Recommendations for public action towards sustainable generative AI systems

Generative AI systems consume massive amounts of water and energy: The paper documents that training large language models produces considerable greenhouse gas emissions and water consumption, demonstrating that the environmental footprint of generative AI remains a significant problem despite growing awareness of digital technology's environmental impact.

Technical solutions already exist to reduce AI's environmental footprint: The research identifies concrete ways to lower environmental impact, including using more efficient processors and optimizing data center energy performance, indicating that sustainability improvements don't require waiting for breakthrough innovations.

Environmental protection is recognized in AI ethics frameworks but lacks enforcement: While environmental protection has been established as a founding principle of AI ethics at international and European levels, the paper emphasizes that this recognition "has not yet translated into concrete measures," highlighting a gap between stated values and actual implementation.

A coordinated, long-term strategy is needed: The paper recommends twelve pragmatic actions for public policy, including building long-term strategies for carbon-neutral AI models and encouraging international cooperation to set common standards, suggesting that isolated initiatives are insufficient to address the scale of the problem.

Regulation and scientific support are essential next steps: The recommendations call for developing appropriate legal and regulatory frameworks alongside continued scientific research support, indicating that addressing sustainable AI requires both governance structures and continued innovation.