Dictionary

24/7 carbon-free energy is an approach that seeks to match electricity consumption with carbon-free electricity supply on an hourly or similarly granular basis.

An AI system that can plan and take actions toward a goal with some autonomy; an emerging layer in software workflows.

The greenhouse-gas emissions associated with AI systems across hardware manufacture, training and operation.

Applying machine learning to assess physical and transition climate risks to assets, portfolios and supply chains.

A false or unsupported AI-generated output presented as reliable; often called an AI hallucination.

Detecting forest loss and degradation from satellite imagery with machine learning, supporting enforcement and zero-deforestation commitments.

The electricity used to train and run AI models, concentrated in data centres and growing with model scale.

Applying AI to species identification, habitat mapping and ecosystem monitoring in support of conservation goals.

A movement and platform, anchored by the ITU, promoting AI applications that advance the UN Sustainable Development Goals.

AI governance is the set of policies, roles, controls and review processes used to manage the responsible use of artificial intelligence systems.

Using AI to balance electricity supply and demand, dispatch storage and integrate variable renewables.

The mining, manufacturing and logistics behind chips and servers, linking AI to critical minerals, energy and labour issues.

Using machine learning to screen and design new materials, accelerating batteries, catalysts and low-carbon alternatives.

Exaggerating or misrepresenting AI capabilities or benefits in claims; increasingly applied to sustainability marketing.

The freshwater consumed by data centres and power generation supporting AI, mainly for cooling and electricity production.

Machine-learning models that generate weather forecasts, increasingly matching or extending conventional numerical methods.

Machine-learning analysis of satellite data to track land use, oceans, ice and environmental change.

Using AI to generate, amplify or give false credibility to misleading environmental claims.

A defined set of rules or steps a computer follows to solve a problem; the building block underlying every AI system.

Holding organisations answerable for the outcomes of the algorithms they deploy, including environmental and social harms.

Independent examination of an algorithm's behaviour and impacts, extending assurance practice to AI systems.

Algorithmic bias is a systematic distortion in automated outputs that can arise from data, design choices, assumptions or deployment context.

The use of algorithms to make or support decisions, with consequences for accountability in sustainability and social outcomes.

A structured evaluation of an automated decision system's effects before deployment, covering rights, equity and environment.

The use of algorithms to direct, monitor and evaluate workers, with implications for decent work in supply chains.

Hypothetical AI with human-level ability across most cognitive tasks; a focus of safety and governance debate.

Machine systems that perform tasks associated with human intelligence — increasingly central to sustainability both as a tool (monitoring, modelling, efficiency) and as a fast-growing source of energy demand.

Using software and AI to calculate organisational or product emissions from activity data, replacing manual estimation.

The use of technology to perform tasks with reduced human intervention, with consequences for employment and just transition.

A system that operates without continuous human control, from monitoring drones to self-optimising industrial equipment.

Using microphones and AI to detect and identify species from environmental sound, tracking biodiversity at scale.

An AI model whose internal reasoning is not interpretable by its users, complicating accountability for its outputs.

AI-driven control of heating, cooling and ventilation to cut building energy use while maintaining comfort.

A data-centre metric relating total carbon emissions to IT energy use, complementing PUE with a carbon dimension.

Scheduling or relocating computing workloads to times and places where grid electricity is cleanest — cutting the carbon footprint of digital services without new hardware.

Software that converses with users in natural language, now commonly built on large language models.

The application of data science and machine learning to climate data and climate-science questions.

A European industry self-commitment to make data centres climate neutral by 2030 through efficiency, clean energy, water conservation and heat reuse.

An interdisciplinary field applying computational methods, including AI, to environmental, social and economic sustainability problems.

AI that interprets images and video, widely used in satellite monitoring, biodiversity surveys and industrial efficiency.

The use of AI tools such as camera-trap image recognition and acoustic sensors in wildlife protection and anti-poaching.

Facilities housing IT infrastructure that consumed around 415 TWh of electricity in 2024 — about 1.5 per cent of global demand — and growing fast with AI.

A critical term for the extraction of data from people and territories, echoing historical resource colonialism.

Annotating data so models can learn from it, often performed by large low-paid workforces — a social dimension of AI supply chains.

Data sovereignty is the principle that data is subject to the laws, governance expectations and control rights of the place or community connected to it.

Machine learning using multi-layered neural networks; the dominant approach behind modern AI and its rising compute demand.

The emissions attributed to an individual's or organisation's use of digital products and services; a contested but widespread term.

The use of digital technologies, including AI, satellites and sensors, to monitor, report and verify emissions and carbon projects.

The responsible design, use and disposal of digital technologies to minimise their environmental footprint — energy, materials and e-waste.

A digital twin is a digital representation of a physical asset, process or system that is updated with data to support analysis or decisions.

EU rules setting minimum energy-efficiency and material-efficiency requirements for servers and data-storage products.

Running AI models on local devices rather than in data centres, reducing data transmission and enabling low-power monitoring.

Inferring individual appliance consumption from aggregate meter data, enabling efficiency feedback without extra hardware.

The European Union's risk-based regulation of AI systems, including transparency and reporting duties with environmental dimensions.

A voluntary European framework setting best practices for data-centre operators to reduce energy use.

The mandatory reporting of data-centre energy, water and efficiency indicators under the recast EU Energy Efficiency Directive.

Techniques making AI model outputs understandable to humans, supporting accountability for AI-influenced sustainability decisions.

