esiil innovation summit

Earlier this year, I was invited to the 2026 iteration of the ESIIL Innovation Summit. ESIIL (the Environmental Data Science Innovation & Impact Lab) is an NSF data synthesis center that brings together groups of environmental data scientists to spread knowledge of emergent technologies and build working groups that use science-based solutions to solve environmental challenges.

This year’s summit was themed “AI for Sustainability: Translating Environmental Data into Decisions.” Sustainable AI is a topic I care deeply about, as I’ve often struggled to morally balance the powerful benefits of AI and Machine Learning in research with the environmental repercussions. The summit also provided a great opportunity to build connections with other researchers in the beautiful city of Boulder, and as early-career scientists who hadn’t seen mountains in a few months, my labmates and I jumped at the chance.

The summit was organized as an “un-conference”, with lecture sessions replaced by workshops and time for teamwork building and sharing ideas. The summit had 4 focus areas with relevant workshops to attend:

1. Co-producing digital twins for environmental futures
2. Building Earth embeddings
3. AI and causal inference
4. Identifying best practices for using large language models in environmental data science

Throughout these sessions we were meant to envision how these techniques could be used in our own work, and be applied to future projects and collaborations. I was instantly struck by the diverse uses of ML and AI in the environmental sciences, and envisioned several future side projects using digital twins and Earth embeddings.

After the workshops, we worked through a chaotic but energizing process to form teams for the rest of the summit. Around 25 sheets of paper were hung on the walls with various topics, and we had to aggregate around the ones that interested us most. As a lot of my work focuses on extracting and harmonizing data from literature, I gathered around the poster talking about agentic data harmonization.

I found some like-minded researchers and formed a team called Data Buffs. We settled on developing a generic workflow that researchers could use to assist in finding data for meta-analyses using an AI agent. It works by having the user specify a topic with a geographic and temporal scope. The agent then looks through open source journals for relevant publications with data, and extracts tables from those papers, harmonizing them into a single usable dataframe.

Some of the pipe cleaner art we made during the summit.

Above: A sketch of our agentic workflow for data harmonization.

Left: The Data Buffs team, featuring a cartoon by Allison Horst.

I got to spend an extra day in Boulder following the summit, and chose to spend it hiking through Chautauqua park on the East side of the city. Through the roughly 6 mile hike, I found three lifer birds (Broad-tailed Hummingbird, MacGillivray’s Warbler and Virginia’s Warbler) and saw many common western birds I had missed seeing since moving to Michigan. There were also a variety of beautiful alpine southwest plants, like the Ponderosa Pine and Prairie Prickly Pear, and great views of the plains below.

As I hiked through the park, for the first time that week staring at the actual Earth instead of a screen, I had a chance to reflect on the tension I’d been sitting with all summit: the gap between AI’s potential to help us understand the planet and the environmental cost of developing and deploying it in the first place. Surrounded by pines and a backdrop that even the most complex models would struggle to replicate, it was hard not to feel that distance. But the summit also reminded me that many of the people doing this work are aware of the cost, and are actively trying to build AI more thoughtfully, through better data practices and smarter collaboration.