The term is discussed in various disciplines from different perspectives.  Due to the AI effect, the term is constantly evolving. Three definitions are given below:

Definition 1: Capability of an engineered system to acquire, process and apply knowledge and skills (ISO/IEC 29119-11).

Definition 2: Computerized system that uses cognition to understand information and solve problems (ISO/IEC 22989).

Definition 3: Artificial intelligence refers to a set of technologies [...] that, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with (European AI Act in draft ¹).

• Machine Learning
• AI effect

ETSI GR ENI 004 computerized system that uses cognition to understand information and solve problems

NOTE 1: ISO/IEC 2382-28 defines AI as "an interdisciplinary field, usually regarded as a branch of computer science, dealing with models and systems for the performance of functions generally associated with human intelligence, such as reasoning and learning".

NOTE 2: In computer science AI research is defined as the study of "intelligent agents": any device that perceives its environment and takes actions to achieve its goals.

NOTE 3: This includes pattern recognition and the application of machine learning and related techniques.

NOTE 4: Artificial Intelligence is the whole idea and concepts of machines being able to carry out tasks in a way that mimics the human intelligence and would be considered "smart".

ISO/IEC DIS 22989 set of methods or automated entities that together build, optimize and apply a model (3.1.26) so that the system can, for a given set of predefined tasks (3.1.37), compute predictions (3.2.12), recommendations, or decisions.

Note 1 to entry: AI systems are designed to operate with varying levels of automation (3.1.7).

Note 2 to entry: Predictions (3.2.12) can refer to various kinds of data analysis or production (including translating text, creating synthetic images or diagnosing a previous power failure). It does not imply anteriority. study of theories, mechanisms, developments and applications related to artificial intelligence (3.1.2).

ISO/IEC TR 29119-11:2020 capability of an engineered system to acquire, process and apply knowledge and skills

ISTQB - CTAI Syllabus The capability of an engineered system to acquire, process, create and apply knowledge and skills (ISO/IEC TR 29119-11)

The case when an AI system is no longer considered to be AI due to the advancement of technology.

ISO/IEC TR 29119-11:2020 situation when a previously labelled AI (3.1.13) system is no longer considered to be AI as technology advances

ISTQB - CTAI Syllabus The situation when a previously labelled AI system is no longer considered to be AI as technology advances (ISO/IEC TR 29119-11)

A system that integrates one or more AI components.

ISO/IEC TR 29119-11:2020 system including one or more components implementing AI (3.1.13)

ISTQB - CTAI Syllabus A system that integrates one or more AI components

A component that involves AI methods.

• component

ISTQB - CTAI Syllabus A component that provides AI functionality

A system that uses Artificial Intelligence and consists of several components (at least one of which is an AI component).

The AIA¹ compromise proposal adopted by the EU Parliament on 27 April 2023 for the trilogue (based on document version of 06 December 2022) defines an AI system as a system (in contrast to the OECD definition²: there as a ''machine-based system'') that makes predictions, recommendations, or can make decisions that influence the real or virtual environment for certain human-defined goals. AI systems can be endowed with varying degrees of autonomy.

• AI component

Software module in which AI methods are implemented.

Capability such as percept, process, act and communicate, which can be realised on the basis of AI methods (see DIN/DKE German Standardization Roadmap on AI, 2nd Edition, Sec. 4.1.2.1).

When using algorithmic and socio-technical systems in a broader sense and machine-learning systems in a narrower sense, fairness as an ethical principle describes the reproducible degree of equal treatment of different persons in all stages of the lifecycle of the system. This principle is also applicable to non-human things, such as animals, environment, nature,... or to natural actors as a whole.

Metamodel for assuring the quality of AI-based systems.

Note: This metamodel is defined in detail in DIN SPEC 92001.

ISO/IEC TR 29119-11:2020 metamodel intended to ensure the quality of AI-based systems (3.1.9)

Note 1 to entry: This metamodel is defined in detail in DIN SPEC 92001.

A software model for the provision of AI services.

ISTQB - CTAI Syllabus A software licensing and delivery model in which AI and AI development services are centrally hosted

Process of teaching an entity a set of knowledge, skills, processes and/or behaviours.

ETSI GR ENI 004 process of teaching an entity a set of knowledge, skills, processes and/or behaviours

ISO/IEC DIS 22989 process to establish or to improve the parameters of a machine learning model (3.2.11), based on a machine learning algorithm (3.2.10), by using training data (3.2.22).

