that I'm not trying to get clicks too, but I try to offer a little bit more nuanced where I'm asking different people for their opinions and sort of putting it in perspective for people, which, Sometimes I'll be introducing concepts that if you're not in cybersecurity, you might not encounter on a daily basis, but you might find interesting to know that that's what these people are thinking about.

14:21Kevin Werbach: You mentioned the issue of trust. This is also something I've thought about a lot and written about in different contexts. What does that really mean? We see all these polls where people say, I don't trust AI, and you hear that reaction a lot. But what is exactly the trust gap that you see in this area? I was just talking to several people about that this morning. I'm going to be moderating a panel in a couple of weeks on this topic. And that's the thing. It means so many different things to different people, and you can come at it from a lot of ways. But I feel like, for one thing, there's trust in the technology itself.

15:02You know, do you feel that, you know, these this technology is trustworthy, that people understand it? And generally people know that the answer to that is in many cases, no, that that there's a lot of things we don't understand about LLMs. Even the researchers building them don't understand that it's a little bit of a black box, even to them, and that there isn't enough transparency there. So there's the trust in the technology. Of course, then you have the trust in the people building the technology. And there, too, there's so many challenges. You know, I recently wrote a piece where I spoke to my nephew, who's a 22-year-old Gen Z guy, and he uses AI.

15:50So he's not against it. What he doesn't trust are the people building it. He really hates the companies themselves, the billionaire CEOs. So you have a lot of that, you know, the trust in the companies building it. And then follow from that, you know, the trust in the companies using it. So if you're at work, you know, do you trust that your company is using it in a good way? Are you worried about your job? Are you worried about the security of your data, you know, personally too, in your personal use? So the outcomes, you know, the output of the models is also a trust concern. And I find all of that really interesting.

16:33I find all those areas of trust really interesting and equally challenging. I'm not sure at the moment how, you know, all of that can be solved, but I'm always heartened by the number of people in all three of those areas that I do think are working on it. You know, if they're not successful yet, there are a lot of efforts in all of those areas to boost trust. But, you know, in this world environment we live in, I feel like it's difficult.

17:07Kevin Werbach: Absolutely. No. And I speak to a lot of enterprises and executive education groups about AI governance. And I always make the point that trust is important and you should want to be trusted and be trustworthy. But I'm curious, you know, what your experience is like, you know, talking to companies, talking to people, how much that actually matters. It's easy to say, oh, yes, sure, sure. We don't want something to blow up. We don't want to get sued. But I feel like there's often a skepticism that this is really going to matter and it's worth making the investments. Yeah, that's one thing I worry about.

17:46I worry that there's a waiting for the one big bad thing to happen before everyone jumps in with the bigger investments in trust or security or safety. because there's still so much FOMO around this. I'm sure you hear that from executives as well. And right, FOMO does seem to override some of this sometimes. But I do think that the OpenAI hugging face incident, as well as like Mythos and Fable and sort of the pace of change that we're seeing now does have executives more on edge. about needing to, and also maybe seeing the hate coming back at the, if their employees or their customers hate AI, that needs to be a wake up call to how they deal with it.

18:48So I'm interested to see how that plays out.

18:53Kevin Werbach: And you mentioned the reporting you've done on data centers. Obviously, this has become a huge political issue, a huge area of concern. How much do you think this is really a story about the energy usage, the water usage, the things on the ground versus really a proxy for all these vague anxieties that we've talked about? Yeah, it's really interesting because when I first started covering this, this was about a year ago. and at first I really did see it as kind of a local issue for the communities involved I definitely said to myself wow I wonder why we're not hearing more about this like this is you know you're taking these what the equivalent of big cities being built in these rural areas it's just astounding like you can't even comprehend the scale and these communities were mostly taken aback, you know, by what has transpired in both good and bad ways, to be fair, but in a lot of really challenging ways.

