Debunking AI's Environmental Panic | Andy Masley

Conor Bronsdon 0:00 AI is destroying the planet or so we've been told. Today, we're tackling one of the most persistent and most misleading narratives in the AI conversation that using AI is destroying the climate and leading us down a path of environmental destruction. I'm your host, Conor Bronson, and thanks to our presenting sponsors, Galileo, for enabling us to do this episode.

Conor Bronsdon 0:23 If you've been at a party outside of tech recently and mentioned your use of AI tools, you might have gotten the look. You know the one. Don't you know how terrible this is for the environment? What are you doing? Maybe someone even threw a statistic at you about water usage or carbon emissions. You may even have felt a twinge of guilt, knowing how to respond, or at least not being able to communicate the sense that look, our

Conor Bronsdon 0:49 world is so technologically connected already. Like, is this really the thing that is causing all these problems? And it's what most of us have heard in popular discourse about AI's environmental impact. Sadly, most of that information is wrong. It's off by orders of magnitude. My guest today is Andy Maisley. Andy is the director of Affective Altruism DC and

Conor Bronsdon 1:13 a former physics teacher who's become one of the most thorough fact checkers of AI environmental claims. Andy, one of the first articles I read from you was early this year titled Using ChatGPT is not bad for the environment. That is a bold claim in today's discourse.

Andy Masley 1:30 Welcome to the show. Great to have you here. Thanks so much, Connor. Yeah, super excited to be here and talking about this topic that I've been pretty obsessed with over the past year. Like you said, I've also been to those parties. Also gotten those looks. A big part of this is just motivated by helping other people in these conversations and easing out like my fellow environmentalists

Andy Masley 1:46 from something that I'm really honestly pretty worried is distracting them and, like, leading them down what I think is a bad path. One other flag here as we go forward in this conversation is that, for any listeners, I'm definitely not asking for zero scrutiny of AI or its environmental impact along every single axis.

Conor Bronsdon 2:01 And basically just saying that a lot of the common wisdom that you hear at these parties is pretty off base. And so basically just trying to correct the record there. I completely agreed. My, first podcast that I hosted was called Growing the Green Economy, talking about climate change and how to have a more effective, green future. And this is something I, you know, have done political activism on at a lower level as well. So it's fantastic to see people who are in this space

Conor Bronsdon 2:25 pushing back because I think you and I both agree like, you know, there are things we need to with AI, we need to guardrail it, we need to make sure it is responsible. We should make sure it's not having horrific climate impacts. But also if I'm really worried about something, it's probably our almond growth, our alfalfa growing rice in California. There are things that are much more water intensive and we're going get into a lot of those statistics and dive into it. But I wanna start with what really made us do this episode, which is,

Conor Bronsdon 2:53 you going a little viral recently. Yeah, yeah, At least in AI Twitter. You published a detailed analysis of Karen Howe's book Empire of AI that many of you have potentially read or heard about, that's been getting a lot of attention. And you found what you called the single most most egregious math mistake you've ever seen in a book about AI and the environment. So, this feels like the perfect place to start because it really crystallizes what's wrong about this conversation.

Conor Bronsdon 3:19 Can you walk us through what you found?

Andy Masley 3:22 Yeah. I I will flag super quick. That's the most egregious mistake that I've personally discovered on my own. I'm sure there have been other things So that I've totally stumbled basically, I write a lot about AI and the environment, as you said, and in the past few months, especially, I've been really doing a lot of deep dives on AI and water use, where I've come to the conclusion that AI data centers just really don't seem to be a significant issue for water in The United States. As you had mentioned, there are lot of other big culprits. My personal big villain in a lot of these conversations is actually animal agriculture.

Andy Masley 3:51 I have my own biases here where I'm a vegan for animal welfare reasons. And so, like, there are these other reasons I'm motivated to do this. But I would encourage your listeners, if you're worried about water use in America, like, maybe just look into that first before you get into the AI stuff. But basically, after covering AI water use in America for a while and coming to the conclusion that it's really not an issue here, I get a lot of feedback from a lot of different readers that maybe this is the case, but they're also under the impression that in other countries, this could be a much bigger deal. Like, I'd recently had a conversation with someone from India who had said that data centers are actually causing problems there. And so I was thinking, okay,

Andy Masley 4:28 I don't know very much about countries that are very different from America, especially very poor countries. I'd like to know the deal. Where should I look for this? And a bunch of people had recommended this one chapter in the book Empire of AI, and I was like, oh, you know, this is a really popular book, seems to be influencing the conversation a lot, so I should sit sit down and read this and just understand the case that these data centers are harming other countries a lot. And I don't have a final take on that, like, could very well be that data centers in other countries are harming things. I haven't done the full deep dive on every last country, but I I have to admit, every single place that I saw,

Andy Masley 5:02 the book cover water in this one chapter specifically, every single time water was mentioned, I came away thinking that the reader would be left with a much worse understanding than they came in with. So this is more than just the one stumble I saw, as someone covering this for a while. There were a bunch of other places, including the very first place where water is introduced, where it said that by 2027,

Andy Masley 5:24 AI will consume about half as much water as all Britain uses. And using just like the normal water management definition of consumption, the study that the book was referencing actually says that all AI together in the world will only be about 5% of Britain, which again is not nothing. But again, Like five it's or 50% is a Yeah. Again, is like an order of magnitude issue. And like that other,

Andy Masley 5:45 like 90% of the water is actually gonna be like taken and used in power plants off-site and then returned to the source unaffected. And I think, again, this is a place where readers are just kind of being left with, a bad misunderstanding. But yeah, the main way that I personally blew up on Twitter in the last few days is I stumbled on a specific mistake that the author made. And I don't attribute malice to the author, by the way. It seems like Karen Howe has done a lot of, like, really great reporting in other ways. I definitely don't want this to be a commentary on the broader book overall, because I haven't read most of it. I had read her, coverage of, like, OpenAI, Sam Altman, like, drama stuff, I found that pretty useful, honestly. So again, don't want to say that the entire book is bad, like She had some legitimately really interesting insider access and perspectives on the people side, whether you agree with her takes on it, fair enough, but like there there's interesting nuggets there for sure. Yeah. And so, I definitely don't wanna make the claim that the entire book is bad, that people shouldn't read it, or that it's invalidated in any other way. But I do really wanna push that if you come away from this one chapter,

Andy Masley 6:46 it's very, very likely that you're actually coming away with a much worse understanding of AI water use than you're coming in with. I was just pretty consistently disappointed with the coverage there. And so, like, people can read my blog post for my full arguments. The main thing I had stumbled over is halfway through that chapter, she claims that kind of the main focal point data center of the entire chapter is using as much water as a thousand times this community of 88,000 people living in this area of Santiago in Chile specifically,