Data that can be used to train a model.

• dataset

ISO/IEC DIS 22989 subset of input data samples used to train a machine learning model (3.2.11)

ISO/IEC TR 29119-11:2020 dataset used to train an ML model (3.1.46)

State of model after extracting knowledge from data in the course of supervised machine learning.

• training

ISO/IEC DIS 22989 result of model training (3.2.21)

The property of being accessible and usable by authorised persons on demand. Characterised by the degree, the extent of availability may depend for example on features such as actuality, interpretability as well as completeness of information.

ISO/IEC DIS 22989 property of being accessible and usable on demand by an authorized entity

[SOURCE: ISO/IEC 27000:2018, 3.7]

A research and application area concerned with understanding the factors that influence the outcomes of AI systems.

ISTQB - CTAI Syllabus The field of study related to understanding the factors that influence AI system outputs

Desired property of an AI system that describes the understandability to a human of the factors that influence automated decision making.

ISO/IEC DIS 22989 property of an AI system (3.1.4) to express important factors influencing the AI system (3.1.4) results in a way that humans can understand

Note 1 to entry: It is intended to answer the question “Why?” without actually attempting to argue that the course of action that was taken was necessarily optimal.

ISO/IEC TR 29119-11:2020 level of understanding how the AI-based system (3.1.9) came up with a given result

ISTQB - CTAI Syllabus The level of understanding how the AI-based system came up with a given result (ISO/IEC TR 29119-11)

An ML model that was already trained when it was obtained for further development / further training.

ISTQB - CTAI Syllabus An ML model already trained when it was obtained

Updating a trained model by training with different training data.

• training data

ISO/IEC DIS 22989 updating a trained model (3.2.20) by training (3.2.21) with different training data (3.2.22)

AI focused on a single well-defined task to address a specific problem.

• general AI

ISO/IEC DIS 22989 AI (3.1.13) focused on a single well-defined task to address a specific problem.

ISO/IEC TR 29119-11:2020 weak AI: AI (3.1.13) focused on a single well-defined task to address a specific problem

ISTQB - CTAI Syllabus AI focused on a single well-defined task to address a specific problem (ISO/IEC TR 29119-11)

Synonym: weak AI

AI that exhibits intelligent behaviour comparable to a human across the full range of cognitive abilities (synonym: strong AI).

• narrow AI

ISO/IEC DIS 22989 AI that addresses a broad range of tasks (3.1.37) with a satisfactory level of performance Note 1 to entry: Compared to narrow AI (3.1.27).

ISO/IEC TR 29119-11:2020 strong AI: AI (3.1.13) that exhibits intelligent behaviour comparable to a human across the full range of cognitive abilities

ISTQB - CTAI Syllabus AI that exhibits intelligent behaviour comparable to a human across the full range of cognitive abilities (ISO/IEC TR 29119-11) Synonym: strong AI

The process of applying the ML algorithm to the training dataset to create a ML model.

ISTQB - CTAI Syllabus The process of applying the ML algorithm to the training dataset to create an ML model

A Type of AI method based on models that uses symbols and structures to recognize patterns.

• subsymbolic AI

ISO/IEC DIS 22989 AI (3.1.2) based on techniques and models (3.1.26) using symbols and structures

Note 1 to entry: Compared to. subsymbolic AI (3.1.36).

Type of AI Methods based on models using a numeric representation and implicit information encoding.

• symbolic AI

ISO/IEC DIS 22989 AI (3.1.2) based on techniques and models (3.1.26) using a numeric representation and implicit information encoding

Note 1: Compared to symbolic AI (3.1.35).

The activation value indicates how strongly a neuron of an artificial neural network is activated by a concrete input. It is the transformation result by the activation function.

ISO/IEC TR 29119-11:2020 output of an activation function (3.1.3) of a node in a neural network

ISTQB - CTAI Syllabus The output of an activation function of a neuron in a neural network

Ability of a system to react to changes in its environment in order to continue meeting both functional and non-functional requirements.

ISO/IEC TR 29119-11:2020 ability of a system to react to changes in its environment in order to continue meeting both functional and non-functional requirements

Degree of temporal validity of data that is relevant in a specific application context.

A tree-like ML model whose nodes represent decisions and whose branches represent possible outcomes.