Training AI models across distributed devices without centralising raw data, relevant to data sovereignty and privacy.

Additional training that adapts a pre-trained model to a specific domain or task, such as sustainability-report analysis.

A large AI model trained on broad data that can be adapted to many downstream tasks, including sustainability applications.

Designing AI systems to achieve acceptable performance with minimal data, compute and energy, countering ever-larger models.

An AI model usable across many applications; the EU AI Act assigns specific obligations to GPAI providers, including energy-related documentation.

AI systems that produce new text, images, code or other content, raising both sustainability-use cases and footprint questions.

Established 2017 guidance for assessing life-cycle emissions of ICT products and services, subject to newer GHG Protocol standards and updates.

A multi-stakeholder initiative, hosted by the OECD, guiding responsible AI development including climate-relevant work.

Research and practice aimed at reducing the environmental cost of AI itself, as distinct from using AI for environmental ends.

The established field of environmentally responsible design, use and disposal of computers and IT systems.

A data centre designed and operated to minimise energy, water and carbon impacts through efficiency and clean power.

Software development practices that reduce energy use and emissions, including carbon-efficient architecture and code.

Human-in-the-loop describes a process in which a person reviews, guides or approves automated outputs before they are relied on.

The combined emissions of information and communication technology, spanning devices, networks and data centres.

Using a trained model to generate outputs; for widely deployed AI services, cumulative inference is a major and growing share of energy use.

The Internet of Things is a network of connected devices or sensors that collect and exchange data about physical conditions or activities.

The international standard for AI management systems, specifying requirements for organisations developing or using AI.

The ITU standard setting greenhouse-gas emission trajectories for the ICT sector in line with the Paris Agreement.

A structured network of entities and relationships used to organise data, underpinning traceability and ESG data systems.

A large language model is an AI model trained on large text datasets to generate, classify, summarise or transform language-based content.

Assessing AI systems' environmental impacts from raw-material extraction through manufacture, use and end of life.

A branch of AI in which systems learn patterns from data rather than following explicit rules; the technical basis of most AI used in sustainability work.

A machine learning model is a computational model trained on data to identify patterns, make predictions or classify information.

Identifying and quantifying methane leaks from satellites and sensors using AI, a fast-growing mitigation tool.

Compressing a large model into a smaller one that keeps much of its performance while cutting compute and energy demand.

The compute-intensive process of fitting an AI model to data; the stage where much of an AI system's energy use is concentrated.

AI models that work across text, images, audio and other data types, enabling richer environmental monitoring and analysis.

AI designed for a specific task, in contrast to AGI; all AI systems in current sustainability use are narrow.

AI techniques for analysing and generating human language, applied to sustainability reports, regulations and disclosures.

A computing model of connected layers of units, loosely inspired by the brain; the basis of deep learning.

A voluntary US framework for identifying and managing AI risks, including those linked to sustainability claims.

Intergovernmental principles for trustworthy AI, adopted in 2019 and updated in 2024 to address general-purpose and generative AI.

An AI model whose trained parameters are publicly released, enabling scrutiny, reuse and local deployment.

The standard metric for data centre energy efficiency: total facility energy divided by IT equipment energy, with 1.0 as the theoretical ideal.

Using statistical and machine-learning techniques to forecast outcomes, from energy demand to supply-chain disruption.

Designing the inputs given to generative AI systems to steer the quality and reliability of their outputs.

Algorithms that rank and suggest content or products, shaping information exposure and consumption patterns at scale.

A machine-learning approach where an agent learns by trial and error, used in energy-system optimisation and grid control.

Remote sensing is the collection of information about land, water, vegetation or infrastructure without direct physical contact, often using aircraft or satellites.

Predicting wind, solar and other variable generation output, where machine learning improves grid integration.

Principles and practices ensuring AI is developed and used in ethical, accountable and socially beneficial ways.

Grounding a language model's answers in retrieved documents, used to tie AI outputs to verified sources.

Satellite monitoring is the repeated use of satellite observations to detect, assess or document conditions and changes on the Earth surface.

A compact language model needing far less compute than frontier models, relevant to lower-footprint AI deployment.

A smart grid is an electricity network using digital monitoring, communication and control to manage supply, demand, reliability and integration of distributed resources.

A Green Software Foundation specification scoring the carbon emissions of a software system per unit of work.

Machine learning trained on labelled examples, used for tasks such as classifying satellite imagery or predicting emissions.

An umbrella for developing and deploying AI within environmental limits and social safeguards across its life cycle.

Artificially generated data used to train AI models where real data are scarce, as in rare-species detection or extreme-climate events.

Machine learning on very low-power microcontrollers, enabling environmental sensing with minimal energy use.

The data a model learns from; its quality, representativeness and provenance shape model behaviour and bias.

Adapting a model trained on one task to a related task, reducing the data and compute needed for new applications.

The neural-network architecture behind modern language and multimodal AI, built on attention mechanisms.

AI systems that are lawful, ethical and robust; the framing used by the EU's high-level expert group.

A global normative framework on AI ethics, explicitly including environmental and ecosystem protection.

Machine learning that finds structure in unlabelled data, for example clustering sites, suppliers or consumers by behaviour.

Capturing and reusing the heat produced by servers, for example in district heating, to improve overall energy efficiency.

A data-centre metric comparing annual water use to IT equipment energy, used to track cooling-related water demand.