Andy Masley 7:15 and that water activists have to be, like, basically clawing the water back tooth and nail and fighting with everything they've got against this really powerful corporation coming in from outside to guzzle up all of the water. This is actually off by a factor of about 4,500, Specifically, there were two big issues that happened. The first one, which the author addressed directly, and I was very grateful that she addressed it quickly, was that her source, the person who had sent her the information, had mixed up liters and meters cubed. Like, a meter cubed is a thousand liters specifically,

Andy Masley 7:46 and this went unnoticed. I was kind of surprised at how few people noticed this because the number that she gave for this community basically implied that the average person there was consuming about point one liters a day, which is something like a fifth of a single bottle of water, which, you know- They'd all be dead. They would all be dead, basically. And like, this isn't just for drinking water. This is like everything that their household uses water for, as well as like commercial buildings and stuff. And so that was the first major error where there was a massive mix up in units of measure. And so, again, I don't wanna say that I know for a fact what was going through the author or her editor's head. Here, like, seems pretty likely that this was an honest mistake. I do have to say that I was pretty stunned that this didn't get caught by anyone doing a sanity check. Like, this book has been out for about six months. Right now, it's been reviewed by a lot of very prominent

Andy Masley 8:32 news organizations and recommended a lot. And again, I'm not saying they shouldn't have recommended the book, but I'm just kind of surprised that no one anywhere just emailed the author to be like, oh, hey, by the way, this central claim you're making is pretty clearly wrong on the water thing. There's Did actually you do some the math, basically? Yeah. Did you do the math, basically? And did you just sanity check this? And so I was pretty stunned that after six months of this book being out, I was the first person to stumble over that. And my general claim in the post isn't so much that, like, this invalidates the book. I'm more focusing on just how bad the current conversation about AI and the environment and especially water is right now because so many of these numbers are just circulating around and so many of these intuitions

Andy Masley 9:10 are bouncing around and leaving people with really wild misunderstandings of what's going on with AI and data centers. And I do think that overestimating AI water use actually carries its own potential negative environmental impacts. This part I haven't actually looked into super closely, but from what I can tell, activist pressure on this data center, potentially based on this misunderstanding,

Andy Masley 9:30 convinced Google to use an air cooling system instead of a water cooling system for the data center. Air cooling actually takes a lot more energy than water, and so as a result, I suspect that, like, the CO two emissions of this data center are actually way higher than they would have otherwise been. And again, it might be that being off by a factor of 4,500,

Andy Masley 9:49 the data center still my claim is that it's using about 25% of the water in this one specific section of Santiago, which is definitely not nothing. But from what I can tell, it's about 3% of the water of the entire municipal water system. And again, if you look at that and you ask, would we rather use about 3% of the water here versus drastically increasing the emissions? My claim is that environmentalists at least need to be really well informed about this to make good decisions because a lot of environmentalism comes with these kinds of trade offs. So Policy is trade offs. And if we don't have the right numbers to understand those trade offs, we're gonna make mistakes, period. Yeah. And so, my claim is that this really matters. This reflects pretty badly. Again, not necessarily on the rest of the book, but definitely on the current broader information environment

Andy Masley 10:34 about AI and the environment, and especially around water use. And so, I was pretty shocked that I was the first reader to comment on this, and I really wanna raise the standards of the debate a little bit. Where, again, I don't wanna say AI deserves zero scrutiny, but I do wanna say that environmentalists being misled matters a lot, because environmentalism involves a lot of really difficult complex trade offs involving numbers.

Andy Masley 10:54 And so,

Conor Bronsdon 10:55 if we're getting these huge questions wrong, there's potentially a lot of other ways that the debate is just getting really tripped up. I have to give Karen some credit for her response to you once you called her out. Like, she did say, look, like, let's look at this and you know, I'm checking on this number. It does seem like the math may be off here. However, she also then followed that up with a variety of

Conor Bronsdon 11:16 arguably not well fact checked additional claims about AI water usage as if she didn't want to give up the point specifically. And Karen, if you're listening to this, you want come on and give your perspective too, we'd have happy to have a conversation, but we need to have a real conversation on the facts, not about the idea of what you want the facts to be. And I worry that

Conor Bronsdon 11:37 the kind of information environment as you put it has taken hold of like AI is bad for the environment because we choose to believe it's bad versus AI has pros and cons to it and we need to make sure that data centers are efficient but there can be a lot of value delivered by it. And I, I worry that this has just become the narrative and that it's hard to push back on it feels like. Yeah. Also definitely want to express a lot of gratitude.

Andy Masley 12:03 Karen was super cordial and very quick to respond in Twitter threads and on my blog about the main thing, and was very quick to send me, here's the source that I was working with for, you know, here's the number that the source provided to me specifically. And so that was all really useful. And I think she's engaged very well. In the past seventy two hours, I've definitely received a lot of Twitter comments that did not engage in good faith and were very goofy. That was exciting. And so Yeah. Definitely. And so I think like, Karen really stood out, especially compared to a lot of other people I've interacted with in the past on this as being extremely nice, very direct about this one issue.

Andy Masley 12:38 And I also just wanna flag that I both wanna make sure that a lot of people know that this issue happened because I think this has really influenced the debate. Like, this book was one of the most popular skeptical takes on AI and the environment in the last year in general. And so, like, I do wanna correct the record on this, which is why I was so public about this. But I also wanna just flag that this must have really sucked. Like, I, you know, I don't wanna like bully or be mean here. And so, like, drawing public attention on this, I wanna try to balance like, hey, it's really important that we correct the record, but I also don't wanna like come off as trying to attack like Karen's public online presence more broadly. And so, I'm trying to balance this all. Legitimately a very good reporter in a lot of areas. I think it it is telling as you put it that, you know, this book that has thousands of Amazon reviews was seemingly peer reviewed. I presume quite well edited, has now become widely cited and discussed

Conor Bronsdon 13:28 that no one else identified this or at least publicly identified this error in the last several months. Like, what what does that tell you about the conversation and the information environment we're dealing with? Obviously,

Andy Masley 13:39 like, I I genuinely think this number issue was a mistake. Like, I I think it would be ridiculous to intentionally put like an incorrect number like this in. I will say that like, Karen does actually have a degree in mechanical engineering from MIT. So I I do think that, like, it's reasonable to expect that we get the math right here. I think a lot of journalists

Andy Masley 13:57 are kind of relying on this built up common wisdom that has come out of a lot of different articles on this in the past. Like, if you just look at a lot of different coverage of AI and water specifically, a lot of the coverage frames it as being a huge disaster right now, where there are all these reports of these kind of abstract, very large amounts of gallons of water being used. It's like totally understandable if you're kind of running with this and you're running with this intuition that most people that you trust are saying is correct. Like it makes sense to potentially gloss over some of the specific facts on this.

Conor Bronsdon 14:27 And So you think it was more of a like, Hey, I trust these people already.

