Latest / Computer Says Maybe / Reporting on AI’s climate injustices w/ Karen Hao (replay)

Transcript

Reporting on AI’s climate injustices w/ Karen Hao (replay) · Computer Says Maybe

• Alix: [00:00:00] Hey there. This is the second in our deep conversation replays with Karen Hao. It's from November of 2024 and it was part of our net zero plus plus series. We add to it from time to time, you can look in the titles and the pod feed and you'll be able to jump into these. Series of discussions looking at the environmental implications of AI technologies. This one focuses on two news stories that Karen reported on at the time, and the main takeaway is that basically all tech journalism that is trying to report on data center infrastructure is basically by definition investigative because the industry is so intentionally opaque with its activity and the implications of what they're building. Hey there. Welcome to the computer, says Maybe podcast. My name's Alex Dunn and I'm your host. In this episode, we're continuing along this net zero plus plus series. In this conversation, we talk to Karen Howe, who I think is globally probably the best person writing about AI and the environment. There's two pieces that we dig [00:01:00] into of hers. The first is a story she did about Arizona. Which when you think of Arizona, you're not thinking about a place with lots and lots of water that they can give to data centers. But you might be thinking about lots and lots of land that they can willingly sell cheaply and maybe a struggling economy. So she did a story to try and better understand the process by which a data center was built and where development was approved by a local council. So she used the Freedom of Information Act to request data that had been given to that. Locality so you better understand one, what big tech companies are telling local municipalities about the energy use required to sustain these data centers. And two, to better understand what the sort of process by which these types of developments are approved. And she hit a wall, the FOIA request was approved. She got the documents and the forms and the, you know, consultation application stuff that the company had to. Submit as part of this process, but it was heavily [00:02:00] redacted. I found that personally incredibly compelling as a story itself, but I think she was really hoping to get more transparency about what energy uses and water use was actually required. And then we'll look to her second story focused on Microsoft, who is kind of touted as one of the biggest sustainability leaders in big tech. She digs into that PR framing and the. Historical truth to that, that actually Microsoft has invested a lot of resources and energy, which we heard a lot from Holly Alpine a few episodes ago. If you haven't listened to that episode, you should, but at the same time, Microsoft was selling its AI services to fossil fuel companies to make it easier for them to find more oil and gas for us to churn through as we careen towards the climate crisis. So she looks at that kind of hypocrisy of you can't be both. You can't be a sustainability leader and also be emboldening of fossil fuel industry. So with that, let's get into it with Karen Howe. Karen: My name is Karen Howe and I [00:03:00] report on ai. Alix: I'd love to zoom in on just why is it so difficult to do good reporting on the environmental impacts of tech? Because I feel like you are one of the only journalists doing it, which, and it's such a huge story and I don't know. I'm presuming the reason it's not happening. You have thoughts on, Karen: I do have many thoughts on this. Okay. So like, Alix: so like why is it, why is it so hard? Or why are people not doing it? Karen: One of the things that's particularly hard about the environmental dimension of. The tech industry and with AI in particular is that every single company that has a substantial footprint and is kind of like worth talking about or reporting on, has become extremely closed off in terms of the amount of transparency and information that they provide. So by default, any kind of story that looks into the environmental impact of digital technologies and of AI in particular is an investigative story. [00:04:00] And it requires like building lots of insider sources to give you information that has been marked confidential by these companies. Those types of stories just require enormous amounts of time and resources. It also requires trust building with sources, but at a baseline. It also requires you to have enough of an established platform that sources are willing to take a risk on you. That in and of itself is like already years of investment in building up your reporting chops, building up your source base, building up your reputation, and having the bandwidth to actually take on a story that could take months or. Even over a year to report. And frankly, like most journalists don't have that position in the media industry anymore, that those types of jobs don't really exist. There are designated investigative reporters and the investigative desks at many media companies has really shrunk over time because of the way that the media business is not doing well. [00:05:00] Um, but also what this beat requires is, is not just like. Investigative journalists. It requires like tech journalists to be investigative journalists, and it's just like that intersection has just become increasingly under-resourced and just really, really difficult. Alix: That's so interesting. I mean, I, I mean, it makes perfect sense, but I hadn't really thought about it in that way, that it's