References
In presentation order:
- MIT Technology Review
- Food and Water Watch
- Epoch AI’s Trends Insights
- ‘Roadmap’ shows the environmental impact of AI data center boom
- Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends
- Our contribution to a global environmental standard for AI [Mistral]
- AI Energy Star Project (Hugging Face): https://huggingface.co/spaces/AIEnergyScore/Leaderboard
- ML Energy Leaderboard: https://ml.energy/leaderboard/ from the ML Energy Initiative
- Grid.io
- Green Algorithms https://calculator.green-algorithms.org/
- ML CO2 Impact: https://mlco2.github.io/impact/
- GreenPT (Eco-friendly AI): https://greenpt.ai/
- EcoLogits Calculator
- ScaleDown (browser extension)
- Offset AI
- SustAIn (browser extension)
- AI Wattch (browser extension)
- Shift web browser
Photo credit: DocChewbacca on Flickr
Additional Resources.
Key Points
These are notes in addition to the material on the slide.
- Slides 3-5: These slides have a lot of numbers and they’re telling you something you probably already know: AI is a very resource-intensive technology. Organizations and individuals are using AI more every day and tech companies are in an arms race to buy chips and build data centers. The bottleneck in all this is energy. As power usage increases 3-4-fold, the available energy won’t be enough to keep up.
Some companies are investing in green and renewable power, which is an important step, but those will take longer to bring online. In the meantime, we’re using quick energy – which is also the dirtiest and more carbon-intensive: coal and natural gas. - Slide 6: I’m not going to tell you not to use AI. Just because tech CEOs have prioritized innovation over environmental impact does not mean that you should be avoiding what is clearly a generational technology. But we SHOULD be responsible for our own use, as much as we’re able.
- Slides 7-16: How can I pick a sustainable model? There isn’t much data out there, but here are some places you can extrapolate how responsible the tech company is being with respect to their climate impact.
If you have your own model, here are some ways you can make responsible estimates of your impact.
If you seriously want to lessen your impact, your best bet is probably doing it yourself, or using a dedicated tool like GreenPT, but that may not be viable for your use case. - Slides 17-21: If I can’t use a truly green model, what are some ways I can use AI more efficiently?
- Slides 22-29: It doesn’t seem like the options available to me will help much. How can I make a bigger impact?
- Slides 30-31: You can make an impact. Don’t stop doing the right thing because it doesn’t seem like enough.
- Slides 32: Thank you.