Andy Masley 14:30 We've a lot of work to do or just kind of moving forward. Yeah. And honestly, also just like perceiving AI as a bad guy more broadly. Like, I think if you perceive like, this is this act of colonialism, it's this like evil new tech that isn't benefiting people and is only harming people. I think, like, you're gonna be more inclined to also potentially jump on numbers that look bad without necessarily triple triple checking them. So that's my message. It's really easy to turn this into a tribalism argument of like, AI bad, AI good, we're not gonna actually look at the details of it. Yeah. And so I will flag like based on all that, like, again, I'm not trying to be mean or attribute any malice. There were like two other places with the reply that I was also disappointed with. The first disappointment was that Karen had attributed a lot of our other disagreements

Andy Masley 15:16 entirely to philosophical definitions, and I mostly don't think this is true. Actually, I think I make a good case in the post that there's one other specific place where there's a claim about how much water AI will use in 2027 that I think based on any reasonable definition of what it means to consume water is actually off by an order of magnitude as well. And so I am asking for that correction as well, basically.

Andy Masley 15:37 And so far, I don't see any indication that that's coming, but maybe that will change. And then the other complaint I've had basically is, yeah, she then went on to share a lot of other articles about how, oh, but AI is still very bad for water use for all these other reasons. And I'd lie I have to say most of the articles that she had shared from Bloomberg, I've had similar issues with. I've actually covered this in the past where they'll basically just make these big maps of The US of where data centers have been built in water stressed areas without sharing what proportion of the water that area the data center is using.

Andy Masley 16:09 And a lot of the times, if you look into it, one, these are very small proportions of the water. Two, there are lot of other industries that use a lot of water in those areas because even cities built in water stressed areas need taxable industries and stuff, and so they build a lot of other stuff. There's also, like, water parks and golf courses and other kind of egregious,

Andy Masley 16:26 like, I I would consider kind of wasteful uses of water in these very water stressed areas. And so We we can do a whole separate episode or just the two of us talking about golf courses and in my opinion, it's good off. Yeah. Yeah. I know some folks here listening are probably big golf fans. I'm not gonna get into that. Yeah. Yeah. Yeah. Again, like, if you want like a big water bad guy, like, maybe just focus a little bit more on golf courses, obviously. So, yeah, those were the other two places that I

Andy Masley 16:48 wish the reply had been a little bit better, where it's like, I would like you to basically acknowledge that if you were off by this magnitude, this tells you something about your broader narrative as well. Like, the central theme of that chapter specifically is that AI water use is this extension of brutal colonialism that happened in this country, where, like, there's a lot of mention of how elders have these memories of really horrific

Andy Masley 17:11 crimes against humanity happening in the region. And like, it's then kind of framed as like, oh, and AI is this new thing that's drinking up all the water by these big colonialist, you know, entities. And I kinda wanna say that like, this claim seems to be somewhat invalidated by the idea that this data center is actually using such a small part rather than such a large part of the overall city's water, basically. And so, I would've liked some kind of acknowledgment, but I realize,

Andy Masley 17:35 the author has also probably had a really wild seventy two hours. Like, I wouldn't have wanted this much attention on myself personally. So I I wanna give a lot of grace here and like, expect that like, you know, maybe in the next few weeks or months, like some additional clarification will this podcast probably isn't gonna come out for three or four days. Yeah, yeah, yeah. It's possible we'll see even updates, since then. And so I'll try to amend and add something to it, but

Conor Bronsdon 17:56 it does feel like a really difficult conversation because to some extent, this debate has kind of turned into the classic two camps of like, you know, this is good, this is bad. In many cases, are just trying to find supporting evidence for their claims without actually interrogating their claims. This is very common where, you know, emotions and viewpoints and your, your preconceived notions

Conor Bronsdon 18:20 can kind of lead you down this path. I've done this myself in many cases in my life. Yeah, same. So, think it's really important to be critical and thoughtful about how do we push back and unwind this narrative because simply saying, hey, look, we're gonna attack you on these facts and it's an attack is not going to work for us long term. If we want to ensure AI is something that is beneficial for, for everyone. So, from your perspective,

Conor Bronsdon 18:49 how do you think people who legitimately want to have the conversation about the potential benefits of AI, what it can do for people and also avoiding the drawbacks? How should they be pushing back on this challenging narrative?

Andy Masley 19:03 I think that AI is going to bring a lot of benefits and a lot of risks and negatives. Like, wanna flag that I'm around a lot of people who are not just gung ho. We should just, you know, build out data science and AI

Conor Bronsdon 19:14 sort of p doom of like, this has potential really negative effects. Exactly. And even outside of like very extreme

Andy Masley 19:20 p doom situations, even if all of that goes well, I can totally see a lot of people saying, oh, AI progress in the next few years might erode a lot of other just important aspects of society, ways we can I have opinions about the surveillance state, I'll say? Like, there's plenty of books of other concerns, and I I think it's really legitimate to have conversations. Every technology

Conor Bronsdon 19:37 Mhmm. Creates a lot of change and that change can cause upheaval,

Andy Masley 19:40 positive and negative. Yeah. And so there's a ton I wanna say about this. I think first of all, if you are going to criticize such a broad general new technology like AI, it's just really important to actually focus on what's causing harm because everybody does have a limited amount of bandwidth for thinking about this stuff correctly. And if you hyper focus on, oh, AI is using some water, one, that might just push more data centers to use air cooling, which in my opinion is actually worse for the environment- Agreed. For the most part. And also, it might just distract from all these other things. Like, I guess, if I think about, like, oh, you know, so and so government is now using a much more effective system of surveillance on its people using advanced AI, and also every time they do that, they use a few drops of water. I actually think that second point kind of dilutes, you know, in kind of a punny way,

Andy Masley 20:29 the main argument. Like, if these things are happening and these bad things are happening, I I don't actually think just slathering on any and all additional criticisms are going, is going to help your side. Like, I've been vegan for ten years personally. I'm pretty worried about the state of animal welfare in America. And if I see someone eating a chicken sandwich, as an example,

Andy Masley 20:49 I actually think it's kind of not super useful to say, oh, and that chicken sandwich also has a slight environmental cost, because it's not actually very bad. Chickens aren't especially bad for the environment. My claim is mostly just that they're they're suffering a lot, basically. And so, because AI is gonna come with all these new huge problems and potential good things about it as well, I think just getting this right is gonna matter a lot, because we're gonna be faced with a ton of trade offs in what to think about and do. And if the debate is being so overwhelmed with things that I just think aren't real and are problems that are basically a distraction,

Andy Masley 21:20 I think that's mostly just gonna cause us to make worse decisions. And this matters for the environment more broadly as well. Something I try to get across in my writing a lot is that very few people actually think about the environment at all. The average American seems to have a few conversations about climate change every year. And so if you're so fired up about climate and energy and water stuff that you're focusing on it enough to really worry about AI, I would say that your time is actually very precious because you're one of a small minority of people who are actually doing this. And I really don't want your attention to be too focused on something that might not actually add up. And then the very last thing I'll say is that, yeah, I think people have unfortunately developed

Andy Masley 21:55 a very kind of, like, team y way of thinking about this where people will see me saying, oh, AI water use isn't a big deal and assume, oh, Andy must be some awful tech bro who's trying to push this technology on us. And I associate him with every other bad thing that's happened with AI, and he's trying to attack one of my side's key arguments, and so therefore, he must be bad. And I just don't think this argument is gonna work as we go forward because AI is just such a general technology that's gonna affect so much in so many different ways that thinking about it as, we need to make sure that our team always wins regardless of the strength of our argument is just gonna lead us into a lot of goofy and I think mostly incorrect conclusions. So these are all worries I have that I think, honestly, the AI water use conversation is really helpful with just because it's I think it's such a glaring massive error in how people are thinking that if we can't even get this right, I'm pretty worried about society's broader ability to just adjust to AI systems coming online.

