We have a power grid problem. Can AI fix it?

0:00When it comes to AI, we spend a lot of time talking about what it takes to move models from research into reality. Enter LTX2 from Lightrix. It's an open-source audio-video foundation model built for synchronized sound and video, native 4K output, expressive motion, and precise multimodal control, all running on consumer GPUs. With over 3 million downloads on Hugging Face, LTX2 gives you full model weights, training frameworks, and evaluation tools to build real production workflows. Don't settle for a first draft. Get the complete creative engine to reach your idea's full potential. Try LTX2 today at ltx.io slash model.

0:48Hi everyone, it's Rana. The future of AI is not built in isolation. It is shaped in rooms where ambitious people challenge each other. Masters of Scale Summit is back October 20th through 22nd in San Francisco, bringing together founders and innovators pushing the edge of what is possible. If you're building what comes next, you should be in this room. Apply now at mastersofscale.com slash pioneers. That's mastersofscale.com slash pioneers. On Pioneers of AI, we often talk about the responsibility that comes with building powerful technology. AI isn't just one thing. It's a force reshaping medicine, education, science, and the stories we tell on screen.

1:36And for that future to work, it needs infrastructure that's built thoughtfully and deliberately. As the essential cloud for AI, CoreWeave empowers pioneers at leading AI labs and enterprises with a purpose-built cloud made specifically to train and run the world's most advanced models. They help accelerate the breakthroughs that matter at scale. The big ideas, the moonshots, the futures we haven't imagined yet. Ready for anything, ready for AI. To learn more about how CoreWeave powers the world's best AI, go to coreweave.com slash ready for anything.

2:17I think AI can definitely be like a really useful tool to help address climate change. It has all sorts of uses. Like better forecasting renewable energy on the grid, like optimizing heating and cooling systems to be more efficient. But I would say that in all of these cases, AI is playing kind of a support role.

2:41What is the biggest misconception that people come into these programs with? I think that there's a big set of people who think that AI is this thing that is going to be developed by a certain set of people. And then like the rest of the world will like procure it and use it. They're not shaping it. They're just consumers of it. When I say democratization, what I mean is that more people are equipped with the skills, tools and expertise to actually shape what AI and the AI ecosystem look like. Totally. I've heard some people and specifically actually Reid Hoffman, who's an investor and entrepreneur and a good friend of mine in the show.

3:21He describes AI as the platinum bullet to solve climate change. Do you agree? No. My guest today is Priya Danti. She's assistant professor at MIT's EECS and LIDS. EECS is electrical engineering and computer science, and LIDS is MIT's lab for information and decision systems. She's also the co-founder and chair of Climate Change AI, a nonprofit tackling climate change using AI and machine learning. Priya was also named by Time magazine as one of the 100 most influential people in AI for 2025. I'm so excited for my conversation with her, so let's jump in. I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

4:18Priya, welcome to Pioneers of AI. Thanks for having me on. I'm so excited to have you on the show, and I'm especially excited that we're doing this in person. That's always the best. So I want to start at the beginning of your journey. You grew up in the United States. That's right. I grew up in Massachusetts. Oh, where? North Andover. Oh, how cool. Yeah, that's great. But as a kid, you would go back and visit family in India. How did these experiences kind of shape the work you do today? Yeah, so I think kind of growing up in the U.S. and then going to India, you both see kind of disparity in sort of, you know, resources and wealth kind of between countries.

4:56and then also in both cases kind of within countries. And in India, sometimes that can feel particularly stark where in a particular block, you have like this really grand building and then kind of people begging on the street kind of in the same block. And so I think I just grew up with this really kind of, you know, inherent kind of sense that like, you know, equity and inequity are just these huge challenges in our society and I really wanted to address them. And so that is actually what kind of inspired my route into working on climate change, because I was lucky to get to take a high school class where the first week of our biology class was kind of turned into a sustainability curriculum.

5:38And it was just emphasized there, like the extent to which climate is not just sort of an environment issue, but it's a people issue and one that kind of has disproportionate effects on those who are already affected by other kind of inequities in society. And so sitting there as a high schooler, I was like, wait, equity is already really bad. And then this thing, climate, is going to make it even worse. Like, oh, my gosh, I need to do something about that. And so that kind of inspired my route into working on climate change. That's amazing. I can really relate to it. So I'm originally from Egypt, and we spend a lot of time going back and forth between the U.S.

6:16and Cairo. And like you, I think there is this stark contrast inequality, not just inter-countries, but intra-countries. And we definitely, like my kids experience that, you know, we go back every summer and that's always something we talk about. Can you unpack for us this kind of disproportionate impact of climate effects on certain communities and what that inequity looks like? Yeah. So when we look at kind of the effects of climate change and kind of the extent to which we'll see kind of more droughts, more flooding, et cetera, et cetera, we tend to see that, first of all, kind of if you think about economies and livelihoods that depend on the weather being kind of what you want it to be, agriculture, right?