kind of a subset of the shrinking journalistic class has the. Ability even to report on this, choosing to position it as such, one of the coolest parts of your reporting in the last few years I thought was the clever use of FOIA in Arizona and just sort of showing how closed off it is. Like I think that that way of thinking about is really interesting. Like do you have a sense of. Are you imagining that mandated disclosures and transparency requirements are gonna kick in in the next few years that shifts the way that journalists can cover the story? Karen: I really hope so, because it's. [00:06:00] Becoming increasingly impossible. Part of the reason why in the story that I did on Microsoft, I showed how the FOIA request came back with basically nothing was because it came back with basically nothing. And I didn't really have, which is a story like I didn't have That's a story information. Yeah. Right. And I was like, maybe this is the story. But um, you know, like I was actually so frustrated reporting that story because. What I wanted was the actual information. I wanted, the actual environmental impact. Shining a light on how hard it is to get that information is certainly helpful, but it's not like I didn't successfully get to the bottom of what is actually the environmental impact I felt. I. Like I had not actually completed what I had set out to do with reporting this story in particular, I really do hope that there is some kind of legislation or regulation or something that makes at least disclosures around data centers more transparent [00:07:00] because. The way that data centers creep into communities that are ready to disenfranchised and already like economically disempowered. The power imbalance just makes it set up such that these tech companies can do whatever they want. They can enter a community wave around like a few million dollars as an incentive, and then set up shop, build a data center, start building power plants to feed that data center and before the community has even realized what has happened, it's too late now. They're stuck. Just hosting a data center in perpetuity. That's how this infrastructure works. It's like installed and then it works in theory for like decades, if not forever. There were so many communities that I encountered while I was reporting on not just this story, but other data center communities in the last few months who repeatedly just told me like we just didn't. Know what we were signing up for. I spoke with a city counselor in a different town in Arizona when reporting on this [00:08:00] story, and she literally said to me, she was like, no one told me that data centers use water. They were one of the earliest communities that a tech company approached these tech companies when they approached these communities, whether because of intention or mal intention, they just don't accurately portray what the costs of the data centers will be like. They have all of these projections and estimates on how much energy it might use, how much water it might use, and it all goes out the window once the paperwork is signed because. Their projections and projections changed based on customer demand, based on the weather, based on a whole host of factors that can't actually be predicted. And so they always give the rosiest picture to these communities. And the communities also have a complete and total reliance on the expertise of the companies because it doesn't occur to them in the beginning to seek independent expertise and there isn't that much independent expertise. So they listen to the tech companies say. We're going to be super [00:09:00] sustainable. We have like the most cutting edge technology for making sure everything is carbon neutral and water positive and whatever. And they don't actually look at the fine print of what that means. And then like the thing gets installed in and no one else is watching. Like no journalists are watching, no like civil society's watching because these companies can enter in through shell companies that are named after animals and things like that. Like I talked to this one community in. New Mexico, they discovered that Facebook had had signed a deal with their city council after the Shell Company called Avar or something. Suddenly like she its name and it was meta like, Alix: that's so wild. So they don't even know that this, no one knows. This is subcontracted piece of infrastructure from for big tech. Karen: Yeah. Like no one knows. In the Arizona case, what was so interesting within city governments, people don't see eye to eye within the government itself. So there's like the economic development team that's really focused on how do we get like big [00:10:00] numbers of investment into our town that we can like tout in press releases In Arizona, every city has like a water administrator that makes sure that the city won't run out of water. Alix: Yeah, and it's for like a set period of time. Right. They have to basically project out. Karen: Yeah. Alix: At least a decade. Yeah. Karen: Yeah. The water administrator needs to prove that they will have enough water for the next 100 years for the current population and the projected growth of the population over 15 years. It's like a convoluted concept. Whoa. But yeah, so like it's based on like the current population as well as the population in 15 years. And can we supply a hundred years of water to. That finite group of people. And so you approve each project based on your water budget. You decide how much water is coming in from like the river, how much are you gonna spend on it, how much of it it's going to go to