Sustainable AI Manifesto
On one hand, AI made outstanding progress and entered every software
development workflow. On the other hand, AI is built on huge energy use,
resource extraction, and underpaid click workers. Despite this, the software
industry massively bet on it. The consequence is that a software vendor can't
avoid it. But what we can do is choose how we integrate it before adding
anything. That's why we wrote guidelines to ensure it is deployed in line with our
values.
This manifesto lays out the limits we set to make it viable in the long term.
AI should be designed with respect for its users.
- Our features will empower production teams and augment their capacity
- We won't add elements to push you to consume more
- We won't implement things whose usefulness hasn't been validated
- We will prioritize vendors with ethical labor practices
- We will disclose limits like hallucination
AI consumes a lot of energy, and we should limit this.
- Our tools will allow tracking token consumption
- Carbon tracking will be included in the reports
- Selecting models, such as local ones, will be possible
- If not, users will be informed about which model is used
- Lighter models will be our priority
AI features should be implemented like any other.
- We will avoid AI colors, icons, assistant names, faces, or magic metaphors to
depict an AI feature
- We will allow you to turn it off and will ask for confirmation before
activating it
- The same AI function will not appear twice
- Our feature descriptions will tell what they do, not what they evoke
- We won't add any pushy onboarding
Thanks to the Limites Numériques collective for the inspiration.