7:01And there is definitely a big relationship between, for example, like countries that are agriculturally dominant in terms of their GDP and kind of developing countries. Right. So already right there, it's kind of the effect of like the weather changing obviously impacts kind of agriculturally dominant countries. In addition, you have kind of people who live in island states and in coastal regions. They are going to be the ones who experience kind of the biggest effects of sea level rise and flooding. And then even if you sort of have two different communities that in some sense experience the same change in weather, the same climate hazard, their kind of existing wealth and infrastructure impacts how well positioned they are to actually kind of deal with that change.

7:44And obviously, like countries that are less resources or locations that are less resourced, they generally have kind of less infrastructure to actually equip them to deal with that change. So kind of in all of these different ways, both sort of the climate hazards being unfortunately distributed, as well as people's ability to adapt being kind of related to the wealth that they have historically had and the infrastructure they've historically had. That's kind of where this disproportionate impacts aspect is driven from. Yeah. I want to zoom out and talk about AI's energy use and how it impacts the planet.

8:15We had Dr. Sasha Liccioni from Hugging Face. Do you know her? Yes, yeah. We had her on the show. She was awesome. She leads all the climate efforts at Hugging Face. And she talked about how hard it is to actually get a sense of the energy use of different models because these AI companies are not disclosing that information. and kind of she's on a mission to actually build climate benchmarks or energy use benchmarks. From where you sit, how bad is AI really in terms of impacts on energy use, water consumption? Yeah. Where do you think we're at? Yeah. And so kind of stepping back, one thing I also want to mention is that I think when it comes to like AI, right, like a lot of people in the general public and in the kind of public discourse have this notion of AI as like this one specific thing.

9:05So they're like, it's chat GPT. It's like large generative models, et cetera. AI in reality, as kind of, I imagine the audience of this show knows, right? Is sort of a diversity of technologies and a diversity of paradigms. And so I think that kind of then when we talk about this public discourse of like, a question I very often get asked is right, like, is AI's energy use gonna be made up for by the fact that you can use AI in the electricity sector? But in reality, the type of AI on both sides of that equation is not it's usually not the same type of AI. And so basically, I think that at a macro level, I am worried about the extent to which kind of the push for like a specific type of large scale technology and like kind of a massive investment in sort of data center installation and like all of the energy and water and materials impacts of that.

9:57I'm worried about that. And I'm also sort of annoyed by the fact that sometimes the narrative that's bolstering that is if we just develop this type of AI, then it'll have benefits for health and climate and such. When sort of the types of innovation and deployment of AI that are necessary to make an impact on climate and health are not largely kind of scaling this particular paradigm of AI. Yeah. Yeah. So basically there are different paradigms of AI that can help us get there, both in terms of fixing the problem, right, and kind of addressing the underlying also causes of the issue. Absolutely.

10:34Yeah. Yeah. I kind of actually want to dig into this more because a lot of your work is rooted in machine learning, but it's also not necessarily these large language models. My understanding is you do a lot of work with small models, smaller models. But what else? Like what are the different approaches that you use in AI that perhaps don't have the same climate impacts? Yeah. So one kind of approach that we use, I'll call it like engineering informed AI, where we basically kind of think about in power grids, which is the area I work in, we have some knowledge of kind of the physics of how this power grid works.

11:14we have some existing knowledge of kind of the engineering constraints associated with like what you can tell equipment to do or not do. And the challenge we're often trying to solve there is if you just write down the physics and engineering constraints and try to sort of solve out like what should I do on my power grid, that tends to be kind of in our modern paradigm of power grids too computationally expensive to do. You have to solve these problems of like what do I do much more quickly and at much more scale because you have more like time varying renewables and distributed devices and all that kind of thing.

11:48And so in that setting, what we try to do is develop AI techniques that can actually help us to optimize the grid faster and more effectively, but while still keeping some of the kind of physics guarantees that we need to make sure we're not breaking the power grid. So there the paradigm when I say engineering informed AI and machine learning is we actually think pretty deeply about where are the opportunities to actually embed like the literal physical equations or kind of robust control guarantees or various kind of things that we we know how to grasp and prove into the way we design our neural network architectures in the first place and that way you kind of avoid also let's say like you you both need to do this if you want your algorithm to satisfy a particular constraint and not hallucinate Right.

12:35Like make up stuff that is not, that can't be applied to the real world. Exactly. Like for a power grid, if you kind of wanted to do the equivalent of not hallucinate, like some of these constraints that it needs to satisfy, it is probability zero that like a machine learning model will kind of land on that particular constraint. So you need to make sure that happens. And then also in doing that, you kind of avoid like wastefully relearning stuff from scratch from data because you already have that knowledge embedded in the model. So you can sort of focus your training cycles and data size and data curation efforts on like what the model actually needs to learn.

