David Dao is a machine learning researcher working on AI for science and the commons, with a focus on nature and climate. He is co-founder and Chief Scientist of GainForest.Earth and co-leads Economies & Governance at Protocol Labs R&D. He holds a PhD in computer science from ETH Zurich and has done research at Berkeley AI Research and Stanford University.
David Dao is a machine learning researcher working on AI for science and the commons, with a focus on nature and climate. He builds datasets, benchmarks and tools for monitoring forests and biodiversity. His most cited work, on data valuation, measures how much each data point improves a model. Earlier, he worked on biological applications of machine learning, including CellProfiler Analyst at the Broad Institute of MIT and Harvard. He co-leads Economies & Governance at Protocol Labs R&D and is co-founder and Chief Scientist of GainForest.Earth.
He holds a PhD in computer science from ETH Zurich, where Ce Zhang and Gustavo Alonso advised his dissertation on data valuation. He has also done research at Berkeley AI Research and Stanford University, and worked as an engineer on autonomous driving in Silicon Valley. He co-authored the GEO-Bench and OAM-TCD benchmarks for Earth observation (NeurIPS 2023 and 2024), and his work has been cited more than 2,500 times (h-index 18).
GainForest works with Indigenous and local communities to close gaps in biodiversity data. With ETH BiodivX, his team surveyed the Amazon with drones and eDNA and won an XPRIZE Rainforest bonus prize in 2024. They also co-designed Taina, a community-owned AI assistant, with four communities near Manaus.
He has been a delegate to the UN climate conferences since COP23 in Bonn in 2017. With the Youth Negotiators Academy, his team built Polly, an AI assistant that youth negotiators from the Global Majority used at COP29 in Baku and UNCCD COP16 in Riyadh, and published AI4COP, a guide to sovereignty-aligned AI for environmental negotiators. They also help delegations run open-weights models on their own hardware.
