Module 2: AI Impacts & Responsible AI, BoK II.A
Environmental harms
AI has environmental impacts across training, inference, hardware production and disposal. Emissions and water estimates depend on model, location, energy source, cooling and system boundaries. A historical estimate for one configuration is not a universal per-model or per-query value.
AI has environmental impacts across training, inference, hardware production and disposal. Emissions and water estimates depend on model, location, energy source, cooling and system boundaries. A historical estimate for one configuration is not a universal per-model or per-query value.
- Carbon emissions depend on computing demand and the energy supply. Compare studies only after checking their dates, configurations, units and system boundaries. Avoid memorizing one historical estimate as the footprint of every large model.
- Energy consumption arises during training and inference. Its impact depends on use volume, hardware efficiency and electricity supply.
- Water use depends on cooling, electricity generation, location and operating conditions. There is no single universal bottle-per-query conversion.
- Lithium extraction → battery lithium demands enormous water usage; future availability is a worry → rapid demand growth, geopolitical concentration, slow mine development.
Sources and study method
This independent lesson uses active recall, spaced retrieval and scenario practice. Read the full study method.