Home / Vocabulary
The 117 defined terms
ISO/IEC 22989:2022 defines 117 terms in clause 3, in seven groups: AI, data, machine learning, neural networks, trustworthiness, natural language processing and computer vision. Every term has a page here with the meaning in our own words, what it binds, the evidence to keep and a test you can apply to your own documents.
The vocabulary is the operative part of the standard for most readers, and the part that is public. This page does not reproduce the text. It states each entry in our words and shows how an AI governance lead or an ISO/IEC 42001 implementer would use it.
Start with the terms the generative AI amendment is most likely to touch: artificial intelligence system, model, machine learning, neural network, deep learning, natural language generation and trustworthiness.
The 2022 vocabulary has no entry for generative AI. That absence is the reason the amendment exists. This site tracks the amendment and gives you the baseline it will change, so the gap is visible rather than assumed.
3.1 Terms related to AI
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.1.1 | AI agent | a self-operating entity that perceives its surroundings, reacts to them and acts to reach the goals it has been given |
| 3.1.2 | AI component | a working part that goes to make up an AI system |
| 3.1.3 | artificial intelligence | the research and development work behind AI mechanisms and the applications built on AI systems; the entry is labelled as the discipline sense |
| 3.1.4 | artificial intelligence system | an engineered system that produces outputs including text, forecasts, suggestions or decisions for objectives set by people |
| 3.1.5 | autonomy | a property of a system able to change the scope of its use or its goal without anyone outside stepping in to control or supervise it |
| 3.1.6 | application specific integrated circuit | a chip built for one particular use rather than general purpose work |
| 3.1.7 | automatic | describes a process or system that works under stated conditions with no person in the loop |
| 3.1.8 | cognitive computing | a class of AI systems built to let people and machines interact in a more natural way |
| 3.1.9 | continuous learning | training that keeps being added to a system while it is in operation |
| 3.1.10 | connectionism | a way of modelling cognition with a network of simple linked units |
| 3.1.11 | data mining | a computational method that looks across quantitative data from several angles to surface patterns, sort them and summarise likely links and effects |
| 3.1.12 | declarative knowledge | knowledge held as facts, rules and theorems |
| 3.1.13 | expert system | an AI system that gathers and packages what one or more human experts know in a domain and uses it to reason out answers to problems |
| 3.1.14 | general AI | an AI system type that takes on a wide range of tasks and performs satisfactorily across them |
| 3.1.15 | genetic algorithm | an algorithm that mimics natural selection, breeding and evolving a population of candidate solutions to an optimisation problem |
| 3.1.16 | heteronomy | a property of a system that works under constraint from an outside party that intervenes, controls or supervises |
| 3.1.17 | inference | the reasoning step that reaches conclusions from premises already known |
| 3.1.18 | internet of things | the connected fabric of entities, people, systems and information resources, plus the services that process and act on what comes from the physical and virtual worlds |
| 3.1.19 | IoT device | a part of an IoT system that senses or actuates in the physical world and communicates |
| 3.1.20 | IoT system | a system that delivers internet of things functions |
| 3.1.21 | knowledge | abstracted information on objects, events, ideas or rules, with their links and properties, arranged for systematic use towards a goal |
| 3.1.22 | life cycle | the path followed by a made thing, whether a system, a product, a service or a project, from conception to retirement |
| 3.1.23 | model | a representation, whether physical, mathematical or logical, of an entity, a system, a phenomenon, a process or data |
| 3.1.24 | narrow AI | an AI system type aimed at defined tasks and one specific problem |
| 3.1.25 | performance | a result that can be measured |
| 3.1.26 | planning | the computational work of arranging a set of actions into a workflow that is meant to reach a stated goal |
| 3.1.27 | prediction | what an AI system chiefly puts out when it is given input data or information |
| 3.1.28 | procedural knowledge | knowledge that spells out the steps needed to solve a problem or reach a goal |
| 3.1.29 | robot | an automation system with actuators that carries out intended tasks in the physical world by sensing its surroundings under software control |
| 3.1.30 | robotics | the science and practice of designing, making and using robots |
| 3.1.31 | semantic computing | a computing field that seeks to work out what content means and what a user intends, and to put that in a form a machine can process |
| 3.1.32 | soft computing | a computing field that tolerates and uses imprecision, uncertainty and partial truth so that solving a problem becomes more tractable and robust |
| 3.1.33 | symbolic AI | AI built on techniques and models that handle symbols and structures under rules written out explicitly, in order to draw inferences |