20:01But then as the months passed by, I definitely started seeing just so much more pushback, a lot of misinformation, a lot of, you know, just swirling and just growing this massive pushback that's turned very political, I do think that it is a proxy for a lot of things that people just hate or feel uncertain about around AI. It's sort of like a physical, yucky manifestation of everything you hate. You just couldn't... Cinematography better for a movie than having some spewing gas generator Elon Musk data center in Memphis, you know, to serve as an example of the excess of AI ambitions for billionaires.

20:53That said, you know, even though there is misinformation around some of the impacts, because part of it is that some of the impacts are as yet unknowable, most of these mega data center sites have not become fully operational yet. So a lot of what you're talking about is people speaking about smaller data centers or ones that have been retrofitted in some way. But, you know, but on the other hand, there are. I mean, some of the impacts are clear. Like you do have, you know, incredible amounts of construction going on. That's total chaos in Richland Parish, Louisiana. You know, yes, there are some winners, but you have people being evicted.

21:40There's definitely rising rents. There's not enough housing. You know, I visited a town near the site, an area near a block near the site where, you know, there were eight houses on the block and now there are 2000 people living in RVs. I mean, it's crazy. So I feel like there's a lot of legitimate, you know, impacts there. But on the other hand, I do think that a lot of people who are not the communities particularly impacted, they don't live there, but they're just against data centers, are definitely bringing in a lot of their own other concerns.

22:20Kevin Werbach: Do you think that there is any likelihood that we'll see an actual pause or moratorium on development? Well, there is already. In several states, They've already passed data center moratoriums at the state level or it could be at the county level. So that's definitely happening. I have to say that I don't really think so because there are communities that will accept them. And it is good news for some communities. And also, like, I do feel that many of these are just going to get billed by hook or by crook. So I would like to see more, you know, clear benefits for the communities involved and that they should be more.

23:12I just feel like there's a better way to do this whole thing. But, you know, I'm not a developer, but I feel like it could be more of a win win. And I do think many of these will still be built because there's a lot of political wherewithal to make that happen.

23:29Kevin Werbach: Oh, and so much investment capital. And all that investment loss. And if these AI labs are right in terms of the scaling laws continuing to hold, they don't have a choice. Exactly. Let's go back to the cybersecurity issue. How worried are you about what's coming next in terms of these threats? I am because when I speak to cybersecurity people that I respect, they are. That doesn't necessarily mean what some people think it means. It doesn't mean that these people think that AI is going to kill us all or that, you know, rogue agents are going to take over the world. But the issue is the speed and scale of what's happening now and whether the defense can catch up.

24:13It's really like cybersecurity has always been this kind of offense-defense thing. You know, you're playing offense to keep the hackers away, but you're also defending your own systems. So it's always been like that. But, you know, you might have time to catch your breath. You might have, you know, in mind a certain period of time that you can take to patch vulnerabilities, for example, or pen test your systems. But now that's really upended because the speed to which hackers can do some very bait, like typical hacker things, not even like rogue agents, but just like crazy amounts of phishing.

25:02They can take phishing emails to the max, for example, or new things that can be done at scale. I'm particularly weirded out by deepfakes in the enterprise where hackers will make phone calls that sound just like your boss's voice or Zoom calls where it looks like your boss's face or something or a customer's face. but as far as the higher the the more even scarier stuff like what the reason they took mythos uh kept mythos to just a few companies to work with and took fable kind of offline for a while was because of these higher level cyber security capabilities that attackers could use to find vulnerabilities everywhere in your system before you could possibly patch them but those same systems can be used to find them and patch them too so if you have access to them you could catch up kind of so that's a lot of what the discourse is now is about let us have access to these more powerful systems so we can catch up but then you have the people saying but if we let more people use them the hackers can find them so it's kind of a a weird dichotomy but some people i speak to think that the attackers need time to catch up too.

26:28So there might be this little window we're in where we still do have a chance to catch up, but the worry is we may not have enough time. So I would say I am worried and I'm really relying on smart people to help us catch up. Mm hmm.

