transcript

Speaker 1:
[00:01] When the AI overlords take over, what are you most excited about?

Speaker 2:
[00:05] It's not crazy, it's just smart.

Speaker 1:
[00:08] And just this year, in the first six months, there have been something like a thousand laws.

Speaker 2:
[00:12] Who's actually building the scaffolding around how it's going to work, how everyday folks are going to use it?

Speaker 1:
[00:17] AI only works if society lets it work.

Speaker 2:
[00:20] There are so many questions have to be figured out and...

Speaker 1:
[00:23] Nobody came to my bonus class.

Speaker 2:
[00:25] Let's enforce the rules of the road. Welcome back to Scaling Laws, the podcast brought to you by Lawfare and the University of Texas School of Law that explores the intersection of AI, policy and of course the law. I'm Kevin Frazier, the AI Innovation and Law Fellow at Texas Law and a Senior Editor at Lawfare. Today, we're looking at how AI is being used as a tool for global emissions accountability. Our guest is Gavin McCormick of ClimateTrace. Gavin has spent his career at the forefront of climate tech, moving from environmental advocacy to leading a coalition of tech companies and NGOs that uses satellite imagery and advanced AI to track human-caused emissions with unprecedented precision. From the halls of environmental summits to the server rooms where these algorithms are built, Gavin's work allows for a verifiable and real-time inquiry into who is emitting where and how much. It's an especially timely development, given that a number of environmental concerns have arisen from America's AI buildout. Stay tuned as we suss out reality from myth and get to the heart of some of the open questions around the environmental ramifications of AI development. Giddy up for a great show and be sure to follow us on X and Blue Sky. Gavin, welcome to Scaling Laws.

Speaker 3:
[01:46] Thanks. It's good to be here.

Speaker 2:
[01:47] Gavin, when most folks think about AI, oftentimes you hear them raise environmental concerns. Hey, it's using a lot of water. Hey, it's using a lot of power. Oftentimes, when you talk to folks in environmental communities, you hear, I don't use AI or I only use AI when it's truly necessary. God forbid I ever say please or thank you to AI. That's something I'll never do because it's bad for the planet. I was so excited to hear you speak at the Ashby workshops hosted by Fathom because you cut through all of those paradigms while being someone who is very much staunchly in the environmental community. For the folks who haven't had the pleasure of meeting you and learning about your work, who the heck are you and what the heck are you doing both at watt time and then with ClimateTrace? We can break this down into a few subparts.

Speaker 3:
[02:39] Yeah, thanks. It's so interesting. I think the key message I have about AI before I get any further is sometimes it's helpful for the environment. Sometimes it's harmful for the environment. It matters. What are you using for? How are you doing it?

Speaker 2:
[02:52] It's not all there. It's as if AI is a tool that has good and bad use cases. Yeah.

Speaker 3:
[02:57] Whoa. I got my start in this space with a different tool that people thought was 100% good for the environment that turned out to be actually often not good for the environment. So sort of the flip version of the story. So a lot of people know this story that renewable energy is of course necessary for being climate change, but sun doesn't always shine, wind doesn't always blow, so you need some kind of batteries to store the energy. A lot of people assume that means that batteries are always good for the environment. I was a PhD student at UC Berkeley when some new research was coming out that said, actually, 90% of batteries are increasing pollution, not decreasing pollution. What? What's going on there? It turns out that what happened is people were deploying large numbers of batteries without asking, what time should we charge those batteries? It turns out batteries only help the environment if you charge it at the time that's good for the environment. If you charge it at the time that's bad for the environment, you just helped out coal plants instead. And so I was involved in some research saying, what time do you need to charge to be on the pro-environment team, so the anti-environment team?

Speaker 2:
[04:06] Yeah, to pause for one second for the folks who, like me, love to ask dumb questions. So this is basically getting to the point that our grid can only handle so much power. And so if you're charging that battery at the wrong time, you're really straining the grid and not doing the world a favor, and in fact leading to deleterious outcomes.

Speaker 3:
[04:28] Yeah, the way I like to think about it is there was a very concerted information campaign effort by electric utilities. It kind of like predates a lot of the AI campaigns to convince everybody that the best thing for the environment was to do whatever was most profitable for the electric utilities.

Speaker 2:
[04:45] Wow, who could have seen that coming?

Speaker 3:
[04:46] Who could have seen that coming? So the whole notion that it's somehow good for the environment to use energy at off-peak times and that you want to not use it when it's peak, that's true if you're talking money. It's not always true for environment. A different way to think about the problem is to say, if I use energy from the power grid at this time, what power plants are going to turn on or off? And so it turns out that research was coming out of that right about 2012. And long story short, I was the grad student assigned to go actually operationalize some of those algorithms. And so we just for fun on a weekend hit upon this idea that a few people from Google and Facebook and me met at a party and we're like, what if we just put this on the internet? What if we just made it possible for anybody to know what time do you have to use energy for it to be cleaner or dirtier? And we hit upon this idea that in addition to charging batteries at the right time, you could use your dishwasher at the time that is good for the environment. That would mean it's running on clean, renewable energy. You could not charge your electric vehicle at the time. That means you're going to have a bunch of coal plants. So, I got involved in this interesting intersection of software and science and kind of telling the truth about environment, where we stood up this nonprofit that had a budget of $0. It was just like, hey guys, everybody should be able to see the best science on what's really true, putting aside all the corporate claims about if you do this or that, how is it going to affect emissions?