| 3.1.34 | subsymbolic AI | AI built on techniques and models that encode information implicitly, learned from experience or raw data |
| 3.1.35 | task | an action that has to be taken to reach a particular goal |
3.2 Terms related to data
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.2.1 | data annotation | the work of adding descriptive information to data while leaving the data itself unchanged |
| 3.2.2 | data quality checking | examining data for completeness, bias and other qualities that affect how useful it is to an AI system |
| 3.2.3 | data augmentation | making synthetic samples by changing existing data or building on it |
| 3.2.4 | data sampling | choosing a subset of samples so that the patterns and trends of the larger dataset still show |
| 3.2.5 | dataset | a set of data sharing one format |
| 3.2.6 | exploratory data analysis | a first pass over data to find its main features and judge its quality |
| 3.2.7 | ground truth | the value of the target variable for one labelled input item |
| 3.2.8 | imputation | a procedure that fills missing data with estimated or modelled values |
| 3.2.9 | input data | data an AI system uses to work out a predicted output or an inference |
| 3.2.10 | label | the target variable given to a sample |
| 3.2.11 | personally identifiable information | personal data: information that can be tied to a natural person, either directly or through a link, or that can be linked to that person directly or indirectly |
| 3.2.12 | production data | data collected while an AI system is running, used by the deployed system to compute an output or inference |
| 3.2.13 | sample | one atomic data element handled in bulk by a machine learning algorithm or an AI system |
| 3.2.14 | test data | data used to judge how a finished model performs |
| 3.2.15 | validation data | data used to compare how well different candidate models perform |
3.3 Terms related to machine learning
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.3.1 | Bayesian network | a probabilistic model that computes probabilities with Bayesian inference over a directed acyclic graph |
| 3.3.2 | decision tree | a model whose inference is laid out as paths from a root to a leaf in a tree |
| 3.3.3 | human-machine teaming | putting human interaction together with machine intelligence in one working arrangement |
| 3.3.4 | hyperparameter | a setting of a machine learning algorithm that shapes how it learns |
| 3.3.5 | machine learning | the process of tuning the parameters of a model by computational means so that the behaviour reflects data or experience |
| 3.3.6 | machine learning algorithm | an algorithm that sets the parameters of a machine learning model from data against stated criteria |
| 3.3.7 | machine learning model | a mathematical construct that produces an inference or prediction from input data or information |
| 3.3.8 | parameter | an internal variable of a model that changes how it computes its outputs |
| 3.3.9 | reinforcement learning | learning a best sequence of actions by interacting with an environment and seeking a reward |
| 3.3.10 | retraining | updating a model that has already been trained by training it again on different data |
| 3.3.11 | semi-supervised machine learning | machine learning that trains on a mix of labelled and unlabelled data |
| 3.3.12 | supervised machine learning | machine learning that trains only on labelled data |
| 3.3.13 | support vector machine | a machine learning algorithm that places decision boundaries so the margin around them is as wide as possible |
| 3.3.14 | trained model | what comes out of training a model |
| 3.3.15 | training | the process of setting or improving the parameters of a machine learning model with an algorithm and training data |
| 3.3.16 | training data | data used to train a machine learning model |
| 3.3.17 | unsupervised machine learning | machine learning that trains only on unlabelled data |
3.4 Terms related to neural networks
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.4.1 | activation function | a function applied to the weighted sum of all the inputs reaching a neuron |
| 3.4.2 | convolutional neural network | a feed forward neural network that uses convolution in at least one layer |
| 3.4.3 | convolution | a mathematical operation that slides a dot product or cross-correlates the input data |
| 3.4.4 | deep learning | an approach that builds rich hierarchical representations by training neural networks that have many hidden layers |
| 3.4.5 | exploding gradient | a backpropagation problem in which very large error gradients pile up and push the weights so far that the model becomes unstable |
| 3.4.6 | feed forward neural network | a neural network where information moves in one direction only, starting at the input layer and ending at the output layer |
| 3.4.7 | long short-term memory | a recurrent neural network type that handles sequential data well across both long and short dependencies |
| 3.4.8 | neural network | a network built from layers of neurons, one or more, joined by weighted links that can be adjusted, taking input data and producing an output |
| 3.4.9 | neuron | a basic processing element that takes one or more inputs and produces an output by combining them and applying an activation function |
| 3.4.10 | recurrent neural network | a neural network in which the current layer receives outputs from the previous layer and from the previous processing step |
