IA-08 Environmental Impact

2 sub-dimensions · 14 risk scenarios

Evaluates the ecological and resource impacts of the AI system on the natural environment throughout its entire lifecycle, from development and training to deployment, including direct computational resource consumption, greenhouse gas carbon emissions, and cooling water consumption, as well as indirect behavioral impacts brought by commercial applications.

🎯 BenefitsAdopts lightweight models and green compute scheduling to reduce energy consumption and carbon emissions during training and inference, supporting corporate ESG sustainability goals.

⚠️ Harms / RisksGiant model training and high-frequency inference consume vast amounts of electricity and cooling water resources and generate carbon emissions, or algorithms excessively induce unsustainable consumer behavior.

Direct Environmental Impact4 risk scenarios · ISO 23894 A.14

DefinitionEvaluates the direct environmental impact of the AI system across its entire lifecycle from inception to retirement, including resource consumption.

🎯 BenefitsEstablishes methods for accounting the compute, electricity, and cooling-water consumption of model training and inference, actively pursuing energy-efficiency optimization and green compute procurement.

⚠️ Harms / RisksLarge-model training and inference consume vast electricity and cooling-water resources that are excluded from ESG accounting, exposing the organization to carbon taxes and local water-resource disputes.

Representative ScenariosS0023 Carbon Emissions from Model Training S0024 Water Consumption by Data Centers S0025 Environmental Impact of Critical Mineral Extraction S0098 Resource Waste from Inefficient Inference

Indirect Environmental Impact10 risk scenarios · ISO 23894 A.14

DefinitionEvaluates the indirect environmental impact of AI system deployment, including both beneficial and harmful effects.

🎯 BenefitsGuards against the rebound effect triggered by falling unit compute costs (Jevons paradox), e-waste from premature hardware retirement, and the cross-regional transfer of environmental costs.

⚠️ Harms / RisksLower unit inference costs spur departments to expand AI use without restraint (rebound effect), or frequent hardware turnover generates large volumes of electronic waste.

Representative ScenariosS0016 Concentration of Technology and Compute S0017 Digital Divide Widening Existing Inequality S0019 Intensifying International Technology Competition S0020 Market Competition Eroding Safety Standards S0021 Global Gap in Research Capability S0022 Imbalanced Labor Conditions in Data Annotation S0204 Rebound Effect Offsetting Efficiency Gains S0206 Electronic Waste from Hardware Replacement S0207 Premature Retirement Worsening Waste S0208 Environmental Costs Transferred to Vulnerable Regions