Speaker 2:
[06:14] This is the most lovely Bay Area story I've heard in a while. A bunch of smart people walk into a house party and say, let's use the internet to expose a bunch of information and make people smarter. I love it.

Speaker 3:
[06:26] This is our concept of fun.

Speaker 2:
[06:28] This is, yeah. I will say, I don't know Gavin if I'm ever going to ask you if you're hosting a party, because I just think we may have different conceptions of fun. But with that in mind, before we get into some of your more recent work, what I want to establish early on is just this trend of people's first response to a new technology. Being the one that really anchors them. Because here, as you noted, it's like batteries are here, they save the world, that's all they do, period. Then electricity companies saying, hey, we have this new paradigm, this is the way the world works, period. Don't look fine. We were supposed to be done with this. We found out the sun wasn't the center of the universe. But what is this temptation even among folks in the environmental community, scientists, empirically driven? What is it that leads to that gut reaction, that anchoring effect?

Speaker 3:
[07:23] I think part of it is just like, hey, the world is a complicated place. It takes some time to figure out exactly under which conditions our technology is good and bad, and that one makes sense to me. The other thing I think is about incentives. Nobody ever got money to found a tech startup where they said, will this make the world a worse place? It's complicated. Further research is needed. You got to say it.

Speaker 2:
[07:45] Yeah, I think that pitch fails nine times out of time.

Speaker 3:
[07:48] Totally. And nobody ever raises money as a non-profit to say, is this thing bad for the environment? It's probably fine, but it's a bit complicated. Further research is needed. So you got a world where some people got to say that everything is good, and some people got to say that everything is bad. There's very few institutions that actually are able to sort of be neutral. Everybody thinks it's universities, but a lot of universities, they need to have the answer be, it's complicated for funding. And what we found at Watt Time is that there's very few people who can just be like, oh, this one's good and this one's bad, end of story, no further research needed. That's not a thing.

Speaker 2:
[08:22] Yeah, and it takes folks a little bit more time to get into that nuance. You have to sit down, you have to have an actual conversation, because it's not the sort of thing that gets covered by The New York Times, right? You're not going to read five paragraphs down into this, you know, huge story and suddenly realize, oh, I'm going to change my mind, I've seen the stars, everything has become clear now. And sometimes, to your point too, Gavin, you have to sit with that uncomfortable truth, because I know folks, and I'm sure you know folks, who they buy that battery, they buy that Tesla, and they think, that's it, I'm the world's greatest environmentalist, you know, mic drop, give me a ring, Al Gore, I'll take your prize now. And they don't want to acknowledge the fact that it's not always that clear cut.

Speaker 3:
[09:07] We get actually more hate mail from people who thought they were environmentalists, and we published a study saying that actually I was wrong than we do from fossil fuel companies. And I'm like, I'm sympathetic. It's not fun to learn that you draw, you were a good guy and you messed up.

Speaker 2:
[09:20] And so for folks who are thinking, oh my gosh, my whole life has been exposed right now. Who the heck is this Gavin guy? I need to go research him frequently. So tell us a little bit more about the kind of validity and rigor of the science you're doing in terms of having this measuring of how the grades are being used and thinking through how you are actually contesting the status quo in a way that is verifiable and reliable.

Speaker 3:
[09:49] So this all started with UC Berkeley research. My innovation was just like, hey, everybody should get to see this research. So this is really Meredith Fowley's research more than mine. Inesh Azevedo at Carnegie Mellon, now Stanford, kind of started doing the same research. So a lot of that started trending. I think this got serious when the state of California got involved. So there was a study that came out that said 90% of batteries in America are probably making emissions worse, not better. And all you would need to do is take those same batteries and charge them at a different time. And the state of California had recently established the nation's most generous subsidy for energy storage because it's supposed to be helping the environment. But the law said, yeah, but it has to actually help the environment or you don't get the subsidy. So, it was actually really interesting. The government had to get involved to answer, well, which batteries are good for the environment and which ones aren't. I was predicting years of bureaucratic infighting. I was all prepared for a very slow government story. It was not like that. What they did was really interesting. They convened a group of the 40 largest companies, NGOs, and universities thinking about this issue and said, if you guys can all reach consensus, if you guys can in a serious time frame figure out a serious answer for how do we know what's good for the environment, what's bad for the environment, we'll go with that. And lo and behold, after a whole lot of meetings, 40 members of the working group reached perfect consensus. I was blown away.

Speaker 2:
[11:08] Took a few months. Where is the biopic on this miracle that happened in California?

Speaker 3:
[11:14] Yeah, I mean, I got to say, I was pretty impressed with the California Public Utilities Commission. They did the job well. It was not a coincidence. They did a lot of listening. They did a lot of calling BS when people were making stuff up, but listening when consensus is trending. And so there's been so much research since then, but I think the key point is there is scientific consensus on how you figure out, use electricity at this time or that time, what happens to emissions. And now we live in a world where instead of that being kind of unknowable, that's a crazy new type of data that people never used to have. Think about you plug in a laptop, you flip a light switch. It's not like it's labeled with you cause this much emissions, but now we know.

