AI Sustainability Rating - Wiebren van der Zee ABNAMRO
When multiple models could do the (AI) trick for your use case, which one would you choose? Do you take an open source small model, the most sovereign solution or will you go for the greenest option? Using Grok via Gas fueled turbines in the US versus selecting Mistral operating on the 'Nuclear' French grid will yield very different carbon outcomes!
For the TNO Future Proof AI Challenge, an AI Sustainability Rating framework was developed to help organizations choose more sustainable AI solutions. This model supports companies in distinguishing between “dirty” and “cleaner” AI, especially when pursuing goals such as energy reduction, net zero, and circularity.
The rating acts as a non-functional design driver that stimulates the dialogue between business and IT about the impact of AI. In addition, the model can be applied in processes such as impact assessments, architecture, and procurement.
An important feature is its connection to EU legislation (such as EED, ETS, and CBAM), creating a financial incentive to reduce CO₂ emissions. Other frameworks (such as CE and ESPR) can also be linked to encourage responsible use of materials.
In short, the framework helps in the selection, design, and assessment of sustainable AI, rewarding environmentally friendly solutions while discouraging more polluting alternatives.
The next step is to validate the criteria and metrics together with experts and stakeholders.
Share this post!