Measuring What Matters: ITU’s Approach to Mitigating Environmental Impact

The ITU has published a report titled “Measuring What Matters: How to Assess AI’s Environmental Impact,” offering a thorough overview of the current methods for evaluating the environmental impact of AI systems. The report, created by the Sustainable AI working group in partnership with the Green Digital Action initiative, was released on 10 July, at the ITU’s annual AI for Good Global Summit.

The review focuses primarily on identifying which components of AI’s environmental impacts are being measured and evaluates the effectiveness and transparency of these practices. The report draws from academic studies, corporate sustainability initiatives, and emerging environmental tracking technologies, examining the current methodologies in place and offering recommendations where limitations are observed.

There were key findings and gaps identified in the report, including:

• Over-reliance on estimates: Current methods of measurement heavily rely on proxies and indirect estimates, creating significant data gaps and hindering the accountability of the AI system
• Lack of standardized methodologies: The tools used to measure AI’s impact are inconsistent across research platforms, lacking standard definitions and reporting units
• Carbon-centric metrics: Environmental assessments focus frequently on greenhouse gas emissions, neglecting other crucial impacts like biodiversity loss, electronic waste, and resource depletion
• Underexplored life cycle stages: Key stages of an AI model’s life cycle, such as the inference and usage phases, are severely under analyzed in current assessments

The report makes recommendations, appealing especially to developers (producers), users (consumers), and policymakers.

The ITU overall stresses the need for a more holistic approach to AI impact analysis, moving beyond current carbon-centric, fragmented, and indirect measurement practices toward more standardized and comprehensive methodologies.

To read the full report, click here.