Speaker 2:
[11:55] And what I love about this whole thing is that none of this would have happened, number one, without a means to measure, right? We have to make sure there's actual mechanisms in place for us to have that data, receive that data. But then, and here's where I'll scream from the mountaintops, sorry listeners, for me yet again, climbing up on my soapbox, having outcome-oriented legislation, right? By virtue of saying this law has to actually achieve its intended outcome, we had this conversation. But that miracle would not have been possible if it was just more Californians should buy batteries full stop.

Speaker 3:
[12:31] Yeah. And, you know, I think I was so impressed because they went in with a theory and they admitted that theory was wrong. They had a theory that more batteries is good. And then when they realized that they had been mistaken, they said the law does have an outcome. This must benefit the environment. What do we have to do to achieve the outcome? And it was remarkable to see.

Speaker 2:
[12:52] And so you proved this model in terms of just hanging out with a bunch of folks in San Francisco, saying we need radical transparency, gathering that data, making it available. You prove that with Watt-time. And in case folks haven't picked up, when we're saying Watt-time, it's W-A-T-T. Watt-time to use better. Watt-time, yes. And then you thought, huh, that worked out pretty well. We changed California policy. How did you create ClimateTrace? And what is ClimateTrace?

Speaker 3:
[13:25] So I think before I can get into the story of ClimateTrace, I need to share one surprising next step that happened with the California story. What was so surprising is that people have this paradigm in their head that when the government mandates you have to do something better for the environment, it's hard, it's expensive, it's painful, we're sacrificing something. The crazy discovery of this working group is that it wasn't any more expensive to do it right than to do it wrong. And so what started happening is that although only one state mandated this, we started hearing from more and more energy storage companies, hey, can we voluntarily do the same thing in Massachusetts, in Missouri, in New York? We said, yep, here's the equivalent data. Then we started hearing from people in China, in Italy. They said, do we have the equivalent data? And we said, call your version of the EPA. And we started hearing, actually, no other country has data as good as the United States EPA. I didn't know that. So people forget that in the 1970s, the United States went way further than any other country on the planet to mandate transparency on emissions, not action, just transparency. And so, we were suddenly in a world where Europe has much more aggressive laws about environment, but America had more transparency. And it actually wasn't technologically possible to reduce emissions cheaply in Europe, but it was in America. And that left us in a really interesting place.

Speaker 2:
[14:49] And just to pause there, too. Sorry, I'm going to make you stop all the time because you're like 14 times smarter than me, and so I just have to break this down in chunks. It is so remarkable to me that we haven't realized this lesson of just you cannot change what you don't measure. And I love that we faulted our ways into it as a result of some decisions made in the 1970s, but I also want to pay attention to the fact that you have to have that infrastructure in place. Gathering data doesn't just happen because you want it, right? It's not something you rub the genie bottle and it shows up. No, you have to really invest in that data collection infrastructure. And I'm really glad we did. And I'm guessing Europeans wish they had to.

Speaker 3:
[15:39] And so we were getting inbound from companies in all over the world saying, what would it take for us to do the same thing? And I said, well, first, you could have a 10-year program to stand up an equivalent of the UPS EPA Grid Monitoring Program. I forget about it.

Speaker 2:
[15:52] Yeah.

Speaker 3:
[15:53] And then we hit upon a kind of crazy idea. So we had some physicists on our group of volunteers. And some of them said, do you think you could point a satellite at power plants in other countries and train a machine learning AI model trained on US data, but using the new satellite data, satellites have come way down in cost, so there's way more than there used to be, and look at photographs of power plants around the world every few hours, and determine and protect what are their emissions, and build a cheap equivalent. And I thought that sounds crazy. But we, for fun, as a training exercise for our staff, said, why don't you write this up for a grant for google.org? Wouldn't that be funny if they agreed?

Speaker 2:
[16:38] Again, your definition of fun is just wild to me, but I'm so glad you answered.

Speaker 3:
[16:45] But we did it with no expectation that we would actually win. Google calls us and said, yes, we will give you millions of dollars to try a crazy might work AI solution to monitor all the power plants in the world and provide a free database of their emissions for anyone to use for any purpose.

Speaker 2:
[17:04] Wow.

Speaker 3:
[17:05] Whoa. Didn't really expect that. It was a coalition of three nonprofits that kind of teamed up on this thing. When we announced the project, Al Gore read the message. Apparently, Al Gore had envisioned for 20 years that if the technology were ever possible to do something like this, he thought this was the key because he was a fierce believer in the power of transparency and technology to change the environmental conversation.

Speaker 2:
[17:32] To Al Gore's credit, he was thinking about tech way before most Americans. It's cool to hear that he was doing. Yeah, no, no, definitely. Okay, so Al Gore is reading these DMs, finding out this random information. How does he then get inserted into all of this?

Speaker 3:
[17:47] So Al Gore literally had had a news alert set up for years in case anyone ever announced that they had built this technology because he saw all this going.

Speaker 2:
[17:56] I want to know what the keywords were, like Gavin, Berkeley, satellite.

Speaker 3:
[18:02] Machine learning, I'm not quite sure.

Speaker 2:
[18:05] That's awesome.

