Tributaries with Subak: A Conversation with Priya Donti
Introducing Tributaries
A fundamental part of Subak’s mission is to shift the status quo by generating public conversation around climate change. When a mass mindset and behaviour change has taken place, the fight against this crisis becomes less theoretical and more concrete. It all starts with how we think. A big part of this new public conversation is not just the magnitude and frequency of these conversations, but who we’re having these conversations with. This is the focus of the Tributaries with Subak interview series.
I’m Miles Ezeilo, a Fellow at Subak. Earlier this month I had the opportunity to speak with Priya Donti, a PhD student at Carnegie Mellon University. Co-founder of the innovative organisation Climate Change AI, Priya Donti is harnessing the power of machine learning and data analytics to solve difficult problems.
Priya Donti and Climate Change AI
Miles: How did you end up in this field of study? Was computer science and climate change always something that you were interested in?
Priya: I knew that I was interested in climate change from a fairly early age. During my first week of high school, my high school biology teacher actually set aside a week or two for a student-led curriculum on sustainability, where we learned about a variety of sustainability related issues, including climate change. The thing that really struck me is that climate change is not just an issue of the literal environment, but it’s also a very human issue. It has huge impacts on the way that humans live. It has a disproportionate impact on the world’s most disadvantaged populations — and as somebody who grew up as part of the Indian diaspora and therefore grew up seeing some of the stark contrasts in wealth in different parts of the world, this really struck me as something I absolutely wanted to work on addressing.
In college, I also got really interested in computer science, so I was trying to figure out a way to put those together. Luckily, I stumbled upon a paper that was called Putting the Smarts in the Smart Grid, written by a group of researchers at the University of Southampton, that talked about how AI and machine learning would be a really critical ingredient in integrating renewable energy into our power grid. So, my academic PhD research has focused on that specific intersection of AI and energy systems ever since. And more broadly, of course, I’ve continued to be interested in the broader intersection of AI and climate change, which is what the Climate Change AI initiative tries to facilitate.
Miles: I talk to a lot of people in the climate space, and they often bring up one-off sessions or week-long topics in school and how that snowballed into the field that they’re in. You never really know what really impacts you until it’s clearly shown itself, that’s such an interesting story. Talk to me about Climate Change AI! What’s the origin of the initiative and how is it going so far. I’d love to hear more about the whole project.
Priya: Climate Change AI, or rather the precursor to Climate Change AI, started in December 2018. One of the co-founders of the initiative, David Rolnick, was chatting with a bunch of researchers at Mila, an AI research institute in Montreal, to try to understand how AI researchers could leverage their skills to address climate change. In parallel to that, one of our other co-founders Lynn Kaack was a PhD student at Carnegie Mellon on the energy and climate policy side and was starting to see large streams of data become available, like satellite imagery that could potentially be useful for climate policy and climate decision making. And then you had someone like me who was working at the intersection of AI and energy from a technical perspective, but sometimes felt a little bit isolated in that work. So these three threads were coming together at a similar time.
In December 2018, David organised an event at NeurIPS, one of the largest machine learning conferences, to try to start some of this conversation within the AI community about how we could apply our work to climate change. That then progressed into writing a paper that we released in June of 2019, called Tackling Climate Change with Machine Learning that detailed the different ways in which AI could be used to both reduce emissions and adapt to the effects of a changing climate. Upon releasing that report, we realised there needs to be much more done in this space both to educate people about what exactly this intersection of AI and climate change means, and also to provide ways to reduce various bottlenecks that are being faced to actually doing impactful work in this area.
Since then, we’ve put together a number of events and resources aimed at education, providing infrastructure, and building community. For instance, we’ve held a series of workshops at the major machine learning conferences to try to build teams and share knowledge about what kinds of work are being done at this intersection of climate change and machine learning. Our recent virtual workshops have drawn something like 2000 participants each from across academia, industry and the public sector. We recently launched a $2 million grants program to fund and provide resources to researchers who are doing impactful work in this area. We are going to be launching a summer school and other educational programs, and also run virtual community-building platforms and networking events. And we’ve also been working with the Global Partnership on AI to write a report for governments about how they can try to alleviate various bottlenecks in the AI for climate change space.
Machine Learning in the Climate Space
Miles: Machine learning is a big element in the missions of a few Subak members. Climate Policy Radar and Open Climate Fix are two examples. For those who may not be familiar with the concept, can you break down machine learning and why it’s so promising in the climate space?
Priya: That’s a great question. In terms of what machine learning is, let’s first talk about artificial intelligence. Artificial intelligence (AI) is an umbrella term that refers to any computational algorithm that can take on a complex task. Usually these are tasks associated with human capabilities in some ways, so things like reasoning or speech or vision. Machine learning refers to a subset of artificial intelligence concerned with analysing large amounts of data in order to infer patterns to make some kind of decision or action.
