The Limits of AI in Solving Environmental Problems

This is the statement of purpose that I used in my application to Stanford. I’m sharing it because it describes what motivated me to leave industry and go for a PhD, and because I hope that as I meet more people I can get some feedback on this new direction.

Solutions to climate resilience will depend on identifying small-scale successes and scaling them up. By building pest reporting systems for regulatory compliance in Norwegian Aquaculture and applying computational tools to climate policy research under Dr. Angel Hsu, I became optimistic that data science can scale solutions to environmental problems by empowering people. However, I noticed that in practice these solutions often depended on strong institutions and expensive data collection. To build systems that empower local environmental leaders in the global south, we need innovations designed to serve these leaders by overcoming data gaps and governance challenges. In light of this, I am applying to Stanford EIPER because I want to learn from the faculty’s successful experience working at the intersection of Applied Economics and Computational Ecology, initiating partnerships across industry, government, and civil society, and building data tools that empower local leaders to amplify their impact in agriculture and environmental policy.

I had my first experience building environmental AI as a data scientist at Aquabyte, a precision aquaculture company analyzing data from underwater cameras to help Norwegian salmon farmers save feed and manage pests. I acted as tech lead for our automated sea lice counting product, building datasets and training computer vision and time series models that take millions of underwater images to detect sea lice and determine daily average counts. Earlier, farmers spent a half day every week taking fish out of the water to count sea lice manually. Now, the system I helped develop gave farmers a dashboard of real-time highly reliable sealice data. Despite early field success, the government initially rejected our methodology for automating legally required reports of lice counts. But through diving into the ecology literature and collaborating with Dr. Ragnhild Valen, Aquabyte’s lead biologist, I reformulated our lice counting methodology by slightly modifying state-of-the-art population models. The subsequent documentation we wrote together was the first and only system ever approved by the Norway Food Safety Authority to replace manual sea lice counting. Built on computer vision algorithms I built, the automated lice counting system improved fish welfare through eliminating handling and increased the sample size of fish 40X, enabling empirically better prediction of sea lice outbreaks. We elevated AI’s visibility in Norwegian aquaculture, and Aquabyte’s footprint grew from 5 farms to over 300 while I worked there. More importantly, by simultaneously eliminating fraudulent lice underreporting while reducing compliance costs for farmers, our data science-enabled policy change increased trust and collaboration between industry and the government. This experience made me passionate about how data science systems grounded in institutional partnerships can increase farmer welfare while improving sustainability.

Inspired by the impact of local policy change on environmental problems, I spearheaded a project doing environmental policy analysis under the mentorship of Dr. Angel Hsu, head of the Data-Driven Envirolab then at Yale-NUS college. Federal inaction has led local and state actors to take the lead on climate action plans, but scaling manual policy analysis for thousands of local policies and targets was an intractable problem. Informed by listening to Dr. Hsu about the importance of ambitious emissions reduction targets, I proposed that using text regression techniques could tell us what factors are associated with more aggressive climate plans. To do this, we built a dataset of city climate action plan texts and expert-derived labels of net zero target ambition to model predictors of aggressive climate plans. By interpreting model coefficients, we learned that terms related to government action, quantifiable metrics, community empowerment, and heating emissions were significantly associated with more aggressive climate targets. Not only did these findings align with much of the qualitative political economy literature on climate policy; they highlighted an emerging vision for data-driven inclusive governance among local leaders. Motivated by this finding, I became interested in learning how policymakers adopt learnings from local innovations in environmental policy to scale initiatives that drive global change.

Despite some successes, these experiences highlighted to me the limitations of data innovations alone. The AI in aquaculture policy was only possible in Norway because of cross-institutional partnerships: regulators trusted research institutes who reviewed our methodology, and we could build on public data infrastructure that farmers used to report aquaculture data. While our data helped improve sea lice management, the root problem was an industrial monoculture vulnerable to pests. Truly sustainable aquaculture would use diverse crops that simultaneously boost pest resilience and income like China’s multi-trophic seaweed farms or Vietnam’s integrated mangrove shrimp farms. Similarly, findings from our computational policy analysis project were skewed by the fact that the vast majority of climate plans in our dataset came from cities in rich countries, as most cities in the global south don’t publish a climate plan. However, Shenzhen’s sponge city strategy for reducing flood risk and the world’s largest organic waste management system in the East Kolkata Wetlands show that Asian cities also have innovative local solutions that should be studied. Though data science can enable trusted measurements and inform effective policy, solutions to fighting climate inequality in the global south primarily depend on empowering people. People are needed to build strong institutions, to experiment with techniques in sustainable agriculture, and to lead communities toward solutions.

By building data tools to empower local innovators in under-resourced contexts, we can use technology to scale adaptive solutions to climate resilience across the global south that overcome governance and resource constraints. Because the impacts of climate change are felt most by populations least responsible and most vulnerable, sustainable development requires serious efforts to support climate resilience in the global south. At the same time, governance challenges and resource constraints mean that approaches to climate resilience must be different from those in the West. In contexts where growing populations and incomes are driving conversion to agriculture and urbanized areas, we need to promote models that enable humans and nature to grow symbiotically in small amounts of land. Indigenous innovations in aquaculture, agroforestry, and ecosystem management in India, China, and Southeast Asia are driving diverse approaches to sustainable intensification of agriculture and nature-based solutions. I am interested in using data to scale up these sorts of innovations because they are often robust to local governance challenges and resource constraints. By using remote sensing data, we can understand how policy may address economic or social bottlenecks to local solutions for green growth, enable rigorous validation of and payments for ecosystem services like carbon sequestration and water quality, and build the economic datasets needed to improve access to finance for farms. I want to apply my data science skills to build the science, finance, and policy frameworks needed to empower local innovators to scale up resilient solutions to land and water management in the global south.

My curiosity has led me to pursue experiences that teach me about how social science, computational science, ecology, and Asian studies can be combined to solve environmental problems. Attending the EIPER PhD would be the perfect next step in deepening my understanding of these disciplines and using that understanding to enable locally contextualized solutions to climate adaptation problems in Asia to scale. In particular, I am attracted to EIPER’s emphasis on applying novel computational methodology to problem-driven research as well as the focus many faculty have on issues relevant to sustainable development in the global south. I also believe attending EIPER is unique in that its faculty are not only able to give academic training, but they also have the experience and platform to help me build partnerships with locally contextualized organizations across academia, industry, the private sector, and civil society. After my PhD, I hope to build on the datasets and partnerships I have developed to construct data-driven policy and finance mechanisms that enable the public and private sector to invest in scaling local natural capital, climate resilience, and sustainable agriculture projects.