housing versus to businesses. It has in theory, worked well for Arizona [00:11:00] for several decades. The system, but also the river that they're sourcing from is drying up. Great. But like within city governments, the different actors don't see eye to eye. Like the economic development People will be like, we have enough water, and the water administrators will be like, are you sure? Because the river's drying up. They both are relying on the company to tell them accurate information to put into their projections about whether or not they should approve the project. And obviously these companies are highly incentivized. To paint the Rosie's picture so that, that it does fall within whatever the water budget or the energy budget or whatever it is, are like the metrics that they need to kind of check off in order to get approval by the city. Yeah. All of it goes out the window after the first brick is laid. Alix: Are these disagreements aired in public? Karen: No. So often they aren't aired in public, which is interesting because I mean, they're, they're like in like city council closed door meetings. I spoke with like [00:12:00] a bunch of. City council members and city representatives and former city representatives in different towns in Arizona that were at sort of like different life cycles of economic development. So one of the things that. What's also interesting about the Arizona case is because water is so finite in the desert, the trade off between like how much water to use and like how much economic development you get shifts over time depending on like how robust your economy is as a city. So like underdeveloped economies. Are the ones that are much more willing to put down lots of water to get more economic development than someplace like Phoenix, which is like fully matured as a city now, and is much more stringent with like its water expenditures. And it's kind of a metaphor for the cost benefit analysis that every small town is facing that ends up needing to. Make the decision about whether or not to welcome in a data center, like every economically underdeveloped town is going to be much more willing to open their [00:13:00] doors to an Amazon, a Microsoft, a meta, a Google. Because the initial like payout that the company gives in terms of the amount of money that they're paying for the infrastructure, the taxes, it just looks so attractive that it's really hard to resist. Alix: That's so interesting 'cause as you were describing almost a predatory approach of big tech companies finding these small towns, you framed it as the sort of finding marginalized communities. And I, I think when I read your piece on Arizona and sort of looking at foia. My first feeling was one of complicity. I found it really shocking that local policymakers were willing to participate in redacting key civic information about a development project. But when you put it in that way, in terms of the outgunned, local policymaker, who's just trying to do right. By their community and generate income and development dollars. It looks different. Karen: Yeah, I mean it's so fascinating because like I was just speaking with someone else the other day, and I'm totally blanking now on exactly the [00:14:00] context of the conversation, but they were saying that in one community that they were looking into also in the US. When the data center came in, they like quickly realized after signing it that even though no brick had been laid yet, that it would like massively increase the energy expenditure of the town, and they would have to build new power plants and they would have all these ripple effects that they didn't actually want. But instead of undoing the deal, they decided to keep it because they felt that if they undid a deal that had already been signed. It would be such a negative signal to any other economic investment that could come to the city that it wasn't worth it. And so, yeah, it, it really is that like these towns are just really disempowered economically and I don't think they necessarily think they are that way in the beginning when they enter these partnerships, they probably don't feel the power gap in the initial conversations. It's also like hard to really say like, are these tech companies. Doing it on [00:15:00] purpose, are they, are they purposely going to the towns in this kind of predatory way? Partly, yes. Partly no. Those are also the towns that have the land. So it's kind of like the intersection of all of these incentives that misaligned in like a very particular way that leads to this super messed up situation. Alix: Yeah, I mean, where you're building data centers in Arizona, it's wild. I mean, going where the resources are. The least available and also most zero sum relations with people who also need those resources. And building these power hungry, water hungry data centers in those places, like it must be the cost of land. It must be the likelihood of these local councils not anticipating what's coming. Is it primarily cost of land and like physical locality in relation to a place like California where they obviously don't wanna buy land in California so they can buy it and a nearby state Karen: It is. And also the availability of power. So Arizona actually has like an enormous amount of power that it can deliver. To a single [00:16:00] point. Alix: Is it solar? Karen: Not just solar. They also have nuclear. They also have like a little bit of hydro, kind of like a mix of a lot of different energy sources. There's that aspect of it. And then there's also the underground infrastructure, like the cables that connect all of