Speaker 3:
[18:07] But so we get a call from Al Gore's office. There's a hilarious moment where at first we don't believe it's Al Gore's office, but it turns out it's a real deal. And so he invites me into his office and he says, you know, what a wonderful project, but Gavin, the climate movement is bigger than just power plants. Is there any version where you could take this innovation and apply it to everything in the world that pollutes? And I said, no, sir, I do not know that.

Speaker 2:
[18:28] No, no, just a small ask. Just, you know, measure everything, Gavin.

Speaker 3:
[18:33] And so, you know, I said, look, no, I don't know how to do that. You would need experts in so many fields. And then the fact this project just happened to be different nonprofits teaming up anyway, we hit upon this idea. What if we just keep adding more nonprofits with more different types of environmental expertise?

Speaker 2:
[18:51] Your own kind of scaling laws, the nonprofit scaling law.

Speaker 3:
[18:54] Yeah. So we hit upon kind of like a Wikipedia model. What if we have, we make it really easy for nonprofits to join when they contribute data to the project. It's got to be free and open source for everybody. But unlike Wikipedia, it's a little bit more structured data, so you can actually use some clever AI algorithms to make the whole thing more accurate. Fast forward a few years, 150 nonprofits and universities and tech companies have joined the Climate Trace Coalition, which is now the largest provider of emissions data in the world.

Speaker 2:
[19:23] Wow. So to break it down for a second, basically what we've got, satellites hanging out in low-earth orbit, taking a look at all of emissions, power plants, data centers, cars, you name it, and by virtue of monitoring that, then training on sophisticated data to understand, okay, whatever I'm detecting, that's going to result in this much emissions. And then having that be refined over time where, for example, if I am a power plant operator myself, I could call up ClimateTrace and say, hey, it's in my interest actually to show people that I'm among the most efficient power plant operators.

Speaker 3:
[20:07] That's why it works. The good ones want transparency.

Speaker 2:
[20:12] So the whole idea here now is, if I can have ClimateTrace verify, hey, power plant A is actually far more efficient than power plant B, that can lead to perhaps lower regulatory fines or regulatory scrutiny, so on and so forth. And even your customers may appreciate that, so on and so forth. And so this sort of transparency actually makes sure that the green watchers of the world, the folks who are trying to sound environmentally friendly, but really weren't, get called out.

Speaker 3:
[20:42] That's totally true. And so one of the big surprises for me, so if you can't tell, I'm a San Francisco Berkeley liberal. I have all my views on government.

Speaker 2:
[20:49] Hey, as a Berkeley grad, I'm a fellow bear, but I just had to walk further up the hill than you. There you go.

Speaker 3:
[20:57] Then I think the big surprise is how much I'm seeing a free market answer in this story. So my surprise as a person who believes in government, actually the government mandates are a very small piece of the story. And the transparency of the US EPA was really important. And the transparency is an equivalent government. Your Uruguay did the same thing on cattle, their way ahead of America on cattle emissions.

Speaker 2:
[21:20] Who knew?

Speaker 3:
[21:21] Who knew?

Speaker 2:
[21:22] Go, yeah, yeah. Moo for Uruguay.

Speaker 3:
[21:25] Moo for Uruguay. And so by stitching together all these government transparency programs, whoever's best in class in every sector, that's our training data set. And then applying all this transparency, I would say 80% of the emissions reductions that result are free market, not government based. It is not a story about mandates.

Speaker 2:
[21:46] And what strikes me as well is that government intervention here would actually inhibit your ability to do this, which is to say when you have the government specify, hey, this has to be the approach or that has to be the approach, you may have missed out on the sort of detection of what was truly more environmentally friendly, and allow for that kind of rapid response as a result of that transparency. Is that a fair take?

Speaker 3:
[22:12] I think that's exactly right. And so just to be very careful, I'm not saying government should sit on the sidelines. I'm saying role of government, and I completely agree with you, it needs to be in terms of outcomes. So the Biden administration, much as I love a lot of what they did, they did push through a rule that actually requires measuring a particular metric they thought was going to be helpful on emissions. Turns out to have no scientific credibility and related emissions, awkward.

Speaker 2:
[22:38] Awkward. Hate when that happens. Hate when that happens. Hate when I bake a bad idea into law and just assume it's going to work out.

Speaker 3:
[22:45] So governments that have instead passed laws like you must measure emissions, you must reduce emissions, let the science go where it goes. Often the Biden administration got this right. There's one case where they really got wrong. That radically outperforms government trying to pick the winner in a fast moving environment.

Speaker 2:
[23:03] And having that sort of outcome-oriented approach is more data intensive. It's more, you know, requires a higher degree of actually paying attention to what you're doing. It's more work and it's more work. And I think we've become accustomed to a sort of set it and forget it approach to the law. But as you've noted and as you're seeing in real time with ClimateTrace, we have access to more and more data. And so in theory, this should be leading to more and more outcome oriented legislation to hold ourselves accountable.

Speaker 3:
[23:35] And so I totally agree. And so I don't have the same legal background you do, but I did used to work in the United States Department of Energy. And one of my big discoveries is that there is a deep seated belief that the thing that is greener surely is more expensive. And it's shocking how often that's wrong. And so we had all these energy efficiency standards where we concluded that the higher energy efficiency technology was also cheaper. And the reason demanded is people were just confused, but there was like literally you give up nothing. And so I think what data and science can do is they can create, I love the name of the Abundance Institute, they can show you how often the right answer is there. It's a false choice. You can have your cake and eat it too. You can do the greener thing. It's actually better business, better operating conditions. There is no catch other than you need to look at data.