13:07Yeah, which also kind of underscores you do not need a generative, like a large language model that can answer what can you have for dinner tonight to solve this problem, right? Like you can train it to focus on this specific problem. I think there's an approach today where the solution to everything is this large generative general model, but that's not necessarily the case. Yeah. So, I mean, I think there's this paradigm of like, do you create like a task specific model that's like for one specific thing? Or do you try to create a quote unquote general purpose model that does everything? And I think in reality, like the true answer is somewhere in between, but closer to the task specific side.

13:46in the sense that even on the power grid, there are different tasks that because they're sharing the same physical like grid and same underlying physical structure, there are clearly some tasks that probably can benefit from sort of sharing kind of, you know, you share data between these tasks, you share like the learning of structure between these tasks and that can enhance learning overall. But I think right now the approach is a little bit like we're gonna start from this paradigm of general purpose and like see what it can apply to. And I think we actually need to go the other way. It's more like you start with your specific task.

14:18You try to understand if there are other tasks that could benefit from sharing knowledge and you kind of build up. And I think the sort of clusters we get are kind of closer to that, like some smaller set of tasks than like this kind of everything all at once with one model. Very cool. OK, so there's a lot of talk about the infrastructure needed to support the increased demand for AI, whether that's building more data centers in the U.S. or transitioning to nuclear power more widely. What kind of infrastructure do you think we will need to sustain this demand for AI? Yeah. So, I mean, I think that kind of generally like with kind of clean powering of data centers, yeah, you clearly need your grid to be clean.

14:58And so that means kind of, yeah, putting together like, you know, solar, wind, nuclear, like it becomes really necessary to do this. But then, of course, that doesn't take care of kind of like the use of water and water stressed regions. So there, I'm less of an expert in this, but like citing data centers in a place that's less water stressed, but also like alternative cooling technologies, etc. So I think there are all these kinds of approaches. But I think fundamentally, like when we look at kind of global decarbonization pathways, these pathways rely not just on sort of like cleaning up electricity, but also kind of being serious about energy efficiency and like not wasting energy.

15:40And so I think it's both about kind of for sure, right, like for the data center growth that we as a society choose to accommodate, we should make sure that's clean. But also, like when I say choose to accommodate, we should also think about this holistically in terms of like that renewable energy we're putting out there. Should that be for data centers? Are there other types of loads and needs that we actually need to make sure we're prioritizing and powering? And what does that mean for kind of planning the overall efficiency of the energy sector? In a minute, how AI can be part of the solution to climate change, and why Priya disagrees with LinkedIn co-founder and investor Reid Hoffman.

16:20That's after a break.

16:33When it comes to AI, we spend a lot of time talking about what it takes to move models from research into reality. Enter LTX2 from Lightrix. It's an open-source audio-video foundation model built for synchronized sound and video, native 4K output, expressive motion, and precise multimodal control, all running on consumer GPUs. With over 3 million downloads on Hugging Face, LTX2 gives you full model weights, training frameworks, and evaluation tools to build real production workflows. Don't settle for a first draft. Get the complete creative engine to reach your idea's full potential. Try LTX2 today at ltx.io slash model.

17:21AI has moved from possibility to practice. The question now is how do today's pioneers build it fast, reliably, and at scale? From medical research and education to scientific discovery and cinematic creativity, AI is already changing how we understand the world. As the essential cloud for AI, CoreWeave is a leader of that transformation. Delivering a purpose-built AI cloud designed specifically for today's most complex AI workloads, it allows researchers, developers, and creators at leading AI labs and enterprises to focus on impact, not limitations, which is how bold ideas become real breakthroughs.

18:04CoreWeave was built for this. Ready for anything. Ready for AI. To learn more about how CoreWeave powers the world's best AI, go to coreweave.com slash ready for anything. If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's teamwork collection. It has planning in Jira, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build.

18:51Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger.

19:06All right. So let's talk about how AI can be part of the solution. I've heard some people, and specifically actually Reid Hoffman, who's an investor and entrepreneur and a good friend of mine in the show. He describes AI as the platinum bullet to solve climate change. Do you agree? No. Okay, great. Tell us why. So I think AI can definitely be like a really useful tool to help address climate change. and has all sorts of uses, right? Like better forecasting renewable energy on the grid, like better allowing us to like monitor crop yields to understand like agricultural adaptation, like optimizing heating and cooling systems to be more efficient, like accelerated science.

19:46I think there are lots of places AI can play a role. But I would say that in all of these cases, AI is playing kind of a support role, like an important support role in some of those cases. And in some cases, like optimizing next generation power grids, I can't think how we would do it without AI. But it's sort of like still part of a broader system and a broader solution. And kind of AI itself is not the thing solving everything. I mean, even to give a very simple example, like many applications that are like forecasting or monitoring, like I mentioned, like forecasting solar power or monitoring crop yields, that's information, but that information needs to help someone make a decision.