Conor Bronsdon 22:48 And I saw a couple of these responses to you where people were basically saying, oh, Andy's a tech bro. Andy is an effective altruist. Sure, sure. We just can't trust him. Know, like that, you know, been proven wrong, ignore Let's maybe talk a bit about your background and contextualize this. I, and I know some folks who aren't familiar with EA as much like, you taught physics for years.

Conor Bronsdon 23:07 Mhmm. You have quite a bit of experience diving into the numbers, whether it's on climate or other topics. But maybe let's start broader. For folks who aren't familiar, what is effective altruism or EA and why were some folks attacking you for that? Yeah. There's a ton to say about that.

Andy Masley 23:26 So EA very generally is a broad movement, that's very heterogeneous. I would say it's actually kind of eight or nine unrelated movements all under the same similar umbrella. The main idea is that you should try your best to use evidence and reason and debate to figure out where you can do the very most good with either your charitable donations or the time that you spend in your career, basically.

Andy Masley 23:51 And so EAs in general, have landed on three big buckets of places that might have the very most positive impact on the world that other people might be missing. Another key aspect of EA before I go on actually is the idea that a lot of people just actually don't think about this very much. They kind of just fall into whatever local To your point about how many conversations about climate change am I having per year on an African American, for example. Exactly. 100. Yeah. And basically, lot of people will kind of just be attracted to, oh, this thing makes a lot of money or this thing will give me a lot of social status, and so this is what I'm gonna do, you know? And I think that's totally understandable. I think the average person is probably having a pretty net positive effect on the world, so I'm not saying that's all bad. But the basic claim is that because

Andy Masley 24:31 either markets or society doesn't price a lot of other problems in in the way that maybe they should be, there's often gonna be a lot of value left on the ground for people who just step back and just look at the numbers. This comes up a lot in global health conversations in EA, about the idea that almost nowhere you can donate, will do nearly as much good as just donating to charities that alleviate malaria, specifically Malaria, for example, yeah. Yeah, exactly. Like something like five hundred thousand people a year are still dying of malaria, and I think that's news to a lot of

Andy Masley 25:00 million people are dying of this disease. So there are probably a lot of other invisible conversations that you can bump into, or invisible problems that you can bump into, if you just step back and look at the numbers, basically. The way that EA gets really controversial really fast, which I I totally, actually, I wanna flag here that e- if you've enjoyed my writing, I'm very grateful for that. I'm not really trying to attribute,

Andy Masley 25:20 like, every EA argument to me or, like, endorse everything that's happened in EA here. Want you to approach this kind of skeptically, where I'm talking about a very specific, very data driven thing. And while I take other EA arguments seriously, I don't really want you to come in being, oh, because I like Andy on this one thing, I should defer to him on these other things as well. So that being said, the way that EA gets really controversial, really fast, is the general claim that,

Andy Masley 25:46 like one other area people might be missing a lot is that if we're able to reproduce a lot of what happens in the human mind around intelligence in machines specifically, this could actually be really drastically destabilizing for a huge number of ways and could even cause like human extinction specifically. I have my own kind of very agnostic takes on a lot of specific EA arguments about this. Like This is my that idea of P doom around AI of like, is society going to end? Are we going to ruin things? And I know many of us have had those conversations. Yeah. Exactly. And like, gun to my head, my own, like, p doom from AI is probably pretty low. Like, there are a lot of specific EA arguments I take a little bit less seriously. Like, that's a whole separate conversation we probably shouldn't get into here right now. But, basically, the reason why a lot of people are skeptical of me is that EAs in general have been associated with pushing this idea that the main thing that matters for AI is preventing human extinction. And there are a lot of different reasons for that. Some EAs will go so far as like, oh, the human race and intelligence

Andy Masley 26:43 more broadly could just last for such a long time, and there could be so much good in the far future that it's very, very important to prevent us from going extinct, basically. And so, everything else has to kind of be subsumed by that worry about extinction specifically. And so, any other issue related to AI, like, say, water, just doesn't matter compared to extinction specifically.

Andy Masley 27:04 And I do take this criticism of EA pretty seriously, actually. Like, I do worry that it's very, very easy to meme yourself into thinking that, like, what you're doing is much more important than what what everybody else is working on. And I definitely encourage some caution and skepticism there. But at the same time, I I just wanna flag that, like, I'm not one of these people who think that there's an incredibly high chance that AI is going to kill us all anytime soon.

Andy Masley 27:28 And I think I'm approaching this honestly, and I actually just do give a lot of attention to other problems besides AI extinction risk. Like, if you look at my blog, I focus on a lot of other just things that are bothering me day to day personally. I think this comes back to the tribalism or teeny arguments that we hear a lot of these days. And Yeah. Part of this, can talk about the echo chamber of the internet and you know, how we fuel, you know, outrage to drive engagement.

Conor Bronsdon 27:50 Plenty of conversations that we had about that. But- Yeah. Regardless, I think it's really easy to tar someone as an EA, as a tech pro, as a Democrat, as whatever you want to and then attach all these other labels- Yeah. That are nebulously associated with them in a word cloud but may not actually matter to that individual and may not actually be true with that individual and how they think and approach the world.

Conor Bronsdon 28:12 And I, I definitely saw a couple of folks who were arguing that because you are at all attached to AAEA as a movement and like, should just ignore your claims around water and climate. Mhmm. Which I would argue reading your work are quite well researched and and thought A, I wonder how how should we approach this debate around climate impacts of AI and actually have a,

Conor Bronsdon 28:41 I guess a serious conversation where hopefully we can talk about the pros and cons. How do we deflate this, this tribalism that's

Andy Masley 28:50 Yeah. There's a lot to say about that. Two things is that I'm really not asking for any kind of difference here. I I say a lot of the times that I'm like a guy with a sub stack, you know, I'm really not flashing my Yeah, credentials exactly. Yeah. Yeah. Yeah. The dream. So, yeah, I'm not really flashing my credentials anywhere. I'm not saying like, oh, you know, because I have x, or z background, you definitely need to listen to me. Like, I will push a little bit that I was a physics teacher for seven years. I know what a watt hour is. I have a lot of experience explaining this to students and stuff like that. And so that's useful to me. But I'm really not trying to flash any kind of credentials here. I'm just saying, here's my argument. Here are all my sources. You can dig into this and see what you think. And so far, has actually been really surprisingly useful to me. The second thing is that I actually think, for me personally, I've tried to really divorce

