Can AI Help Fight Climate Breakdown? A Conversation With MIT’s Priya Donti

Can AI be used productively for the greater good? More specifically, can AI be used for the urgent goal of improving the Earth’s climate and reducing carbon dioxide emissions? Might AI help slow down, or even stop, some of the climate catastrophes that are already under way? 

EcoWatch spoke with Priya Donti, the co-founder and chair of Climate Change AI, a non-profit dedicated to exploring the intersection of climate change and machine learning.

Where would you say AI is right now in relation to climate change?

AI is being used in all sorts of ways, from helping us to better forecast wind and solar on power grids, to monitoring greenhouse gas emissions from space, to helping us optimize heating and cooling systems and buildings to be more efficient, and to helping us accelerate the discovery of next-generation batteries. 

I’ll point specifically to the use of AI for climate modeling and weather forecasting. We’ve started to see some powerful approximate models for weather and climate forecasting that are scalable, that run very much more efficiently than their physics-informed counterparts. And I think we’ve also started to see a lot more commercialization or deployment of some of these techniques in the wild. You see a lot more startups that are doing building-energy optimization using AI and machine learning, or insurance companies that are assessing by using AI to understand climate risks.

I would say the general understanding of where AI can be impactfully used for this is still lower than I think it should be.

Right now, is AI having any impact on these older oil or fossil fuel companies that are still extracting oil, which might be argued are the major problem when it comes to climate change? And what about other industries?

I think it was back in 2020 that Greenpeace released their Oil in the Cloud report, which talked about ways that AI was being used to facilitate extraction, exploration, extraction, process optimization, and advertising. And I think their estimate at the time was that AI would generate $425 billion of revenue for the oil and gas industry by 2025. I haven’t seen sort of the retroactive assessment of those numbers, but that’s a big one. 

AI is also used for targeted advertising of on-demand delivery, all these things that really increase the amount of societal consumption without necessarily always making us happier but in ways that certainly increase emissions. AI shapes how we consume information online and so has implications for how we’re consuming climate information or misinformation and the generation of that information as well.

And AI is also a key driver of technologies like autonomous vehicles, which are not often thought of in the context of climate, but depending on how they go, could be good or bad for the transition of the transportation sector.

Would you say it’s still early times in terms of AI impact?

It is, and I think that right now, a lot of the debate is around the fact that it also has its own carbon and hardware footprint. And that’s real as well. That’s something we absolutely should be paying attention to.

Give us a rundown of some of the sectors where AI is most effective in climate change right now.

AI is being used across virtually every single sector in climate action. AI is being used for things like forecasting solar and wind on power grids. There’s some research at this point on how you actually use AI to better optimize the power grid itself to accommodate the fact that lots of variable renewables are coming in.

AI is being used in the buildings and city planning sector, not just for things like energy efficient heating and cooling, but also speeding up simulation models that are showing how wind is flowing through the cities. You can understand the energy efficiency characteristics of the overall city and as a result plan the city overall better. 

In the agricultural sector, there’s a lot of work using AI for large-scale crop yield monitoring and crop-type mapping in order to understand what is being grown where, what are the yields looking like, and how does that shape agricultural policy to better adapt to the effects of climate change. And there are also applications of AI for precision agriculture to try to both increase productivity of agriculture, but also, depending on how it’s deployed, reduce its impacts. 

In the climate science space, there’s a lot of work trying to basically approximate all or parts of climate models using machine learning to help them run faster, which allows you to try out more scenarios and hopefully get a finer-grained understanding of what’s going on. And on the disaster response side of things, the UN satellite centers are already using AI to do real-time flood mapping in order to understand from satellite imagery the extent of flood and how they can best respond real-time.

Is there an example where AI is responsible for reducing emissions at all? 

DeepMind famously had an article come out that said, we’re optimizing Google’s data center using AI, and this is how much energy we were able to save by turning on and off the cooling system. Building heating and cooling optimization is an example I gave in a way that has a direct impact on reducing greenhouse gas emissions.

How about concrete making, a huge emitter of emissions, or other industries in need of climate modernization? 

With concrete, with batteries, with electro fuels, the general thing that happens is that you have people who are trying to figure out how to synthesize this new thing, and they must try to figure out what thing I’m going to synthesize, what experiment am I going to run. It’s going take some time to do that. Where AI and machine learning have been used there is to analyze the outcomes of past experiments and more intelligently suggest which experiments to try next to try to reduce the number of design cycles it takes to create a better battery, or cement, whatever. 

What is the barrier to implementation of a change like that from the point of view of business owners? 

I think it’s this coupling of technologies with business models, right? So even if you have a technology, who is selling it? What’s your pathway to deployment? Is it only a new building that you can do, or is it compatible with retrofitting an old building? I think some of these commercial and deployment-oriented aspects need to be ironed out.

What are some recent success stories with your nonprofit in the climate change space?

Our summer school last year had 10,000 people registered for it. In terms of specific projects, we have, for example, entities working on projects as wide-ranging as working with the government of Fiji directly to improve their flood forecasting, to a future project that is developing better and more localized sensing for the Ghanian power grid to identify places where there are large inefficiencies, which can then be reduced, which means saving emissions there. 

How does AI dovetail with actual legislation to reduce emissions? 

I think the biggest way AI can play a role is facilitating information transparency. A commonly cited example is in the UN climate negotiations. Different countries will submit their own emissions inventories that they have compiled domestically in order to enter the conversation. And, you know, I think many countries are doing this honestly. But of course, they may be limited by certain tools in terms of gathering these inventories faithfully. 

In principle there is an incentive to not truthfully report your emissions because of course that has different implications for how your progress is tracked.

Initiatives like Climate Trace are trying to come up with an independent third-party estimate of greenhouse gas emissions using satellite imagery on-the-ground data in a publicly available and publicly auditable way. And the way you do that at scale is by leveraging AI for large-scale data analysis. I think this information transparency angle is important for policy.

I have a lot of skepticism towards AI and climate change. 

I think a healthy amount of skepticism is warranted. AI is not a silver bullet. It’s not magically going to create this scenario where all of a sudden, it’s like, we’re saved, we don’t have to make hard choices, nothing like that. But there are certain situations where a climate-related workflow is facilitated by being able to analyze data at scale or to have a more precise forecast or to optimize a system more efficiently. AI can play a role in those kinds of solutions. 

AI is a support. The reason I work on AI for climate is I think it’s a really powerful tool. Addressing climate change requires us to leverage all the approaches and tools we have at our disposal in the ways where they’re best suited to just make it all work and move forward quickly. And AI is among that set of tools. I don’t think AI necessarily deserves to be put on a pedestal above other tools. It’s a support alongside everything else. 

People will ask things like: Can I just put all of the world’s data into an AI algorithm and have it spit out what policy I should do? Policies involve value judgments, hard tradeoffs. You’re not just going to get away with putting data into an algorithm. There’s no objective answer to these things.

Priya Donti is an assistant professor and the Silverman (1968) Family Career Development Professor at MIT EECS and LIDS. She is the co-founder and executive director of Climate Change AI.