Dr. John Bistline Head of Science Watershed
WATCH NOW
AI for sustainability: What Watershed AI fellows learned in 8 weeks
Q. WHAT CAN A COMPANY ACTUALLY DO WITH THIS TODAY, VERSUS WHAT HAS TO WAIT FOR BETTER PROVIDER DATA?
» A company can use whatever data it has today, even a dollar figure on an AI vendor invoice, and get a defensible estimate using our lowest tier. As they get more granular data, like token counts and provider-reported factors, they move up the tiers and sharpen the estimate. The tiers are designed so that better data never means starting over. What has to wait is more precise interaction-level data, including hardware specifics and provider-reported emissions per token. That disclosure is improving, but slowly. In the meantime, the two biggest levers to reduce emissions, model choice and grid region, don’ t depend on it at all.
Q. HOW IS WATERSHED’ S SCIENCE TEAM DIFFERENT?
» Most of what we build, we publish. Our team publishes the methods behind our work so that other companies – including our competitors – can use, inspect, and challenge them. We’ re focused on making the field of sustainability better. That includes work on building credible AI emissions factor mapping, the AI emissions framework itself, and ATLAS, a spend classification benchmark for estimating Scope 3 carbon emissions.
Watershed also works to keep critical datasets like CEDA and USEEIO open to the public and regularly maintained, which we do as part of the Cornerstone data initiative with Stanford and ERG. Those datasets sit underneath a large share of corporate carbon footprints worldwide.
sustainabilitymag. com 79