ISO/IEC DIS 22989 model (3.1.26) for which inference (3.1.22) is encoded as paths from the root to a leaf node in a tree structure

ISO/IEC TR 29119-11:2020 supervised-learning model (3.1.46) for which inference can be represented by traversing one or more tree-like structures

ISTQB - CTAI Syllabus A tree-like ML model whose nodes represent decisions, and whose branches represent possible outcomes

An iteration of ML training on the whole training dataset.

ISTQB - CTAI Syllabus An iteration of ML training on the whole training dataset

Incremental learning is a machine learning method that uses constantly changing input data to further train the model and continuously expand the knowledge of the existing model.

ETSI GR ENI 004 learning from a continuously changing source of data (e.g. streaming data) that arrives over time .

Knowledge that was created based on inference / reasoning, using evidence provided.

ETSI GR ENI 004 knowledge that was created based on reasoning, using evidence provided

The generation of an ML model that corresponds too closely to the training dataset, resulting in a model that finds it difficult to generalise to new data.

ISO/IEC TR 29119-11:2020 generation of a ML model (3.1.46) that corresponds too closely to the training data (3.1.80), resulting in a model that finds it difficult to generalize to new data

ISTQB - CTAI Syllabus The generation of an ML model that corresponds too closely to the training dataset, resulting in a model that finds it difficult to generalize to new data (After ISO/IEC TR 29119-11)

Creation of new data points based on an existing data set.

ISTQB - CTAI Syllabus The activity of creating new data points based on an existing dataset

A system that functions over long periods of time without human intervention.

ISO/IEC TR 29119-11:2020 system capable of working without human intervention for sustained periods

ISTQB - CTAI Syllabus A system capable of working without human intervention for sustained periods

The generation of an ML model that does not reflect the underlying trend of the training dataset, resulting in a model that has diffculties with accurate predictions.

ISO/IEC TR 29119-11:2020 generation of a ML model (3.1.46) that does not reflect the underlying trend of the training data (3.1.80), resulting in a model that finds it difficult to make accurate predictions (3.1.56)

ISTQB - CTAI Syllabus The generation of an ML model that does not reflect the underlying trend of the training dataset, resulting in a model that finds it difficult to make accurate predictions (ISO/IEC TR 29119-11)

The ability to track propositions or systems, for example by granting access to data, documents or (AI) systems.

In machine learning, hyperparameters usually refer to all parameters that are not directly defined or influenced by the training process. These include model parameters such as the number of layers of a neural network or the step size of the training process, but not the learned weights. Basically, hyperparameters can be differentiated into algorithmic and model-specific. Algorithmic hyperparameters influence the performance of the learning algorithm, Model-specific hyperparameters, on contrast, influence the mathematical or statistical model used in the learning process.

ETSI GR ENI 004 learning parameter that is set before the learning process is started

ISO/IEC DIS 22989 characteristic of a machine learning algorithm (3.2.10) that affects its learning process

Note 1 to entry: Hyperparameters are selected prior to training and can be used in processes to help estimate model parameters.

Note 2 to entry: Examples of hyperparameters include number of network layers, width of each layer, type of activation function, optimization method, learning rate for neural networks; the choice of kernel function in a support vector machine; number of leaves or depth of a tree; the K for K-means clustering; the maximum number of iterations of the expectation maximization algorithm; the number of Gaussians in a Gaussian mixture.

ISO/IEC TR 29119-11:2020 variable used to define the structure of a neural network and how it is trained

Note 1 to entry: Typically, hyperparameters are set by the developer of the model (3.1.46) and may also be referred to as a tuning parameter (3.1.53).

ISTQB - CTAI Syllabus A parameter used to either control the training of an ML model or to set the configuration of an ML model

A role, played by an external entity (human or machine), which interacts with the subject of a use case. An actor is always a type of stakeholder, but not vice versa.

ETSI GR ENI 004 role, played by an external entity (human or machine), which interacts with the subject of a use case

NOTE: An actor is always a type of stakeholder (but not vice versa). See stakeholder.

In the context of AI, an agent is a decision making and acting system that can interact with its environment and other agents.

ETSI GR ENI 004 computational process that implements the autonomous, communicating functionality of an application

ISO/IEC DIS 22989 automated entity that perceives its environment and takes actions to achieve its goals.

Note 1 to entry: An AI agent is an agent that maximizes its chance of successfully achieving its goals by using AI techniques.

Deviations in data characteristics from expected states.

ETSI GR ENI 004 measurable consequences of an unexpected change in state of a datum, or set of data, which is outside of its local or global norm.