Andy Masley 29:35 this specific talking point from EA more broadly. Like, I do think that basic effective altruist intuitions are helping me here quite a bit. Like, it's important to actually dig into the numbers, to understand where the real problems are, and those two intuitions are guiding me a lot in my basic writing on AI and water use. Like, the basic idea is, oh, a lot of people can be getting some simple number stuff wrong. And if you just look into this, you can actually make a pretty compelling case. So in that way, I'm being influenced a lot by effective altruism. But I wanna be clear too

Andy Masley 30:02 that I'm really not trying to push the idea that, oh, if you take me seriously on this, you should also buy my arguments about AI extinction risk or other catastrophic issues with AI or other things related to EA, specifically. So one way of kind of toning down the tribal conversation, I think, is trying to just be like, I'm not really trying to make this a win for my side. Like, those same people who are attacking me were also writing as if I was trying to completely destroy Karen Howe's whole reputation

Andy Masley 30:28 or whatever. And what I'm trying to do is basically just address a really, really big central claim in this one chapter that has a huge amount of influence on the broader conversation. And I'm not really invested in litigating Karen Howe's personal character outside of this. And, I think she's been a pretty great reporter in a lot of other cases. I think she, along with a lot of other reporters, kind of just meme themselves

Andy Masley 30:50 into a bad take on this one specific topic. I'm also trying to get across that I'm not looking to like dunk on anti AI people here or say that, oh look, anyone worried about this is bad and we should ignore that and just Oh, we need to get our facts right. Exactly. And one other thing I had said is that most misinformation often harms your own side. I think like some people had commented and replied to me and said like, oh, effective altruists have also said a lot of, like, really wildly off or really silly or stupid things, and I totally agree with that. Like, there's been a lot of goofy things that that has happened in a space so worried about these kind of weird speculative far off risks specifically.

Andy Masley 31:26 And, like, a lot of people have said things that I really, really don't endorse. Like, some person had pulled out, this one quote that I think got a basic take about climate change wildly off that implied that even 15 degrees of warming, we would basically be okay. I think that's a crazy take. To be clear, I think that's really wildly off. But that actually kinda got my point across in some ways where it's like, yeah, misinformation actually primarily harms your own side, and we should actually be really careful to police this stuff. And I see a lot of allies in people who are pretty worried about AI for other reasons, you know, like people who see it coming for their jobs, who are worried about long term, like labor relations and other things or surveillance and things like that. And I would like these people with very valid concerns about AI to actually just have the facts right and not push things about, oh, and it uses x amount of water, where if you actually just poke at that and discover that's not a problem, that might actually be somewhat invalidating of other claims that they're making. So, yeah, I I am pretty big on just like, we're all very prone to tribal thinking, it's bad, but I actually just think that tribal thinking is gonna be really

Andy Masley 32:24 it's gonna really harm everyone in the debate as all these questions come more and more online. Just like getting the facts right is gonna matter because that's gonna correlate in some ways with whether we're gonna be able to fix the actual, very real problems with AI, even outside of like kind of wacky, far off like extinction risk stuff and stuff like that.

Conor Bronsdon 32:42 And I'll say, I hope listeners of the show know this already, but I personally feel it's important to have these conversations about the details because it is very easy for transformative technologies to cause negative externalities if we don't address them and think them through. And I think we've all seen examples of AI being ineffective or causing harm in some cases. You know, the obvious examples of this are things like the psychosis engagement around chatbots that some people have had, like there are legitimate worries here. There are people who were already struggling, who have been taken out negative paths. We have to, as an industry, be able to talk about how do we make this more responsible? How do we guardrail against us? How do we avoid this? Because AI has this amazing opportunity to improve things for us, but we have to put the work in to have the infrastructure be correct, observe and understand, guardrail and protect against these, these negative challenges.

Conor Bronsdon 33:36 And understanding what it costs to, you know, both from a dollar's perspective, both from a climate perspective, actually build these systems is important as part of that. And as I mentioned at the start of the show, one of the things that I first saw from you was this article using chat GPT is not for the environment. And that is obviously fighting against a really strong current. Can

Conor Bronsdon 33:56 you tell me a bit about this broader discussion piece? Because obviously we've zoomed in a bit on the recent conversation.

Andy Masley 34:02 Yeah. So back in January, I was having the same experience you described in the intro where I was going to a lot of parties and a lot of people almost every time ChatGPT came up, the question that people were really worried about, the thing that would come up over and over is, oh, it uses so much energy or it's so bad for the environment. How could you ever possibly use that? I was finding in conversations that, you know, I didn't consider myself an expert at the time, but I was noticing right away that a lot of the basic points people were making were very strange and basically reflective of a bad way of thinking about climate more broadly. So as an example, people would say, oh, it's using 10 times as much energy as a Google search. And my reaction to this is always,

Andy Masley 34:42 if I had told you a few years ago that I had done a thousand Google searches today, would your first reaction to me be, you must not care about the environment you sing of. No, I'd ask you like, what were you researching? Like, man, you have spent a lot of time on Google today. Yeah. In fact, doing a thousand Google searches would probably be a sign that my emissions were actually way lower that day because just sitting around on your computer, Did you I drive somewhere that day? Yeah. Exactly. Yeah. It uses way less energy than normal. And so, was finding that people didn't have these intuitions about,

Andy Masley 35:11 oh, basically anything you do on the computer is usually gonna carry way less of a climate cost than most other things, just because computing is just so wildly energy efficient compared to most things that we do. And I was finding that I actually just couldn't really get a lot of these basic intuitions through to people at parties where, you know, I would go into physics teacher mode a little bit and be like, oh, this is how much a watt hour is. Here's a comparison to other things. But I think they had that same teamy reaction to me where they're like, oh, this guy's probably just some tech bro. He doesn't really care about the environment, so he's trying to trick me. Because every other

Andy Masley 35:43 article I've read about this, has told me the opposite of what this guy's saying. So he must be lying. So I started to do some deep dives into how this was being covered. The first ever article I read about this was from earth.org. This was written way back in 2023. And even there, there was this really weird framing where the article announced that all of ChatGPC together is now using two times as much energy as a whole person.

Andy Masley 36:06 And there, I was thinking, oh, well, that's not a surprise. This is a really popular app already. It's taken the world by storm. The idea that this is using as much as two people just doesn't that shouldn't be a big gut show for ChatGPT. Obviously, since then, ChatGPT has completely exploded. It's now using on the order of maybe a 100,000 people. I don't have the exact number, obviously, but, you know, that's wildly off. But even there,

Andy Masley 36:30 I was starting to notice that everywhere I went, I was trying to find some article that explained, hey, here's putting this number in context. Just if you are worried about the climate, how should you worry about these incredibly small impacts that you're having? And I wasn't finding that, and so I started to put together my own argument. And I kind of wanted this to just be something that I would keep in the back of my pocket if conversations came up. I didn't actually expect it to get much viewership. If people look back at my blog, you'll notice I've made a few other big arguments. Like, I have a big piece on animal welfare,

Andy Masley 36:58 where I was trying to just be like, you know, I don't expect to get too many readers, but if anybody asks me, I wanna have my case basically in one place. And this full argument with a bunch of different angles and a bunch of different opinions I shared about how to think about this actually really blew up.