the data centers. Once you build like a data highway. You start to get clusters of data centers around that highway. Mm-hmm. Because it's just really expensive to lay another highway. So you try to like milk as much out of the existing cables as possible. And Arizona happens to have a massive data highway, a massive cluster of cables running underneath it. The same is true for. Santiago Chile, which is another community that I like Dove deep into, and the data center extension that's happening there, it was Google's very first data center in Latin America. They decided to plop in Santiago, and then shortly thereafter, Google backed a huge underwater cable that [00:17:00] cuts from Australia to the Americas through Santiago. So now all of the data centers are like rushing into Santiago if they want to enter Latin America because that is where they can like piggyback off of that infrastructure. Alix: Whoa. It's kind of reminding me of who framed Roger Rabbit. I dunno if you've seen that, where the, where he, he destroys, uh, all of the trolley car system in San Francisco and then at the end, the villainy is the highway system and his vision for the highway system, uh, and how imperial it is. Um, that's so interesting. I, I hadn't, I hadn't connected those dots of underwater sea cables and data center localities, if that makes. That makes loads of sense. And then you kind of zoom out and you see this huge layer of infrastructure that big tech is building globally, and then there's no visibility into it because of the kind of micro. Opacity of these deals. And it also makes me feel like if local municipalities feel [00:18:00] pressure to say yes, they're not gonna be pushing the envelope in terms of what data disclosures are required. Does it have to be global in nature to, to sort of require environmental reporting on what these. Systems are requiring of the physical infrastructure around them or like how do you imagine a transparency standard working so that investigative journalism just becomes normal journalism on this, this topic? Karen: Yeah. It has to be everything. Like it has to be local, state. Country and international standards because international standards, like who's gonna enforce them? So we can't just have international standards. But what was interesting reporting on Chile and Uruguay versus the US is actually like in Chile and Uruguay. There's much better transparency at the local level, and so there were actually extremely effective civil society protests against data centers, consuming water in the midst of a massive drought for both countries. From that perspective, I think having [00:19:00] transparency at the local level is actually like really key because it can mobilize people on the ground who actually understand their community and can protect their community in those two countries cases like. The main disclosure was just how much water would the data center use, and I believe in one of them there was also how much energy the data center would use. I'm not as positive on that one, just that one single number being like readily available and transparent as part of the each country's environmental. Agency requiring businesses to report on their environmental footprints. Just that alone was enough to spark an important democratic discussion about like, do we actually want these data centers? If they use that much water, can we modify the data center infrastructure to use less water? And that's really important. And also just allowing companies from entering communities in with shell organizations, like I think that is just a level of opacity that is so easy to get rid of. Like I don't really [00:20:00] see any reason why Alix: it should be allowed. Yeah. Karen: Yeah. Alix: I mean beneficial ownership data in the US I think mandate. That there'd be a disclosure of who owns Shell companies. I don't know if it's enforced, but as of like two years ago, I think so. I wonder. Yeah. But yeah, I mean, I, I can only think it, it shields liability and it's just a natural order of, of the kind of way these companies operate just by virtue of like anytime they set up a new project, they spend up an LLC, but that negative effects it has when they don't disclose the ultimate owner of that LLC is obvious when you're talking about these micro level negotiations. And also the cheekiness of the names is also like, feels, feels quite glib for what's happening. So the Microsoft story you did recently with Holly Alpine and others and sort of thinking about this sort of hydra or like, I guess Janice face of like we're a sustainable organization and also we're selling our services to extremely emission heavy organizations and not marrying up those two views of [00:21:00] ourselves. It feels like that's intentional. Do you feel like these companies are too big to sort of have a unified view on their environmental impacts and that this actually isn't, it's just like kind of a large organization being different in different contexts, or do you feel like there's this bigger play at like for as long as possible obfuscating, the fact that big tech is not green tech and green growth? Karen: I think part of it is that they are so big that they. Have a very contradictory motives in different parts of the company. But honestly, I do also think that that's a bit too easy of an explanation. It lets Microsoft off a little bit too easily in the sense that Satya Nadella, like the person at the very top of Microsoft, he was the one that signed off on. Tens of billions of dollars of investment into generative ai. And when you invest that much into a technology, you