GOLESTAN (Sally) Radwan
Chief Digital Officer of the United Nations Environment Programme
Sustainable AI can be understood as an equation balancing inputs and outputs. On one side are the resources involved in developing and operating an AI model or system, including electricity, water, critical minerals and the electronic waste produced in the process. On the other side are the economic, social and environmental benefits that AI generates. AI cannot be considered genuinely sustainable until both sides can be quantified systematically and the benefits at least balance, and ideally outweigh, the associated resource and environmental costs. On the input side, there is still a lack of common global standards, methodologies and comparable data. On the benefits side, AI faces a serious scalability problem. Large amounts of funding go into small, rapid and highly visible prototypes and pilot projects, but many cease to work when they are scaled because the necessary funding, data or people are unavailable. This widespread condition can be described as “AI pilotitis.” Sustainable AI is also highly contextual and use-case-dependent. Using AI to identify new fossil-fuel resources and using it to support renewable energy involve very different inputs and outputs. Governance should therefore begin with specific problems that can genuinely be measured and controlled. By addressing many smaller problems, we may eventually connect the dots and build a broader framework for sustainable AI.