Conor Bronsdon 37:12 Got mentioned in a lot of different places, was getting a lot of readers very fast and a lot of people giving me feedback. So, we're definitely gonna link these articles in the show notes because I I think they're great sources to go read and I I don't want to have a, you know, a few minute conversation speak for a a well researched and written article. But if you were to give folks a couple of quick comparisons for when they do have that conversation at a party. Let let's say I'm I'm the one and I I come up to you and I go, Andy, I really think that using ChatGPT as much as you are or using another AI product as much as you are is is bad for the environment. How are you responding to me? Yeah. So there are two different ways to go. I actually usually step back and say,

Andy Masley 37:52 based on everything that we know right now, I try to add up every last way that using a chatbot harms the environment. So it's not just the cost of inference, the in the moment answer that you get to your question. You also have to add in the cost of cooling, maybe dividing the cost of training by the number of prompts the chatbot will receive, the cost of transmitting the data to your computer,

Andy Masley 38:15 and the embodied carbon of actually making the AI chips. And once you add up all these together and you try to get the absolute most costly account of the median chat GPT prompt, my best estimate based on all the numbers we have is that that prompt is probably going to cause about one one hundred and fifty thousandth of your daily emissions. Like, after all of that, this is about what it adds up to. And so

Andy Masley 38:39 if you think about it that way and you think, okay, to increase my daily emissions by 1%, I would need to send about a thousand regular chatbot prompts. Once you start to think about it like that, it actually seems very, very hard to raise your emissions or water use at all using these chatbots. Because if you're sitting all day just sending a thousand prompts and reading all the replies, my best guess based on how long Americans take to read, like, the average amount of text.

Andy Masley 39:08 Altogether, this is probably gonna take you about seven to eight hours or or something. And if you spend eight hours on something that only uses 1% of your emissions, your emissions are actually probably way lower for that day than they would otherwise be. And that's the main thing I'm trying to get across. And I can compare it to other things where like, yeah, using

Andy Masley 39:25 ChatGPT, sending a prompt probably emits as much as printing a fifth of a single page of a book, or it's like running a space heater for about half a second, or driving your car for about four feet or something like that. Those all get across the idea too. But I think in that second case, people are thinking like, oh, even though it's very small, it still adds to my emissions. And so any little bit is still gonna So thinking about the replacement is important. Exactly. Yeah. And so I think just putting this all in the overall scale of your daily emissions and water use, the water is actually even crazier, honestly. Like my best estimate right now is that a ChatGPT prompt is probably about one eight hundred thousandth of your daily water use,

Andy Masley 40:03 mainly because I think most people aren't really factoring in their full water footprint here and just all the different ways water is consumed, not only in their homes, but in the electricity they generate, in the food that they eat, and things like that. And once you add all of that up, it's just really hard to think it would ever be good advice for anyone who's trying to cut their emissions or their water use. Like, hey, you should, you know, cut out your chatbot use. Like, that kind of sounds to me as if I met someone who's really worried about the environment, and I saw them fill a pot of water to boil, you know, to make spaghetti or something. And then all of a sudden, they take this little tiny water dropper, and they're like, oh, I need to save a few of these drops to make sure I'm only using the water I really need,

Andy Masley 40:42 and I'll save these drops for later. Like, in that case, I would say, oh, there are just so many other things you could worry about. This is just the small, really sad distraction compared to, like, you know, building out renewable energy on the grid and stuff like that, and you just shouldn't fill your life with these just tiny, really insignificant changes that aren't actually gonna do any good, but just make you feel really guilty. And so for the same reason, I definitely don't think everyone needs to use chatbots. Like, you don't want to, that's fine. Like, maybe you have other valid criticisms of them. But if you're if you do wanna use them and you're worried about AI and the environment,

Andy Masley 41:15 that specifically I say, you're just getting that wrong. They're not going to add to your personal emissions or water use in any significant way. I'm as confident about this as I can be about most things that I believe in general. And I think the common wisdom on this, including from very high up people who make important decisions, is just completely wrong. Like, I read so much about, you know, universities and businesses that are telling people to be, like, quote unquote, responsible and thoughtful with their chatbot usage to not use it too much or in friv frivolous ways. And, like, for me, this kind of reads like they're telling people not to send emails frivolously

Andy Masley 41:48 or to, like, play games on their computer frivolously or something like that. Like, they're really not factoring in just how small of a climate cost this is compared to almost everything else that we do. And so, I would really like us to move on from what I see is like a really massive misconception that's still really driving the conversation in a lot of different places, both at parties, but also in like big institutions.

Conor Bronsdon 42:08 Yeah. It feels like this idea also of focusing on the demand side, which is so spread out compared to the supply for data center energy is so silly. I would love to have a deep conversation about like, okay, we should be using nuclear, we should be using these other sources of energy instead of, you know, coal or natural gas or oil to power these data centers, or we should be trying to use using hydropower.

Conor Bronsdon 42:33 And you know, honestly love, love to have a deeper grid conversation. That I think there's a legitimate conversation to be had. We should be talking about building baseload anyway. We need to talk about that because there are a variety of externalities from these major industries, but it's not about an individual's usage of these products. And, it just seems like to your point earlier, like we have a limited ability and bandwidth to have these conversations. Let's have them about the actual impacts.

Andy Masley 42:59 Back when I was in college, there was a general sense that focusing on individual small lifestyle changes just isn't really going to add up compared to big massive systematic changes we need to make to the energy grid specifically. Like, I definitely do try to make personal cuts for the climate, so I'm very confident that my personal CO two emissions are significantly lower than American averages,

Andy Masley 43:22 where, you know, I live in a city, I live in an apartment, I don't drive super often, and I've been vegan for ten years, all of these things kind of add up in different ways. But all of this, I think, pales in comparison to just a few other things I've been able to do in my life around, like, maybe trying to keep a nuclear power plant open for a little while longer or advocating for solar power in different places and things like that. Like,

Andy Masley 43:45 people really have no idea sometimes of the scales involved where if you're able to just be a small part of a team of people who get a new solar plant built or who are able to keep a nuclear power plant open, you can have literally hundreds of thousands of times as much impact as I've had over my entire ten years of being vegan, as an example. It's just the numbers are so different that they're comical,

Andy Masley 44:07 honestly. And so when I see people who are worried about the environment, and they're spending even a fraction of their time trying to litigate hundreds of thousands of their daily emissions, and trying to be like, oh, but if I can just get this down to zero, that will be great. I think that's really sad, honestly. They're missing all these huge, powerful ways that they could be influencing the environment instead. And just this hyper focus on policing your individual

Andy Masley 44:30 lifestyle emissions and changes, I think, is mostly a mistake for the climate movement. Like, are just so many huge wins we could be having elsewhere. And this kind of bleeds into the broader data center conversation. So, so far we've mainly been talking about your personal emissions. But yeah, like data centers are definitely at least more of a conversation about now we're talking at the general grid level and what's affecting the grid both on the demand and the supply side.

