The problem

Projects that use AI to optimise things rarely count the AI's own energy. Without that number, a 'green' AI project can be net negative.

The fix you can use today

• Estimate the AI's energy per run from published provider figures.
• Set it against the saving, over the years the result is deployed.
• Skip runs where the return doesn't justify the inference.

GreenCode worked out the energy cost of its own AI, and it turns down work that would not pay that back.

Tools that help

• CodeCarbon: a Python package that estimates the energy and CO2 of a run from hardware power and regional grid intensity
• AI Energy Score (Hugging Face): comparable energy ratings for AI models on common tasks

How GreenCode helps

GreenCode helps tackle this by setting its own AI footprint against the saving it makes, in its climate case. It measures reductions in absolute terms and shows owners where capacity can be retired, so a saving is banked rather than spent. More on this.

Tags

#llm#esg#governance

Sources

• The climate case for GreenCode: the footprint of the AI itself
• Green AI and energy-aware computing

All GreenCode Tech Tips