better have a game [00:22:00] plan for recouping those costs as like a publicly traded company. And there are really not that many industries in the world that can pay that kind of money for a technology that's like. Unproven like, yeah, you're like trying to find, that's a nice word. You're trying to find customers that are willing to lay down billions of dollars and like tons of customers that are willing to lay down billions of dollars each in experimental technology. And so there's no way that. The Microsoft executives made that decision without thinking through the fact that they would probably end up recouping at least some of their costs from the oil and gas industry because. This is the one of the few industries that can pay out that kind of money. There's also like, I mean, I don't know, like healthcare, retail, airline, and that's about it. There's really not that many. I hadn't Alix: thought about it like that, but yeah, the market is small of large organizations willing to. Pay up for [00:23:00] something. Yeah. That is that premature. Yeah. Interesting. Yeah, Karen: and, and when you look at the way that like cloud companies are organized, they are organized around the verticals that are able to pony up that kind of money. So like Google Cloud until recently, I don't know if it's true today, but like they were organized around Google Cloud, retail, Google Cloud, pharmaceuticals like Google Cloud, whatever, Microsoft, they already had an oil and gas, like a healthy oil and gas business before this investment. Because they were already like providing cloud technologies to oil and gas companies. So like if you already have a vertical, you're for sure going to be pedaling your latest technology that you've invested so much money in to all of your verticals and to telling every single vertical salespeople to shove this technology down like their client's throats. Alix: Yeah. You're not gonna be picky. Karen: Exactly. Like I don't think that Microsoft Yucks were unaware that this was going to happen. Alix: I hadn't thought about that at all, but basically the, just like the, yeah, the number of [00:24:00] entities that would be capable also of sustaining the, the bubble. I'm curious, so we kind of zoomed in on data centers and thinking specifically about that sprawl of big tech. Are there other questions or reporting that you wish was happening in journalism on environmental impacts more broadly of big tech and ai? Karen: I think there's been some already some like incredible reporting coming out about the way that the energy demands for. Data centers and, and for both training and running AI models has led to the life extension of gas and coal plants. I feel like that story needs to be done repeatedly as like every single community, every single country that is facing this issue needs to have stories like this announcing to the world that this is. The trade off that we're making right now with developing an experimental technology and completely reversing the energy transition in the process, and a critical side effect to that. I would love to see more stories on [00:25:00] the way that the air pollution then starts to affect communities and the water pollution, the water exhaust that comes out of these data centers. It doesn't come out the same way that it came in, so there's probably gonna be. Interesting consequences from that as well. There's so many other localized impacts to the environmental story that I think are really important to highlight because the global story of like climate change is here and we are literally accelerating it in one fell swoop with this technology is a, a really important one to hit on. But of course like. The negative consequences are not gonna be evenly distributed. It's always going to bear more heavily on communities that have already been suffering from various intersectional harms, and teasing out that more I think can really help. The average reader connect more with the story. Alix: I hope you found that as interesting as I [00:26:00] did. One of my big takeaways is that there's so little data transparency about the environmental implications of these data centers and other. Physical infrastructure of big tech that any journalism about it requires investigative journalism. And investigative journalism is expensive. Uh, and there's not that many people that have the resources to do it. And I think this observation leads to the other part of our conversation, which my takeaway from our conversation, which is we just need more structured, consistent, transparent data. About what it requires to operate the physical infrastructure of ai. We can't continue to rely on scrappy journalists like Karen, who spend sometimes months, sometimes years, trying to better understand what's going on. This should be disclosed if as a society we're gonna have a clear-eyed conversation about the costs and benefits of ai and. Also to just be more mindful about where we put this infrastructure, because right now it feels like we're flying blind. That's not great, uh, for outcomes, especially given, uh, the current situation we're in with our energy demands and the energy transition. [00:27:00] We'll leave it there. Thanks for listening. And if you liked it, hit subscribe. Um, share with some friends and also go check out Karen's pieces because if you don't already follow her, she is, I think, at the forefront of the best reporting on this topic. Um, so please do go give her stuff a read. Thanks to Sarah Myles, our producer and Georgia Iacovou for helping structure this whole series, and we will see you next time.