Field NotesWhat we're learning about AI and the environment

What environmental problem are you working on?

Read what other teams tried, what happened, and what they would do differently, including the things that didn't work.

Try: making sense of sensor data data center water use wildlife cameras solar forecasting

Browse by theme

Latest learnings

Surprise

Whale codas turned out to combine features, but what they mean is still unknown

From public sources: the team expected to find a fixed set of click patterns. Instead, whales combine rhythm, tempo, small timing shifts and extra clicks into many variations that depend on the conversation. The authors stress that this shows structure,...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.
Worked

AI recommendations cut cooling energy by up to 40% in live tests

From public sources: DeepMind reported cooling energy savings of up to 40%, which works out to about a 15% cut in total energy overhead once other losses are counted. The authors noted that every data center is built differently, so a model tuned for one...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.
Surprise

Underwater servers failed at one-eighth the rate of servers on land

From public sources: Microsoft reported a failure rate underwater of about one-eighth of a matching set of servers on land. The team's leading explanation was the sealed nitrogen atmosphere, which corrodes less than air, and the absence of people bumping...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.
Surprise

It worked, and Microsoft stopped anyway

From public sources: in 2024 a Microsoft executive told DatacenterDynamics the experiment worked but that the company was not building undersea data centers, and would apply what it learned about operations and vibration elsewhere. A successful pilot is...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.
Surprise

About 75% of industrial fishing vessels were invisible to public tracking

From public sources: comparing satellite images from 2017 to 2021 with ship-tracking broadcasts, the team found about 75% of industrial fishing vessels and over 25% of transport and energy vessels were missing from public tracking. Not all boats are...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.
Worked

Early results suggest up to 30% fewer stops, with caveats

From public sources: Google describes early numbers showing potential for up to 30% fewer stops and 10% lower greenhouse gas emissions at coordinated intersections. It notes the figures are early, averaged, and modelled on a single vehicle type, and that...

⚠ Not yet claimed by the project owner or their organization. This information may not be completely accurate.

16 projects and 10 learnings so far.

Did your team try something?

Upload a report or paste a link, and we'll draft the entry for you to check. No account needed, and nothing is required beyond a project name and your email address.

Share a project Share a single learning