Conor Bronsdon 44:53 And you brought this up earlier, but there are obvious conversations to be had around efficiency of data centers. You know, the the one with Google was a good example of, look, they've moved from probably a more efficient water cooling to now air cooling of this criticism. And you can look at like Microsoft's new data centers where they're trying to be extremely efficient about water cooling. And there is a variety across these data centers we should be pushing for efficiency. There's a reason there are things like LEED certified buildings that are supposed to be like significantly more,

Conor Bronsdon 45:21 or less climate intensive and much more efficient in energy usage. That's a great use case for business. There's a huge opportunity around that. There there's in fact a whole industry that can be built around making things more, climate efficient and reducing emissions and long term reducing costs. In fact, we're see we're seeing that in many areas, But too much of this conversation around energy usage and AI is getting bogged down in this

Conor Bronsdon 45:43 kind of, almost like disassociated moral argument,

Andy Masley 45:48 instead of in what are the actual policy impacts we can do. Mhmm. Yeah. The disassociated moral argument, I think, is also pretty difficult for people where they'll I think this is one of the first times people have been thinking about, like, a very large new use of energy and water that they personally do not believe is valuable at all. And to be clear, if I thought that AI didn't have any value value at all, I would be objecting more to big data centers. Like, I think AI has a lot of complex trade offs but there is a lot of value.

Andy Masley 46:16 But if I thought it was completely valueless, I would both think that, oh, yeah, we sho- probably shouldn't be building like these massive new data centers that are producing nothing. But I also really wouldn't be litigating the personal use of chatbot thing at all for the sake of the environment because there are all these other things we also do in our everyday lives that are kind of quote unquote useless. Like, if you boil just a little bit too much water or you shower for just one second too long, you're probably gonna use way more water than the average ChatGPT user

Andy Masley 46:44 uses in a given day, on the chatbot. And so even there, I think if I believe this tech were completely useless, litigating the personal chatbot use, I just think is a complete waste of time, and people should at least be thinking more about data centers rather than, like, what people are doing online. In general, computers are just so wildly energy efficient based on what they're trying to do compared to most other things, like delivering information,

Conor Bronsdon 47:06 that I just think that almost everything that has to do with computing on a personal level is just a terrible target for the environmentalist movement. Yeah. There is so much opportunity from digital systems, and I I it is important for us to have these broader conversations. I'll say if there's anyone who's listening who is a, you know, on the data center infrastructure side of things, who wants to come on and have a long conversation about energy side and how to make them more efficient, would love to do that. Getting mired down in these arguments about a few

Conor Bronsdon 47:32 Google searches versus a ChatGPT prompt versus what you all, you're spending your day on. It's just not an effective place for the argument to be. And I hope that we see progress in the coming months from a lot of the activists who I think have gotten distracted, because of their own personal feelings around this. A lot of the AI industry would love to talk about making data centers more efficient.

Conor Bronsdon 47:52 Like, we, we think that we want more data centers. We want to make them more efficient so it costs us less to, to run them so that we can, you know, build better technology and make more of an impact and like, let's have that conversation. I I'm hopeful we kind of start transforming it to a positive one. And Andy, I really appreciate your, your writing and hopefully pushing that in a direction and the conversations you're having. And,

Conor Bronsdon 48:14 to your point about the to, to impact, like, I think it's clear that the work you're doing here, is making one just simply about the conversation that has sparked with Karen. So hopefully we can see the start to turn here in the coming weeks and months. But I wonder if you have any closing thoughts about the future of this conversation or the future of AI and your own perspective on it.

Andy Masley 48:33 Yeah. A lot to say. I think that my basic ask for most people thinking about this is I suspect that the vast majority of AI's actual impact on the climate will not happen because of data centers specifically. It's probably going to happen both for good and bad because of the way AI is actually used in the real world. I kind of see this as the early days of Amazon as an example, where if we were trying to think about how is Amazon going to impact the environment,

Andy Masley 49:00 and my assertion was, oh, I mainly wanna look at how much energy it costs to run the Amazon website in data centers. I think that would be a huge mistake. Right? Because almost everything that we do on computers has really outsized influence on the physical world. You know, it determines, you know, where packages get delivered or new, like, routes for delivering them or, like, how much people buy. Like, I suspect that people are buying maybe a lot more frivolous stuff that they might not actually need because of Amazon and stuff. And these are all never done that. I don't know what you're talking about in Amazon. Of course. Yeah. Me neither.

Andy Masley 49:32 And so I think that for anyone really worried about the climate impacts both positive and negative for AI, I really want them to mostly think about how is using this technology going to impact the environment. And, obviously, again, you know, if I thought that AI were completely useless, I would also be more opposed to the data center build out in general. But I do also want people to be aware that data centers themselves are incredibly energy optimized in general. And so, like, any opposition you have to data centers kind of has to be focused on whether you build them or not in the first place. Like, if you think that AI companies are blowing through huge amounts of energy,

Andy Masley 50:10 that they don't have to be, that they could optimize things much more, I really think you're massively underestimating just how much money there is in optimizing AI chips and optimizing like what happens in the data center themselves. So this is kind of a separate question from whether the data center should exist in the first place. But at least if you're worried about this, you kind of just have to make the decision at that level of should this data center be built or not. Like, you really shouldn't assume that AI companies

Andy Masley 50:35 are just choosing to waste huge amounts of energy and could optimize it much more. Like, that's it becomes kind of comical when you actually read like it feels like this, like, huge, ridiculous engineering achievement that data centers are able to be as efficient as they are. And so, I think, like, for anyone listening to this, again, if you're much more critical of what's happening inside data centers than I am, you should mainly just be thinking about, like, should these be built? Should these not be built? What will there be their outside effects on the grid? Can we make sure that the energy they draw from is greener?

Andy Masley 51:04 But it really shouldn't be focused at all on like any kind of pressuring AI companies to optimize their energy use more just because, from my perspective at least, the winds are just so tiny there. They're highly incentivized to optimize that energy use already. Like, they Yeah. Yeah. And so, yeah, my two general asks for people thinking about this are one, look into whether AI will on net be good or bad for the climate in the way that it's used.