20:26And so like who that decision maker is and what our decision making processes are, if you don't also think through that, then you don't actually have a solution. This is just an algorithm lying somewhere. So yeah, AI is super, super helpful, not a silver bullet. Yeah. Okay. I love that. Okay. So I want to kind of tie that to your work then. So a few years ago, you started Climate Change AI, actually back in 2019. So that was before the whole generative AI moment. Why did you start this organization? And I also love your take on how How has the mission changed kind of pre and post, you know, November 2022?

21:00Yeah, totally. So I would say that when we started Climate Change AI, that was at a moment in time when people were not putting AI and climate change in the same sentence. And yet where we definitely saw that there was a lot of potential for AI to kind of serve what is one of the most challenging problems we have to solve, like to address climate change. And so we wanted to bring more awareness to kind of the topic of like, where is it that AI can play an impactful role? But also, what does it mean to actually do this work in a way that's impactful? And that means, for example, not kind of going in with sort of a hammer looking for a nail.

21:38It means really kind of co-scoping solutions like among technologists and kind of domain experts and kind of on the ground users. It means responsible and ethical AI considerations. generations. So we were trying to both kind of raise awareness of like, where is it that AI can play a role and also provide guidance about like how to actually go about doing this. And so we started by writing a paper called Tackling Climate Change with Machine Learning, which was the brainchild of my co-founder, David Rolnick, because he was actually trying to kind of coming from the AI side, get into the climate side and was like, well, if I'm going to do the homework to figure out like what I should be up to, like, why not kind of share that homework with everyone?

22:14And so he kind of got together a group of people, including myself, to write this Tackling Climate Change with Machine Learning paper. I love this paper, by the way, because there's this table in it where you outline all the different ways AI can be applied to a climate. And I think it's very, A, it's a great survey of kind of what's happening in this space, but also in a way it's quite actionable, which is very cool. I appreciate that. Thanks. And yeah, and exactly. And so we put that out there and we ran a workshop alongside it. And, you know, there were, you know, hundreds of people at that very first workshop, like lines out the door, but also lots of questions about, okay, you've told me, you've intrigued me, you've told me this area is important, but like, what do I do?

22:54How do I get involved? How do I find collaborators? How do I fund funding? And so that's what kind of inspired the nonprofit. And so since then we've run, continued to run this workshop series so people can exchange knowledge. We've run kind of a summer school program to help kind of like democratize education on AI and climate and help more people kind of upskill and gain knowledge on the students. Or high school students or college students? So the program tends to be geared towards people who are like, let's say like post-college, but like kind of spanning and late college as well, but kind of like kind of early professional, like late grad student and kind of all the way up.

23:29We definitely have like senior people also kind of participating in these programs. And so, yeah, and then also like grants programs. So basically trying to kind of provide ways for more people to participate and to kind of actualize solutions on the ground. And so then to your question of like, how has the mission kind of changed over time? I would say that at the beginning, because honestly, like many fewer people also in the general public even knew what AI was. Often in these settings, we were talking to like an audience that was like, you have your sort of initial niche audience that sort of is more in the know, that already knows like some of the technologies and values around this, or you are their first entry point to that.

24:09So like if we're talking to a climate person, it's like, we have the opportunity to be like, okay, this is what AI is. This is what responsible AI is. Like this is how to think about AI in the context of climate. Now, of course, you have many more people who are gaining many more entry points into this topic. Their entry point might be chat GPT, so they might come in with a specific or narrow topic. And then also, whereas we started in a realm where AI and climate weren't used in the same sentence, now it's used in the same sentence all the time for better and for worse. And so there's an extent to which some aspects of our mission, I think, have stayed very similar.

24:43It's still about kind of ensuring that people have the tools and skill sets and community and resources to like do AI for climate in an impactful and responsible way. But because the broader context has changed, we have to also be much more responsive to kind of conceptions and misconceptions that people might come in with, as well as ways that maybe people try to, let's say, co-opt this message that AI can be useful for climate and use it to maybe like push forward other agendas. What is the biggest misconception that people come into these programs with? Yeah, so I would say the kind of one of the ones I started with, which is that like AI is one thing.

25:17So that when we're saying like AI for climate, it's like LLMs for climate. So that's definitely one major misconception. And yeah, I think this sort of, I think many more people are coming in with this like AI is a silver bullet like aspect. And again, AI is super useful, but I think kind of helping people to situate like when and where is it useful. And then I think another one that maybe is more like subtle, but comes from this like AI is one thing. I think that there's a big set of people who think that AI is this thing that is going to be developed by a certain set of people. And then like the rest of the world will like procure it and use it.