Using AI without adding to the problem

1

Light on compute by default

We prefer text and data work to generative images and video, which use far more computing power for what they produce. We will not adopt those tools for routine work until their footprint is shown to be low.

2

Use the smallest tool that does the job

A script, a search or a small model comes before a large one. We avoid long-running agents and repeated runs where one careful step would do.

3

Measure what we use

We are setting up a record of how much AI we use and the energy it is estimated to cost, so we can review it and act on it. Until it is in place, we say so and do not claim figures we do not have.

4

Publish what we measure

Once we have numbers, we will share them with their method and source, including when they are not flattering.

5

A person checks what we publish

AI can help draft and sort. A person reviews anything before it appears on this site, and hub entries show who reviewed them and when.

6

Sources before synthesis

We never present AI output as evidence. Figures and claims link to the original publisher so you can check them yourself.

7

Use it where it reduces harm

We use AI where it helps people understand and act on climate information, and we do not use it to produce volume for its own sake.

8

Review and revise

The evidence on AI's environmental cost is changing. We revisit these principles as it does, and we correct them when we find we have fallen short.

Managed by Menlo IQ LLC