Artificial Intelligence and the Environment: Ethical Challenges and Strategic Opportunities

Vocabulary flashcards covering the ethical concerns, resource consumption metrics, regulatory environments, and strategic opportunities of artificial intelligence and the environment from Marabelli and Davison (2025).

Net-Zero Emissions

An environmental goal wherein the quantity of greenhouse gas emissions released into the atmosphere by an organization or entity does not exceed the amount removed by that same entity.

Carbon Dioxide Equivalent ($\text{CO}_2\text{e}$)

The standard metric unit for quantifying carbon footprints, calculated by converting the atmospheric warming impacts of various greenhouse gas emissions into an equivalent amount of carbon dioxide.

Greenwashing

The practice of publishing deceptive, vague, or unsupported environmental claims—found by a 2022 European Union study to characterize $53\,\%$ of corporate green claims, with $40\,\%$ lacking supporting evidence—to appear sustainable without implementing substantive operational changes.

Carbon Offsetting

A compensatory mechanism in which organizations purchase verified credits or tokens from third-party environmental projects (such as reforestation or renewable energy) to neutralize their own greenhouse gas output without altering their resource-intensive operations.

Water Security

The capacity of a population to safeguard sustainable access to adequate quantities of acceptable quality water for sustaining livelihoods, human well-being, and socio-economic development, ensuring protection against water-borne disasters, and preserving ecosystems.

Water Insecure Countries (UN African Designation)

A group of thirteen African nations identified by the United Nations as lacking adequate water security: Chad, Comoros, Djibouti, Eritrea, Ethiopia, Liberia, Libya, Madagascar, Niger, Sierra Leone, Somalia, South Sudan, and Sudan.

Generative Artificial Intelligence (GAI) Resource Intensity

The heavy demand of GAI systems for electricity and water, driven by complex models requiring constant training across their entire lifecycle; server racks running AI consume up to $100\,\text{kW}$, compared to $7\,\text{kW}$ for standard web or mail server racks.

AI Cannibalism

The degradation risk and systemic phenomenon where generative artificial intelligence models are recursively trained on synthetic content and data generated by other AI models rather than authentic human-generated data.

Model Distillation

A training process used by organizations such as Deepseek to create smaller, cheaper, and less energy-demanding AI models by transferring learning from massive large language models into compact systems.

Adverse Digital Incorporation

A conceptual framework formulated by Heeks (2022) describing the inequitable or disadvantageous inclusion of Global South populations, workers, and cultures into global digital structures and platform economies.

Smart Cities

Urban settings embedded with sensors and automated systems to optimize municipal services—including traffic flow, waste processing, and water supply management—with the goal of curbing resource usage and greenhouse gas emissions.

Digital Twin

A virtual computational model of a physical environment (such as the one created for the city of Barcelona) that uses real-time sensor streams and AI to simulate, analyze, and forecast operations like municipal traffic and energy usage.

Smart Farming

Also termed digital or e-agriculture, this practice uses machine learning algorithms (such as regression models) to evaluate environmental conditions and determine optimized farming interventions, such as precise crop water allocations.

GraphCast

Google DeepMind's medium-range weather forecasting artificial intelligence model, trained on $40\text{ years}$ of historical atmospheric data, capable of outperforming conventional numerical weather forecasting systems.

TECTONIC Project

A research initiative funded by the European Commission that applies artificial intelligence techniques to forecast the location, depth, and intensity of earthquakes with higher precision than conventional seismic analysis.

Project Natick

A research project conducted by Microsoft that deployed experimental underwater data centers off coastal cities to leverage oceanic water for natural cooling, thereby conserving electricity and terrestrial fresh water.

The Great Carbon Divide

A 2023 research report by The Guardian, Oxfam, and the Stockholm Environment Institute showing that an individual in the bottom $99\,\%$ of global population would require $1{,}500\text{ years}$ to produce the carbon footprint generated by the wealthiest billionaires in one year.

Artificial Intelligence Environmental Impacts Act

US federal legislation introduced on February 1, 2024, by Senator Edward J. Markey to require standardized investigation, measurement, and reporting of the ecological and environmental impacts of artificial intelligence systems.

Shallow Neural Networks

Simplified neural network models capable of learning from small datasets (such as a threshold of $10{,}000$ data points) in scenarios with predefined outcomes, functioning as a less compute-intensive alternative to deep GAI architectures.

Secondhand Environmental Effects of AI

Indirect ecological damages stemming from algorithmic deployment, such as e-commerce nudging that triggers excess consumer buying, frequent expedited deliveries by gig drivers, extra vehicle fuel consumption, and aquatic tire-wear microplastic runoff.

Lonestar

A US-based data storage and recovery enterprise that explored deploying data centers in outer space in March 2025 to take advantage of continuous solar power and radiative heat dissipation.

Movement for a Sustainable Uruguay (MOVUS)

An activist organization that campaigned in 2023 against Google's development of a large-scale data center in Uruguay due to concerns over intensive commercial exploitation of domestic freshwater supplies.

AI Value Chain Offshoring

The operational practice of delegating manual data labeling, prompt validation, and model grooming to low-wage contractors and gig workers (frequently including minors) in Global South nations such as India, Pakistan, and Venezuela.

EU AI Act Risk Tiers

The European Union's risk classification framework (low, medium, high) based on direct human impact, which scholars argue fails to capture indirect and proxy environmental harms generated by computationally intensive systems.

Executive Order 14110 Revocation

The formal termination of the 2023 US executive order on trustworthy AI development through Section 5 of the January 23, 2025 executive order 'Removing Barriers to American Leadership in Artificial Intelligence,' which removed safety guardrails including environmental reviews.

AI and the Environment Roadmap (Figure 1)

A conceptual framework by Marabelli and Davison mapping how GAI's resource consumption drives ethical challenges (continuous training tweaks, nontransparent reporting, unequal global impacts) and strategic opportunities (smart cities, predictive disaster models, nuclear energy), mediated by national and global regulations.