Andy Masley 51:27 I lean much more in the direction that specific deep learning apps will actually probably have much bigger positive impacts on climate than anything that's happened in data centers themselves. This is more tenuous, and it's also not directly connected to chatbots specifically, but I'd really like people to form their own opinions there instead of just hyper focusing so much on the cost of individual computations in data centers themselves. And secondly, yeah, just understanding that data centers are already really wildly efficient. And the weird thing about them is that they concentrate such wildly efficient tasks in one place,

Andy Masley 51:56 rather than the fact that each individual task is burning through too much energy compared to what it could be. So those would be my two, like, general takeaways for people. And obviously, also, like, would appreciate if you subscribed to my blog. I write about all of this in a ton more detail there. I've done a ton of deep dives, so if you're curious about what I'm thinking I highly actually recommend Andy's Substack. Is very and

Andy Masley 52:15 and interesting. Appreciate it. Thank you. And like, if you've had any disagreements, about what I've said here, I am super open to engaging on this, either in like the comments of my blog or over email and stuff like that. So, I'm just very grateful to be like a big part of this debate in general. It's been really fun for for me and like, yeah, very grateful to be here as well, Connor. Thank you for hosting me. My pleasure. Thank you for for coming on and sharing your your insights and your thought process. And I appreciate the equanimity

Conor Bronsdon 52:40 with with which you've approached the conversation. A lot of folks can learn from the approach of like, look, we can criticize a chapter in a book. That doesn't mean we're criticizing everything else. I do want to ask you about this RL impact piece around, reinforcement learning applications. You said you think there are massive potential benefits from that. Just curious, since it's obviously related to your thought process around the climate impacts of AI, what do you see coming there? The basic

Andy Masley 53:02 fundamental thing that deep learning specifically is, is basically a way of getting a computer to pick up on complex subtle patterns and react to them in ways that humans can't necessarily program into them. So, like, as an example, I can give a computer I can type out, like, really clear instructions for how to play games of tic tac toe or something like that. But there are a lot of much more subtle kind of tacit knowledge questions that we ourselves don't know how to program into computers,

Andy Masley 53:30 but that deep learning can kind of get computers to at least basically understand. And so completely separate from the question of chatbots, I think there are just these really wild possibilities for using this specific technology in a huge amount of areas related to AI. Like, if you put a deep learning system in charge of, like, optimizing a building's energy use, it can significantly reduce it because it can can pick up on patterns

Andy Masley 53:53 that would kind of require a human constantly watching to see what happens and stuff. AI can, help us a lot with, like, irrigation, detecting leaks, and stuff like that. I actually suspect that deep learning has already probably saved more global water on that than has been used in data centers. Again, this is much more contentious, so I would maybe have to circle back on that to triple check.

Andy Masley 54:15 In general, the International Energy Agency has a really great report from 2024 where they try to estimate AI's overall climate impacts. And, again, I wanna flag here that this is, like, including all the energy used on chatbots in data centers, and their impacts are mostly coming from non chatbot things, like other deep learning systems. I do think there's a lot of spillover into these other deep learning things just from all the attention that chatbots and AI have been getting and all the money that's flowing into AI more broadly.

Andy Masley 54:43 But even if you think chatbots are completely useless, the IEA specifically estimates that AI more broadly will probably prevent three to four times as many emissions as are used in all data centers, both for AI and the data centers that support the global Internet as well, just because of all these different potential ways of optimizing things. The other thing that I've been really excited about for about ten years now is called smart grid technology where Yes. The energy grid itself,

Andy Masley 55:08 kind of knows where to move and save energy. Like, if you actually look how grids have worked throughout most of the twentieth century, it's actually kind of really insane and stupid where, like, you just have this one really big, power plant that's just kind of constantly on and delivering huge amounts of energy everywhere, and it's kind of thoughtless about where it's actually delivered specifically.

Andy Masley 55:31 And if you're able to deploy a lot of really simple deep learning models or other just basic programming tricks, it doesn't have to be associated with AI at all even. Like, the grid can just kind of understand, oh, if demand fluctuates this much, I should now store this much energy in batteries and stuff like that instead of just trying to send it to this household.

Andy Masley 55:49 And this will actually enable green, tech and green energy build out much more because the main issue with solar power right now is that it only really works when it's sunny, wind power only really works when it's windy, and if we can have a grid that's like, hey, this is a place where I'm getting a lot of supply of power and maybe not too much demand. So I know to move this all here to these batteries for later. And that can just use really simple deep learning. It might not even rely on AI at all. Just like simple algorithms and stuff like that.

Andy Masley 56:17 All of that can just really enable like a much broader green energy build out. So there are a bunch of specific studies on this, like two that I like. Maybe we can link this in like the show notes later, that I would recommend to people. Let's do it. I think that like most people focused on the climate impacts of AI are just way underestimating the potential of these other deep learning systems to just have really wild, outsized climate impacts. Just because honestly, so much of the way we use energy and water right now in general is just incredibly wasteful and really suboptimal.

Conor Bronsdon 56:46 And we could just make that much better if we just deploy these in a few key places. Absolutely. And Andy, thank you so much for this conversation. Hopefully it can prove valuable to many of our listeners who are maybe struggling to articulate their thoughts and rationale for why some of these concerns are overblown, I think it's really important we start shifting the tenor of this conversation. And you've contributed a lot towards that. So so thank you so much for coming on the show and spending your time with us. And and hopefully, if if listeners have enjoyed this and maybe found it valuable, if you know someone who's maybe been a little skeptical, this is a great starting point for a conversation. There is such an opportunity at this industry for, for positive change and we should all be part of shaping that as people who are working in it or interested in it.

Conor Bronsdon 57:27 And we need to make sure that the debate around it is happening in a way that is actually conducive to these these positive movements. So Andy, thank you so much for the efforts you're putting in to ensure that it's happening on a factual basis and that we are actually focusing on the right things. Yeah. This was so much fun, Connor. I really appreciate the conversation. Thank you. Yeah. Thanks for coming on. For anyone who's listening and really enjoyed this episode,

Conor Bronsdon 57:47 let us know who else you think we should have on. We want to do more of these conversations that are more specific to individual areas of AI. And we would also love to hear from you in the ratings and review section of your podcasting app of choice, whether that's Apple Podcasts, where we deeply appreciate five star ratings in particular, but any ratings are great. And any reviews or feedback you want add or in the comments on Spotify,

Conor Bronsdon 58:08 where you can also rate us five stars, I'll note. So, thank you so much for listening. And if you enjoyed the episode, please go ahead and leave that rating or review or just drop a comment, let us know who you want hear from next. Thanks again, Andy. Yeah. Thanks so much, Connor. Thanks to Galileo for sponsoring this episode. Their new 165 page comprehensive

Conor Bronsdon 58:25 guide to mastering multi agent systems is freely available on their website at kaleo.ai and provides you the lens you need to understand when multi agent systems add value versus single agent approaches, how to design them efficiently, and how to build reliable systems that work in production. Download it for free at the link in the show description to discover how to continuously improve your AI agents, identify and avoid common coordination pitfalls,

Conor Bronsdon 58:57 master context engineering for agent collaboration, measure performance with multi agent metrics, and much more.