Speaker 2:
[24:25] Exactly. Okay, well, I usually try to find a bumper sticker for each episode, and there is no catch. You just need data. That is perfect. I'll send you one when I finish it.

Speaker 3:
[24:36] Awesome.

Speaker 2:
[24:37] So let's move now into the highly contested territory of data centers and AI development, because everyone and their mom has heard about how data centers are destroying the planet and how they are using just incredible amounts of power and really putting a strain on our energy grid. What is your perception of that story and that narrative from ClimateTrace's view?

Speaker 3:
[25:05] So first of all, every data center that causes pollution, ClimateTrace is tracking. So in our database, we've got the world's data centers. And so we look a lot at the data. And the short version is one half of that story seems to be right and one half of that story seems to be wrong. And so do data centers use a lot of electricity? Heck, yes. It is actually true that some of the AI companies now use more electricity than entire countries. I believe Amazon is the same as Ireland now.

Speaker 2:
[25:38] Oh my.

Speaker 3:
[25:39] That's real.

Speaker 2:
[25:41] That is bonkers. Okay, so Ireland and Jeff Bezos, or excuse me, sorry, Andrew Jassy, that is insane.

Speaker 3:
[25:50] So that's happening. People are not wrong about how much power this thing is using. Here's the catch. When people ask, is this thing polluting, they typically look at the thing that is easy to see. So we know that using power is supposed to be bad for the environment, people are using a lot of power, they assume therefore that pollution is coming from these things. What they don't know is that Amazon was also by far the world's largest builder of renewable energy last year, for the third year in a row. And so if you ask, are they using a lot of power? Yes. If you ask, are they using fossil fuels? Well, actually most of that was wind and solar. So hold up, if you build a whole bunch of wind farms and you use a whole bunch of power, you're not going to use up the wind. It's not like if we have more sailing ships, we run out of wind. Right.

Speaker 2:
[26:42] Thankfully. That was good for the Mongols, I heard.

Speaker 3:
[26:45] Yeah, there you go. That's right. That's right. So typhoons aside, in all seriousness, the question isn't how much electricity are people using. The question is how much dirty electricity are people using. And when you look at data centers that way, it is a completely different story than people know.

Speaker 2:
[27:02] And this is where I get so frustrated, not just by environmental headlines, but you can apply this to a lot of different domains, but how quickly the first headline you shared, right, which is Amazon and Ireland use the same amount of power and people are instantly just gut reaction. Amazon's the worst, right? Like stop data centers, this has to be bad. And yet it's just the second and third question that if you go and ask, you can realize an entirely different answer.

Speaker 3:
[27:32] Totally. And so what I'm seeing is a lot of misunderstanding. And by the way, you know, Microsoft is doing even better. So like a lot of these companies are great. So like Apple and Microsoft and Meta now have essentially hit zero emissions from data centers. And it's all going as renewable energy. And a lot of other companies are investing heavily in renewable energy. There's a lot of innovation happening. Google invented this thing called a, essentially popularized this thing called a virtual power purchase agreement that allows you to build a wind farm not on site, a little further away, so you can like go where the wind is better and stuff. I'm thinking, not thinking quickly enough about the other innovations, but like a lot of innovations are happening by these folks.

Speaker 2:
[28:16] And just to pause there too, to go back to your free market point earlier, it's in the incentive of the hyperscalers to have these be as efficient as possible, right? Because I have a lot of tired lines, but one of them now is today's data centers are the least green data centers we'll ever have. Is that a fair statement now that I've got an actual environmental scientist? Hold me accountable, Gavin, please.

Speaker 3:
[28:41] Well, you know, the punchline is the market is telling us very clear that what data centers are going to do is they're going to build these things. You know, whatever your opinion is on AI, and I'm no expert on the rest of it, but like it's going to get built. And they need the power. And the truth is right now, it's increasingly cheaper to use clean energy than dirty energy. And it is increasingly faster to build clean energy instead of dirty energy. People miss this part. It's slow to build a gas turbine.

Speaker 2:
[29:09] So it's also pretty hard to, you know, build a mountain out of coal. I've heard that, you know, just takes a few million years, and then you've got one.

Speaker 3:
[29:17] Oh my god, refreshing coal. And so, you know, there's such a fundamental logic to these guys' power at all in clean energy. And yes, it's good for the environment. And yes, also it benefits them. Good for companies to make more money if the way they do it is clean energy, I say. But they're not all doing it.

Speaker 2:
[29:35] They're not all doing it. So let's go there. And then I want to get to the fact that other users of power are learning from, let's call them the good ones. But let's get to the bad ones first. Tell us about all data centers not being equal.

Speaker 3:
[29:48] So I guess what I wanted to say is that I think the story is, like we're saying at the beginning, it's not that all AI is good. So Elon Musk, to his credit, and the XAI company have not pretended otherwise, but they're powering Colossus 2 on gas, on fossil fuels. And I appreciate that they're at least being honest about it. Good for them.

