Environmental applications
AI can classify satellite images to monitor forest change, identify unusual air-quality readings and help forecast electricity demand. These applications may be useful, but an algorithmic output is not independent scientific evidence. Field observations, calibrated sensors and expert checks remain necessary. A school project should document the date, scope and missing observations in any dataset before presenting a result as reliable.
Electricity demand and climate
The International Energy Agency estimated in its 2025 Energy and AI report that all data centres worldwide used around 415 TWh of electricity in 2024. Its base-case projection for 2030 was around 945 TWh. These are figures for data centres as a whole, not for AI alone. The climate impact of an AI service depends on electricity generation, location, time of use, hardware efficiency and demand. No single fixed carbon value applies to every AI question.
Water, minerals and electronic waste
Cooling can require water directly, while electricity generation and semiconductor manufacturing add indirect water demands. A 2026 United Nations University study argues for assessing carbon, water and land footprints together. Low-carbon electricity is not necessarily the option with the lowest water footprint. Manufacturing chips and servers also requires materials and energy. Equipment lifespan, repairability and the handling of electronic waste must be included in a life-cycle assessment.
Measuring net environmental benefit
Start with a baseline method that does not use AI. Compare the accuracy, energy needs, staff time and material use of both approaches for the same task. A faster result is not proof of lower environmental impact. When a service becomes cheaper and is used more frequently, total demand may rise: this is known as the rebound effect. Local grid constraints and water stress also matter when evaluating infrastructure.
Responsible use in education
Do not present AI-generated flood or wildfire images as genuine photographs. Verify generated statistics, graphs and citations against their original sources. A spreadsheet or smaller tool may be sufficient for a simple calculation. Individual efforts to avoid unnecessary computing are useful, but they cannot replace providers' responsibility to disclose resource consumption and reduce the environmental impact of their operations.
A seven-day classroom exercise
On day one, choose an environmental question and a trustworthy dataset. On day two, test a non-AI baseline. On day three, try a suitable AI-assisted method. On day four, compare accuracy. On day five, document sources and uncertainty. On day six, investigate lower-resource alternatives. On day seven, share the findings with methods and limitations. Never assume that unmeasured electricity or water use is zero.
Verifiable sources
IEA Energy and AI (2025)United Nations University (2026)UNEP environmental impact of AIProjections are not observed outcomes.
Explore with e-NetCoM
Green Intelligence Workshop and environmental literacy resources
e-NetCoME-MerkezInstagram ↗YouTube ↗X ↗