3.5 Terms related to trustworthiness
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.5.1 | accountable | answerable for what was done, decided and performed |
| 3.5.2 | accountability | the state of being answerable in that way |
| 3.5.3 | availability | the property of being reachable and usable on demand by a party that is authorised |
| 3.5.4 | bias | a systematic difference in how certain objects, people or groups are treated compared with others |
| 3.5.5 | control | a deliberate action taken on or within a process to meet stated objectives |
| 3.5.6 | controllability | a property of an AI system that lets a person or another outside agent step into how the system runs |
| 3.5.7 | explainability | a property of an AI system that lets it set out the important factors behind its results in terms people can follow |
| 3.5.8 | predictability | a property of an AI system that lets stakeholders make dependable assumptions about its output |
| 3.5.9 | reliability | the property of behaving and producing results consistently with what was intended |
| 3.5.10 | resilience | the ability of a system to get back to an operational state quickly after an incident |
| 3.5.11 | risk | the effect that uncertainty has on objectives |
| 3.5.12 | robustness | the ability of a system to hold its level of performance whatever the circumstances |
| 3.5.13 | stakeholder | any person, group or organisation that a decision or activity may affect, that may affect it, or that sees itself affected |
| 3.5.14 | transparency | a property of an organisation that communicates the right activities and decisions to relevant stakeholders fully, accessibly and understandably |
| 3.5.15 | transparency | a property of a system that makes appropriate information about it available to relevant stakeholders |
| 3.5.16 | trustworthiness | the ability to meet what stakeholders expect in a way that can be verified |
| 3.5.17 | verification | confirming, with objective evidence, that stated requirements have been met |
| 3.5.18 | validation | showing by objective evidence that the needs of a particular intended use or application have been met |
3.6 Terms related to natural language processing
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.6.1 | automatic summarization | a task that shortens a piece of natural language content or text while keeping the important meaning |
| 3.6.2 | dialogue management | a task that picks the right next move in a dialogue from the user input, the history and other context, to reach a goal |
| 3.6.3 | emotion recognition | a task that computationally identifies and sorts emotions expressed in text, speech, video, an image or a mix of them |
| 3.6.4 | information retrieval | a task that finds relevant documents or parts of them in a dataset, usually from keyword or natural language queries |
| 3.6.5 | machine translation | a task that automatically turns text or speech from one natural language into another with a computer system |
| 3.6.6 | named entity recognition | a task that finds the proper names by which entities are denoted, labels them and sorts them into categories across text or speech |
| 3.6.7 | natural language | a language in active use, or once in active use, in a community, with rules worked out from how people use it |
| 3.6.8 | natural language generation | a task that turns data carrying meaning into natural language |
| 3.6.9 | natural language processing | information processing in a system that rests on natural language understanding or natural language generation; the entry is labelled as the system sense |
| 3.6.10 | natural language processing | the discipline concerned with how systems take in, process and interpret natural language; the entry is labelled as the discipline sense |
| 3.6.11 | natural language understanding | a functional unit takes text or speech in natural language, extracts information from it, and describes both the input and what it represents |
| 3.6.12 | optical character recognition | turning pictures of text, whether typed, printed or handwritten, into machine-encoded text |
| 3.6.13 | part-of-speech tagging | a task that gives a word a grammatical category such as verb, noun or adjective |
| 3.6.14 | question answering | a task that works out the most suitable answer to a question put in natural language |
| 3.6.15 | relationship extraction | a task that finds relationships between entities named in a text |
| 3.6.16 | sentiment analysis | a task that computationally identifies and sorts opinions in text, speech or an image along a range such as positive to negative |
| 3.6.17 | speech recognition | converting a speech signal, by a functional unit, into a representation of what was said |
| 3.6.18 | speech synthesis | the generation of artificial speech |
3.7 Terms related to computer vision
| Clause | Term | Meaning in our words |
|---|---|---|
| 3.7.1 | computer vision | the capability of a functional unit to take in, process and interpret data that represents images or video |
| 3.7.2 | face recognition | automatic pattern matching that compares stored face images with a live face image and reports any match and any identifying data |
| 3.7.3 | image | graphical content in digital form that is meant to be looked at |
| 3.7.4 | image recognition | an image classification process that sorts objects, patterns or concepts in an image |
Last updated: 2026-10-21