Measure of the average information density of a variable, e.g. in data pre-processing.

Set of statements for explaining an observation that is not yet known to be true.

ETSI GR ENI 004 set of statements for explaining an observation that is not yet known to be true

Level of understanding how the underlying (AI) technology works.

ISO/IEC TR 29119-11:2020 level of understanding how the underlying (AI) technology works

ISTQB - CTAI Syllabus The level of understanding how the underlying AI technology works (ISO/IEC TR 29119-11)

In a narrower sense, supervised learning refers to machine learning methods that are trained with concretely specified target outputs (so-called ''labels''). In a broader sense, it includes methods whose learning goal is determined by concrete specifications, even if not at the level of individual outputs. This broader sense includes procedures such as GANs and reinforcement learning.

• unsupervised ML

ISO/IEC DIS 22989 machine learning (3.2.9) that makes use of labelled data during training (3.2.21)

On the one hand, ontology is a philosophical discipline that deals with classifying concepts of the world into category systems that are as meaningful as possible. On the other hand, in computer science, concrete category systems of this kind, for example consisting of concepts and relations for algorithmic use, are called ontologies as well.

ETSI GR ENI 004 explicit specification of a conceptualization

NOTE 1: As defined in [i.14].

NOTE 2: This definition is the basis for definitions in OneM2M and SmartM2M.

language, consisting of a vocabulary and a set of primitives, that enable the semantic characteristics of a domain to be modelled

Action of interacting work systems to produce a specific overall outcome.

cf.: DIN EN ISO 6385:2016

System that involves the interaction of individual or multiple workers with the work equipment to perform the function of a system, within the workspace as well as the work environment, under the conditions specified by the work tasks.

Usually, rule-based systems incorporating symbolic knowledge processing.

Example: If-then rules.

Note: E.g. symbolic, formal representation of knowledge in AI systems, with the ability to infer new knowledge from formal knowledge by means of reasoning based on logic.

ISO/IEC DIS 22989 AI system (3.1.4) that encapsulates knowledge provided by a human expert in a specific domain to infer solutions to problems

ISTQB - CTAI Syllabus An AI-based system for solving problems in a particular domain or application area by drawing inferences from a knowledge base developed from human expertise.

Principles that determine the moral behaviour of a human being or a machine (according to ETSI). Across domains, ethics is the scientific study of morality. Furthermore, ethics reflects and philosophises on various moral concepts, analyses and systematises, examines and questions their justifications and principles. There are various moral concepts, systems of norms, principles, values or dispositions, all of which claim to be the basis of a right action.

• virtue ethics
• ethisches Dilemma

ETSI GR ENI 004 set of principles that govern the moral behaviour of a person or machine

Constellation in which every possible decision leads to a violation of one or more ethical principles (e.g. trolley problem).

• ethics

ETSI GR ENI 004 situation in which any available decision leads to infringing on one or more ethical principles.

Autonomy is the absence of heteronomy. Without autonomy, rational behaviour is not possible. In relation to humans, autonomy means free will and corresponds to a basic principle of digital ethics.

ISO/IEC DIS 22989 autonomous characteristic of a system that is capable of modifying its operating domain or goal without external intervention, control or oversight.

Note 1 to entry: In jurisprudence, autonomy refers to the capacity for self-governance. In this sense, also, ''autonomous'' is a misnomer as applied to automated AI systems, because even the most advanced AI systems are not self-governing. Rather, AI systems operate based on algorithms and otherwise obey the commands of operators. For these reasons, this document does not use the popular term autonomous to describe automation.

ISO/IEC TR 29119-11:2020 ability of a system to work for sustained periods without human intervention

ISTQB - CTAI Syllabus The ability of a system to work for sustained periods without human intervention (ISO/IEC TR 29119-11)

A value that converts the result of a prediction function into a binary result of either above or below the value.

ISTQB - CTAI Syllabus A value that transforms the result of a prediction function into a binary outcome of either above or below the value Synonym: discrimination threshold

Ensemble learning is using multiple learning algorithms to obtain better performance in predicting results, than what is possible from using any single learning algorithm.

• random forest

ETSI GR ENI 004 use of multiple learning algorithms to obtain better performance in predicting results than is possible from using any single learning algorithm.

Part of an IoT system that interacts and communicates with the physical world by sensing or triggering.

Note: An IoT device can be both a sensor and an actuator (see also Internet of Things).