Speaker 2:
[30:09] And for the folks who don't know about Colossus, the name is accurate here, it is incredibly expansive. I believe it's in Mississippi or is it Tennessee?

Speaker 3:
[30:20] I think it's near Memphis.

Speaker 2:
[30:22] Okay. Let's go with near Memphis. I think you're absolutely right. But Colossus and being powered by incredibly dirty energy, but unabashedly so.

Speaker 3:
[30:34] Yeah. And so people like me appreciate the transparency because you can at least have a conversation about how do you fix it, although that's a lot of pollution. And so what I wanted everybody to see is that all of the hyperscalers as this race to get power faster and faster is heating up. They're all at a crossroads. Are they going to keep investing in clean energy or not? And I think we should not use the paradigm of all AI is bad for the environment, all AI is good for the environment, but which one did they choose? And I think there's a lot of history of pushing false narratives about clean. There's a lot of weird definitions of clean flying around there now, but at the end of the day, they either did or did not use clean energy. And if you just measure it in emissions with regional science, we are going to know because we can track this all with satellites, what was the answer? And it's not clear yet which one they're going to pick.

Speaker 2:
[31:31] And what is the determining factor there? Why are we on that precipice point deciding clean or not? What factors should people be paying attention to?

Speaker 3:
[31:43] So some of the complicated things that factor in these companies' decisions are it can be more difficult to get approval from an electric grid to build clean energy. That's a problem. We've got to deal with that as grids.

Speaker 2:
[31:55] It sounds like lawyers fuck something up yet again.

Speaker 3:
[32:00] You know, some of this is legal innovation. And so there's also questions like there's this concept of 24-7 energy floating around. It was a very well-meaning attempt to raise environmental standards, but it actually isn't scientifically credible that adopting 24-7 reduces any more emissions. And the problem with 24-7 is that it is about six times more expensive. So one of the trends we're seeing is that if a bunch of well-meaning people from the public kind of attack any AI company that's not using 24-7, well, then you're going to make renewables six times more expensive, and suddenly the math doesn't pencil out for it's cheaper than fossil fuels.

Speaker 2:
[32:40] And so by 24-7, you're meaning that the public's screaming, hey, if you're building this power plant, building this new data center, we want you to be operating all of the time, using all of it extensively. Is that the correct understanding of what they're clamoring for?

Speaker 3:
[32:55] That is what people meant when they started this thing. But a little detail of 24-7 is you have to build the clean energy physically nearby the data center. That's just not how power grids work. It's very common you turn on a light switch and you turn on a power plant thousands of miles away. And so a well-meaning effort, hey, you got to build the wind farm and the solar panel nearby so we know it's real. That is the heart of what is making it so expensive suddenly to power data centers on clean energy. And it doesn't do anything for the environment to have the solar be nearby or far away.

Speaker 2:
[33:30] And I want to hit on a point that you and I discussed earlier, which is we're actually seeing that some of the power regulators themselves are beginning to learn from the best practices of the cleaner AI labs. So what are those innovations that are now spreading, that are actually helping us be more green across the board?

Speaker 3:
[33:50] Yeah, yeah, yeah. So my favorite first example was the California Public Utilities Commission. In the battery story I told you, the tech companies actually got a little bit involved in that story. So it is not totally a coincidence. Microsoft was playing around with how to use batteries at a cleaner time and was part of the conversation of California regulators realizing, if you're going to have batteries anyway, what is the correct time? But then it gets more subtle. So another thing right now is the California Air Resources Board. This is my last California story.

Speaker 2:
[34:24] That's okay. Hey, being the fifth largest economy in the world and 40 million people, they do merit some degree of outsize attention.

Speaker 3:
[34:34] They sure like environment. So, yeah, the Californians right now think about regulating. It will be the first US mandatory emissions reporting program at the state level. And how should they define emissions is a live conversation right now, where they are very much looking at what the tech companies are innovating and doing. And like one of the things that I think surprised regulators at CARB is how often companies are willing to voluntarily reduce more emissions than they were forced to. If they can report it in a way that better matches the latest science and data, which is to say like thinking at the five minute level and the local level as opposed to vague annual guesstimates. And so regulators were realizing that companies would support a more ambitious environmental regulation if it just used better data. That's different than your old assumptions about how data works.

Speaker 2:
[35:31] Wow. And so you have this great trove of data. You're improving over time. You're holding folks accountable. And you're making all of this information accessible. Are people listening?

Speaker 3:
[35:43] Yeah. I mean, my experience has been you're always going to get pushback from... Interestingly, it's not all fossil fuel companies. It's the oil industry. For whatever reason, the coal industry tends to be open and honest about, yeah, we're a terrible environment. It's always the oil industry that tries to trick us. Don't know why.

Speaker 2:
[35:59] So the coal guys, cool cats, fully transparent, but don't trust an oil man?

Speaker 3:
[36:04] It's bad.

Speaker 2:
[36:07] Fascinating.