26:47Kevin Werbach: You did a piece recently talking about how AI safety and cybersecurity were coming together. And I wonder if you could talk a little bit more about that and what's what's the gap between those two communities? Yeah. So when I first started covering AI, this was one of the most confusing things to me, because if you think of AI safety, isn't cybersecurity about safety? But in the world of AI, if you're talking about people in Silicon Valley and just in the AI world, they think of AI safety as the safety within the models themselves as far as their capabilities and making sure that the models themselves are aligned with what you do or don't want them to do.

27:34And so any fixes there would be within the models themselves, whereas cybersecurity has always been more about keeping things in control and in bounds, even if something is doing something that you don't want it to, which happens all the time with hacking, that you can keep your system safe and not let anyone in. It's also just two different cultures. You have a real Silicon Valley philosophical belief system culture around AI safety that some people might find off-putting or just very academic and research-oriented, whereas cybersecurity has been around for decades. It's really hands on and like, you know, in the in the enterprise, for example.

28:26So I have found in my coverage and going to different conferences that these groups of people don't tend not to understand each other. You know, the AI safety people feel like the cybersecurity people are not taking the model risk seriously enough. And the cybersecurity people think the AI safety people are just kind of like in a cloud, like not really understanding that you need to like control and put boundaries around scary things and that you can. So it's really interesting to see these two groups coming together now, which they do have to, because AI security is a little bit of a different thing.

29:09You're not just talking about, you know, software. You're talking about these models and having AI agents that are reaching out to different applications and potentially working autonomously, which is quite different, again, at a speed and scale that we haven't seen before. So, you know, everyone I speak to that I respect that kind of falls in the middle of this says we need both folks on that wall, you know, to make this work.

29:39Kevin Werbach: Yeah, ironically, it may be that these scenarios that sounded like science fiction, rogue agents finding a way to communicate, breaking out, hacking, the fact that that's now something real and tangible that has happened in some way opens the door, I think, to the people who are dealing with on the ground issues all the time getting involved and saying, OK, let's let's practically figure out what to do about this as opposed to just theorizing. Right. And I think the reverse is true as well. Like if you're dealing in your enterprise or in a cybersecurity standpoint about with AI agents that can escape their bounds and do things they're not supposed to, you might be grateful if a model company made sure that the models were more, quote unquote, aligned with human values and all of that.

30:31So it does feel like two sides, you know, these things both have to happen, it sounds like to me.

30:38Kevin Werbach: And last question, what are the stories or even one story that you think is not being covered enough around AI? It's probably an unfair question because you're going to go and cover it at ground level. But what do you think people are not paying enough attention to?

30:59That is a really good question. i i guess i do feel that most people aren't quite understanding how quickly things are changing because i do think that in a lot of ways it seems like it isn't in our daily lives i'm not sure that most people are feeling tremendous change in their lives from these things and it can seem like nothing's happening. And not that you need to pay attention to everything and it's not meant to, you know, be off-putting, but there is, I feel, a real speed and scale change to what's going on. So I do feel like how that kind of like floats out to the rest of the world is something that I'm keeping an eye on and I think people need to pay more attention to.

31:57All right.

31:58Kevin Werbach: Well, you're doing a great job in terms of telling those stories, and it's great to have you out there as a voice. So, Sharon, many thanks. Appreciate the conversation. Thank you so much, Kevin. This has been The Road to Accountable AI. If you like what you're hearing, please give us a good review and check out my substack for more insights on AI accountability. Thank you for listening.

32:23Kevin Werbach: If you want to go deeper on AI governance, trust, and responsibility with me and other distinguished faculty of the world's top business school, sign up for the next cohort of Wharton's Strategies for Accountable AI Online Executive Education Program. Featuring live interaction with faculty, expert interviews, and custom-designed asynchronous content. Join fellow business leaders to learn valuable skills you can put to work in your organization. Visit execed.warden.upenn.edu slash ACAI for full details. I hope to see you there.