ISO/IEC DIS 22989 entity of an IoT system that interacts and communicates with the physical world through sensing or actuating

Note 1 to entry: An IoT device can be a sensor or an actuator.

[SOURCE: ISO/IEC 20924:2018, 3.2.4]

Learning algorithm that interactively presents labelling data to a user, so that the training process can be optimised by limiting it to selected entities.

ETSI GR ENI 004 learning algorithm that can query a user interactively to label data with the desired outputs

NOTE: The algorithm proactively selects the subset of examples to be labelled next from the pool of unlabelled data. The idea is that an ML algorithm could potentially reach a higher level of accuracy while using a smaller number of training labels if it were allowed to choose the data it wants to learn from.

The deliberate use of adversarial examples to cause errors in the output of a model. Especially artificial neural networks are prone to this kind of attacks.

ISO/IEC TR 29119-11:2020 deliberate use of adversarial examples (3.1.7) to cause a ML model (3.1.46) to fail

ISTQB - CTAI Syllabus The deliberate use of adversarial examples to cause an ML model to fail

A type of bias caused by a person favoring the recommendations of an automated decision-making system systematically over other sources.

ISTQB - CTAI Syllabus A type of bias caused by a person favoring the recommendations of an automated decision-making system over other sources

Synonym: complacency bias

In machine learning, labels or annotations are the parts of the training data set that indicate the desired ideal output of the model for a corresponding input datum for training purposes. In a broader sense, the actual outputs of a model in operation are also referred to in this way.

• ground truth
• data labelling

ETSI GR ENI 004 identification of an output value for a given input.

NOTE: Supervised learning uses labelled data; semi-supervised learning uses labels for a portion of the training data (the remaining training data are not labelled); unsupervised learning is based on training data that are not labelled.

ISO/IEC DIS 22989 the target variable assigned to a sample

Testing approach based on the attempted creation and execution of adversarial examples to identify defects in a model in order to increase the robustness and reduce the fault tolerance.

Note: Typically applied to neural networks.

ISO/IEC TR 29119-11:2020 testing approach based on the attempted creation and execution of adversarial examples (3.1.7) to identify defects in an ML model (3.1.46)

Note 1 to entry: Typically applied to ML models in the form of a neural network (3.1.48).

Search, selection or application of the best solution (with respect to a set of criteria) from a set of available alternatives.

ETSI GR ENI 004 set of mechanisms that select a best solution (with respect to a set of criteria) from a set of available alternatives

Scientific discipline concerned with the understanding of interactions among humans and other elements of a system. Moreover, ergonomics corresponds to a profession that applies theory, principles, data and methods to optimize the human well-being and overall system performance.

Note: This definition is consistent with that given by the International Ergonomics Association

cf.: DIN EN ISO 26800:2011

The Internet of Things (IoT) links a variety of diverse (edge) devices (see IoT device), central data platforms, connecting systems, services, people and information from the physical and virtual worlds. This has enabled the development of new business models in addition to new applications and services.

ISO/IEC DIS 22989 infrastructure of interconnected entities, people, systems and information resources together with services which process and react to information from the physical world and virtual world.

[SOURCE: ISO/IEC 20924:2018, 3.2.1, modified – ''…services which processes and reacts to…'' has been placed with ''…services which process and react to…'', and acronym has been moved to separate line.]

A type of logic based on the concept of real-valued, continuous truth values (within the closed interval $[0,1]$). Contrary to classical, binary logics, where only the truth values ${0, 1}$ apply, fuzzy logic allows to express uncertainties.

ETSI GR ENI 004 fuzzy logic: type of many-valued logic that allows a truth value to be any real number between 0 and 1 inclusive.

NOTE: Fuzzy logic is most often used to reason about the degree of truth, or probability, in a system.

ISTQB - CTAI Syllabus A type of logic based on the concept of partial truth represented by certainty factors between 0 and 1

Unsupervised learning refers to machine learning methods that learn a function without relying on concretely specified targets (e.g. ''labels''). There are different opinions as to the degree of concreteness of external targets that can no longer be referred to as unsupervised learning.

• supervised ML

ETSI GR ENI 004 learning a function that maps an input to an output without the benefit of the data being classified or labelled

ISO/IEC DIS 22989 machine learning (3.2.9) that makes use of unlabelled data during training (3.2.21)

ISO/IEC TR 29119-11:2020 task of learning a function that maps unlabelled input data to a latent representation

ISTQB - CTAI Syllabus Training an ML model from input data using an unlabeled dataset