Speaker 3:
[36:08] I don't know why, but there's a much longer history of oil companies pulling funny business and arguing. So we spent a lot of time showing satellite images of like, hey, if your emissions were the numbers you say, guys, then this oil field shouldn't exist. Like here is a photo of an oil field that can't exist if your numbers are real. That's the kind of argument we're often in. And then I do find that there are a decent number of people who are making money selling some product that they probably honestly think is green. And when data comes out and says, actually, it's not clear that's good for the environment, that really can be tough for people. I won't name names, but there are definitely plenty of startups that perfectly well-meaning innovation, science advanced, turns out that's not good for the environment anymore, and it's really hard to go to your VC board and say, I think the company should just not make money anymore.

Speaker 2:
[36:57] Yeah. And so what this really speaks to me on is the need for just making space for more consideration of data in terms of our legal agreements, in terms of our public discourse. How do you think through that sort of translation effort of how do we make folks more empirically driven? And I'm not, I know that seems like a duh statement, but you can go talk to a behavioral scientist, and they'll tell you folks are compelled by stories, folks are compelled by narratives, so on and so forth. So do we need that sort of legally mandated outcome-oriented approach? Or what is your kind of wish list of things that we can either do culturally or legally to try to make your job easier?

Speaker 3:
[37:44] I think my biggest lesson is if you look at what real money is in putting money behind the things we need to stop climate change, you know, the money to pay for EVs, pay for wind farms, pay for green steel. Most of it is not driven by government mandates. Most of it is driven by companies want to look green. They want to be responsible to their investors and so on. But what governments need to do is they need to set the rules of the road of what is green and what isn't green. And it is so important when they do that, that they use proper science and not some buzzword they found on the Internet. And so I think, you know, at the end of the day, one of the biggest roles of governments is to set the rules of the road. Like imagine if we didn't know if you have to drive on the right or left side of the road. We got to have standards.

Speaker 2:
[38:28] I don't think I'd imagine that because I'd be dead. But it's a disaster.

Speaker 3:
[38:33] And so nobody needs to force you to drive on the right side of the road. It's more about coordination problem. And I think that we have this paradigm that we need to push companies harder to be green. And governments can actually be in a helping, supportive role, saying like, guys, these are the rules. And it's based on science. And if you, a company, voluntarily invest in going green, you won't be caught flat-footed suddenly against the rules. That's a very different vision for what law can do in an environment. And it's coming from a liberal here. So I'm, you know.

Speaker 2:
[39:06] Hey, hey, I love the vision. Well, and I love that it's not liberal, it's not democratic, it's not conservative, it's not Republican. It's just empirical. It's just what works. And that's something we should all be wildly excited by. And I think recognizing too, just to zoom out for a second, you all leveraging AI in such a creative fashion would not be possible without, A, the collaboration that you've seen across different nonprofits, that sort of willingness to get involved, that willingness to share information. But then, secondarily, being willing and open to share data. And this is why I have a lot of issues with folks who say, AI sucks. My first response is, okay, but what data was available? I'm sure you can testify to, and I'm eager to get your feedback on. I'm sure your first algorithms for, let's say, measuring cattle in Missouri, probably weren't great, right? Because you didn't have as much data. Yeah. But this act of, actually, if we want better AI, you have to give it more data. It's not just a magical process of more compute or more algorithms. That data is going to be- It's still fundamental. So how much of that is your, are you a sort of like data hoarder right now? Do you just like go knock on doors, beg people for more data or what's that process like?

Speaker 3:
[40:32] So I mean, I really think you're on to something. And so I would say that- So ClimateTrace is this big coalition, right? We look at satellite data. We are the 20th project that we know of that tried to get everybody to share all their data to build better AI and satellites and measure emissions. The first 19 that we know of failed, and they all failed for the same reason. They all failed because the project model was there is a guy, because it's always a guy, who's in charge, and everybody else needs to give him data, because it's always a him, and there is not really a reason why. And so what we saw is that a notion that one plus one equals three, that joining data makes us all more powerful, it breaks down if the model is someone is in charge and everyone else is subsidiary to them. And so...

Speaker 2:
[41:21] So get men... So eliminate the men first.

Speaker 3:
[41:24] And starting with the gender thing, but it is striking how it's usually not a woman.

Speaker 2:
[41:30] This is empirical as well. 19 out of 20. Right.

Speaker 3:
[41:34] And so we don't have a leader of ClimateTrace. There is no CEO of ClimateTrace. All of us have equal access to the same data. Certainly people spend more time and less time on it. And certainly the fact that everybody respects Al Gore means his opinion kind of counts more than your average person. But we do things as votes. And the key to it is that the design of ClimateTrace gives everybody who's there a reason to be there. And it's not a you are just here to support me. We're all there to support each other. And that is a total paradigm shift of why people would share data. And it's why the project's still here.

Speaker 2:
[42:12] Well, and what's crazy, too, about your model is that once you put data at the heart of governance and once you put data at the heart of this sort of new way of thinking about a policy area, you create the incentive for more transparency and for more data gathering. And so the fact that I'm pulling from a separate podcast you were on, I apologize, but you mentioned, for example, that there is a construction company that actively shares data with ClimateTrace because they want to expose their competitors as being less green. And so that's only possible or that's only in their self-interest if they know that sharing that data then leads to a positive outcome. And so we have to think through in all of these other domains, I hope it keeps occurring in the environmental front, but in education and in health care and in so on and so forth. How do we make data and AI actually at the heart of that process so that we have better outcomes?