There are a few themes that emerge for the different ways that AI and machine learning are being used for climate action. One of these is distilling decision-relevant information. There are lots of places in which additional information could help make a better decision on climate policy, or help better operate a climate-relevant system. For instance, Open Climate Fix is working on trying to map the locations of solar panels in order to inform energy forecasting. Other entities are using satellite imagery to pinpoint deforestation. Climate Policy Radar is working on trying to derive insights from large amounts of policy documents to better understand what worked and what didn’t. The theme here is that we can take large, raw streams of data like satellite imagery and text documents that are hard to analyse manually, and extract certain kinds of insights that can fill critical gaps in information.
Other themes include forecasting of climate-relevant quantities like solar power, crop yields, or extreme events, or optimising real-world systems like electric power systems or supply chains. A last big theme is accelerating scientific experimentation. If you’re trying to develop a better battery and you have prior experiments that you’ve tried, machine learning can help analyse how those prior experiments went in order to then figure out what design it makes the most sense to try next. The kinds of ways machine learning can be used are very diverse.
Open Data
Miles: I want to speak on open data. The idea of data being open so anyone will be able to access it and with their expertise and experience, can help climate action in whatever way they see fit. What is Climate Change AI’s experience with the procurement of data? And what in your opinion is the importance of open data and data accessibility towards climate change?
Priya: Data is obviously a huge, huge part of this particular space. At Climate Change AI, we’ve put together a dataset wishlist keyed to the applications in our Tackling Climate Change with Machine Learning paper, which represents datasets that either don’t already exist, or exist in a form that isn’t necessarily usable by a large variety of stakeholders. For instance, maybe the data that is needed for a particular problem is scattered and distributed among different sources. Maybe it’s incomplete or not generally maintained. Maybe it exists, but it’s proprietary. Or maybe it needs to be collected in the first place. We’ve made that wishlist publicly available on our website, so that entities who are explicitly concerned with the creation or collation of datasets in this space can use that as a starting point to identify the relevant datasets that might facilitate a particular area of work. Another thing that I think is really, really important to think about is that many datasets and the capacity for dataset collection are often centered in the United States or other places in the Global North. It’s really important to expand the representativeness of various datasets or simulation environments to make sure that we’re not just working on problems that are applicable to one part of the world.
Closing Remarks: Collaboration, Advice, and Community
Miles: Thank you so much. We’re going to switch into a bit of a lightning round. Are there any mantras or sayings that you go by?
Priya: Collaboration is key. There’s no one entity or person in this sphere that is going to be able to do this alone. Thinking about the ways in which different organisations can boost each other up and bolster each other’s work is going to be really critical to tackling this problem.
Miles: What are you reading or listening to right now?
Priya: Merchants of Doubt by Erik M. Conway and Naomi Oreskes, which goes into this notion of how the climate denial industry took not just playbooks but also literally people from the tobacco industry, and how that played out. It’s simultaneously depressing but also really important to understand what kinds of techniques are used to sow disinformation in this area in order to understand how to contend with those.
Miles: What advice would you give to a young person who is thinking about entering the computer science climate change intersection right now?
Priya: Be sure to be learning from different fields and perspectives and people and geographies as you go through this process. The greater the extent to which you can widen your horizons, the better you’ll sort of be at navigating the space and figuring out what needs to be done.
It’s also very important to know that computer science in the climate change sphere is not a silver bullet, which is to say that we need policymakers, we need people who are trained in the humanities, we need people who are trained in technical areas that are not computer science, and so forth in order to make progress on tackling climate change. So if you’re somebody who is wanting to work on climate change, there are a variety of ways to get involved.
Miles: Finally, as Subak is all about the community, shout out 3 people who support you in any way. Could be family, someone on your team, someone you just met, anybody.
Priya: I definitely have to give a shout out to my high school biology teacher, Mr. Furnari, who made the space to teach us about sustainability, which launched my interest in this area.
I am going to cheat and clump my research mentors from undergrad and graduate school into one category. So these are my undergrad research advisor Jim Boerkoel, and my current advisors Zico Kolter and Inês Azevedo, who have mentored me very directly on research and career advice, and helped me find the directions that I’m pursuing today. I also want to make a shout-out to Carla Gomes, who started the computational sustainability movement and showed me that it was possible in practice to to merge computer science, sustainability, and climate.
And I obviously have to give a huge shout out to the rest of the Climate Change AI team, notably my co-founders, David Rolnick and Lynn Kaack, but also the broader team. Often when it comes to the story around a particular initiative or project, it’s often tempting to put forth a hero narrative, that it was one person who “solved the problem” or launched a whole thing. But it takes a village, consistently. I’m just so grateful to the entire Climate Change AI team for everything that we have been able to accomplish together.