Speaker 3:
[43:13] And so I hear two things in that story. From the perspective of the people running the algorithms, the key breakthrough on ClimateTrace was instead of starting with the assumption that people will give us data, data are valuable and takes time. Why would you join and the coalition needs to go wherever it makes sense for the people contributing? The other thing is why would the people who are being measured want to be measured? And I think there I always go back to what I saw at US. Department of Energy. This is actually a story that happened in Japan, but we were tracking it. Pepsi and Coke got locked in a battle about who could be greener. And so they started competing on clever ways to go greener. And then under Japanese law, you always have to be as green as the business company was five years ago or something like that, under the top renit runner program.

Speaker 2:
[44:03] Okay. Good for Japan. Sounds fascinating.

Speaker 3:
[44:05] Clever program. Why does this matter? Because anytime they could invent a new way to go greener and not tell the other company, they got like a little leg up on their competition. And so what I saw is this amazing runaway trend of companies had a hard business reason to go green because it gave them a little bit of edge in the competition. But then the only way to beat them was to have an even greener thing, and so my data discovery is that in almost anything you're measuring, somebody's going to be above average and somebody's going to be below average. And if you can set up a system where everybody who's above average has an incentive to donate data, and then they have to keep working to stay above average, you build this like runaway collaboration train that doesn't need anybody, any laws forcing anybody or any carbon taxes.

Speaker 2:
[44:53] And no laws to force that. And you get the innovation of finding out that new thing, right? Instead of saying this is the mandatory mechanism you have to use, now Coke and Pepsi, of all institutions, are experimenting, are innovating.

Speaker 3:
[45:11] They're innovating, discovering what works. Sometimes they're wrong and then the other guy gets a win for a minute and then they catch up. And I think that that is a bigger story than environment. And I think it's a bigger story than, you know, it was cool in Japan, there was this mandate. But I think what I find is usually companies just need a way to tell their customers, look at us, we're better than the competition. And that's all it takes.

Speaker 2:
[45:34] That's all it takes. We're pretty simple. We just want to know what the best is and we'll pay for it. And, you know, to your point, it is bigger than the environmental sector. People want that in education. People want that in healthcare. People want that in mental health. All of these areas where I get that it becomes more sensitive, right, when folks think about sharing health information or sharing educational data. But this is all just a matter of creating new institutions and creating new norms. And to me, this is just a matter of creativity, right? It required you, Gavin, hanging out with some homies in San Francisco and deciding, hey, let's take a risk, let's be creative, let's build this new approach. And the other 19 failed because they didn't have the same degree of imagination and dedication to this new model. But there's no reason why we can't copy you, Gavin, in so many other domains. Oh yeah, totally. I apologize for, you know, it's a sincere form of flattery, but I am going to share this episode with everyone, and their mom, and their daughter, and anyone else who will listen.

Speaker 3:
[46:39] I'm already tracking in the biodiversity space, and there are hints of a ClimateTrace-like play, and maybe they are going to invent it better. And so what I love about this model is that we sometimes have this mental model that competes as a problem. But if these are issues that all of us want to succeed, and you know, health care is a great example, where I'd love to see other people succeed, it just kind of flips the script. If it's like anybody who can collaborate better, it's not a problem, it's a good thing. And I really think that somebody's going to come along with an even better version of the story tomorrow, and great, I'll celebrate their win.

Speaker 2:
[47:08] Yeah, move fast and collaborate. There we go, right? We'll change the SF paradigm. So, Gavin, before I let you go back to saving the world, any final thoughts, any final message you want folks to know? Where can they find you? Where can they monitor what hyperscalers are up to? Give folks the information they need.

Speaker 3:
[47:29] So, climatetrace.org has free information available to everybody. Our hyperscaler dataset is not public yet, it will be soon. You can always e-mail us if you want, find us on the website. But if I can leave you a second hidden message, please, I want to give Walmart a shout out. Walmart is my favorite example of a company that is not just going greener. They are saying, actually we want two things. We want to go greener and we want it to cost less than not going green. And so I think the big new trend that we're seeing that Walmart kind of kicked off is saying, if anybody can invent a better system to go greener that actually costs less than today, we commit, we'll buy it. And I think this is the future because I think if there's one story people need to know about AI and environment is, it's not harder to do it right. Get rid of the idea that it must be more difficult to do the right thing. Just check what the right one is and buy it and don't pay more for it.

Speaker 2:
[48:25] I love this. Advanced purchasing agreements have a heck of a lot of influence. Thanks Abundance. And Gavin, thank you for all the work you're doing. Maybe I will go buy some Walmart products, who knows. But in the interim, I hope to see you in San Francisco sometime. Thank you again for coming on Scaling Laws.

Speaker 3:
[48:43] Sounds great. Thanks for having me.

Speaker 2:
[48:48] Scaling Laws is a joint production of Lawfare and the University of Texas School of Law. You can get an ad-free version of this and other Lawfare podcasts by becoming a material subscriber at our website, lawfairmedia.org/support. You'll also get access to special events and other content available only to our supporters. Please rate and review us wherever you get your podcasts. Check out our written work at lawfairmedia.org. You can also follow us on X and Blue Sky. This podcast was edited by Noam Osband of Goat Rodeo. Our music is from Alibi. As always, thanks for listening.