Definitions

Question 1

Accountability

Answer 1

The obligation and responsibility of the developers, deployers, and distributors of an AI system to ensure the system operates in a manner that is ethical, fair, transparent and compliant with applicable rules and regulations (see also fairness and transparency). Accountability ensures the actions, decisions and outcomes of an AI system can be traced back to the entity responsible for it.

Question 2

Active learning

Answer 2

A subfield of AI and machine learning in which an algorithm selects some of the data it learns from. Instead of learning from all the data it is given, an active learning model requests additional data points that will help it learn the best.
→ Also called query learning.

Question 3

Adaptive learning

Answer 3

A method that adjusts and tailors educational content to the specific needs, abilities and learning pace of individual students. The purpose of adaptive learning is to provide a personalized and optimized learning experience, catering to the diverse learning styles of students.

Question 4

Adversarial attack

Answer 4

A safety and security risk to the AI model that can be instigated by manipulating the model, such as by introducing malicious or deceptive input data. Such attacks can cause the model to malfunction and generate incorrect or unsafe outputs, which can have significant impacts. For example, manipulating the inputs of a self-driving car may fool the model to perceive a red light as a green one, adversely impacting road safety.

Question 5

Agentic AI

Answer 5

Agentic AI are AI systems that are designed to autonomously make and act upon decisions with limited human oversight. Agentic AI is often referred to as being able to pursue complex goals with limited supervision.
→ Also known as AI agents.

Question 6

AI assurance

Answer 6

A combination of frameworks, policies, processes and controls that measure, evaluate and promote safe, reliable and trustworthy AI. AI assurance schemes may include conformity, impact and risk assessments, AI audits, certifications, testing and evaluation, and compliance with relevant standards.

Question 7

AI audit

Answer 7

A systematic review of an AI system or a portfolio of AI systems to ensure it operates as intended and complies with relevant laws, regulations and standards. An AI audit can help identify and map unforeseen risks during the impact assessment and development phases.

Question 8

AI governance

Answer 8

A system of frameworks, practices and processes at an organizational level. AI governance helps various stakeholders implement, manage and oversee the use of AI technology. It also helps manage associated risks to ensure AI aligns with stakeholders’ objectives, is developed and used responsibly and ethically, and complies with applicable requirements.

Question 9

Algorithm

Answer 9

A computational procedure or set of instructions and rules designed to perform a specific task, solve a particular problem or produce a machine learning or AI model.

Question 10

Artificial intelligence

Answer 10

Artificial intelligence is a broad term used to describe an engineered system that uses various computational techniques to perform or automate tasks. This may include techniques, such as machine learning, in which machines learn from experience, adjusting to new input data and potentially performing tasks previously done by humans. More specifically, it is a field of computer science dedicated to simulating intelligent behavior
in computers. It may include automated decision-making.
→ Acronym: AI

Question 11

Automated
decision-making

Answer 11

The process of making a decision by technological means without human involvement.

Question 12

Autonomy

Answer 12

The ability of an AI system to operate independent of human intervention.

Question 13

Bias

Answer 13

There are several types of bias within the AI field. Computational bias or machine bias is a systematic error or deviation from the true value of a prediction that originates from a model’s assumptions or the data itself (see also input data).
Cognitive bias refers to inaccurate individual judgment or distorted thinking, while societal bias leads to systemic prejudice, favoritism and/or discrimination in favor of or against an individual or group. Either or both may permeate the model or the system in numerous
ways, such as through selection bias, i.e. biases in selecting data for model

Question 14

Bootstrap aggregating

Answer 14

A machine learning method that aggregates multiple versions of a model (see also machine learning model) trained on random subsets of a dataset. This method aims to make a model more stable and accurate.
→ Sometimes referred to as bagging.

Question 15

Chatbot

Answer 15

An AI system designed to simulate human-like conversations and interactions. Modern versions use natural language processing and large language models to understand and respond to text or other media. Because chatbots are often used for customer service and other personal help applications, chatbots often ingest users’ personal information.

Question 16

Classification model

Answer 16

A type of model (see also machine learning model) used in machine learning that is
designed to take input data and sort it into different categories or classes.
→ Also called classifiers.

Question 17

Clustering

Answer 17

An unsupervised machine learning method where patterns in the data are identified and evaluated, and data points are grouped accordingly into clusters based on their similarity.
→ Also called clustering algorithms.

Question 18

Compute

Answer 18

The processing resources that are available to a computer system. This includes the
hardware components such as the central processing unit or graphics processing
unit. Compute is essential for memory, storage, processing data, running applications, rendering graphics for visual media, powering cloud computing, among others.

Question 19

Computer vision

Answer 19

A field of AI that uses computers to process and analyze images, videos and other visual inputs. Common applications of computer vision include facial recognition, object recognition and medical imaging.

Question 20

Conformity assessment

Answer 20

An analysis, often performed by an entity independent of a model developer, on an AI system to determine whether requirements, such as establishing a risk management system, data governance, record-keeping, transparency and cybersecurity practices have been met.

Question 21

Contestability

Answer 21

The principle of ensuring AI systems and their decision-making processes can be
questioned or challenged by humans. This ability to contest or challenge the outcomes, outputs and actions of AI systems depends on transparency and helps promote accountability within AI governance.
→ Also called redress.

Question 22

Corpus

Answer 22

A large collection of texts or data that a computer uses to find patterns, make predictions or generate specific outcomes. The corpus may include structured or unstructured data and cover a specific topic or a variety of topics.

Question 23

Counterfactual

Answer 23

A hypothetical alternative in which an input variable is changed. A counterfactual is often used to play out what-if scenarios.

Question 23

Data drift

Answer 23

The shift that occurs when a dataset being used as an input for an AI system changes
over time so that it does not statistically resemble the dataset the model was trained on. This could lead to reduced performance as the model is not compatible with a different
dataset.

Question 24

Data leak

Answer 24

An accidental exposure of sensitive, personal, confidential or proprietary data. This can be a result of poor security defenses, human error, storage misconfigurations or a lack of robust policies around internal and external data sharing practices.

Question 24

Data poisoning

Answer 24

An adversarial attack in which a malicious user injects false data into a model to manipulate the training process, thereby corrupting the learning algorithm. The goal is to introduce intentional errors into the training dataset, leading to compromised performance that results in undesired, misleading or harmful outputs.

Question 25

Data provenance

Answer 25

A process that tracks and logs the history and origin of records in a dataset, encompassing the entire life cycle from its creation and collection to its transformation to its current state. It includes information about sources, processes, actors and methods used to ensure data integrity and quality. Data provenance is essential for data transparency and governance, and it promotes better understanding of the data and eventually the entire AI system.

Question 26

Data quality

Answer 26

The measure of how well a dataset meets the specific requirements and expectations for its intended use. Data quality directly impacts the quality of AI outputs and the performance of an AI system. High-quality data is accurate, complete, valid, consistent, timely and fit for purpose.

Question 27

Decision tree

Answer 27

A type of supervised learning model used in machine learning (see also machine learning model) that represents decisions and their potential consequences in a branching structure.

Question 28

Deep learning

Answer 28

A subfield of AI and machine learning that uses artificial neural networks. Deep learning is especially useful in fields where raw data needs to be processed, like image recognition, natural language processing and speech recognition.

Question 29

Deepfakes

Answer 29

Audio or visual content that has been altered or manipulated using generative artificial intelligence. Deepfakes can be used to spread misinformation and disinformation.

Question 30

Diffusion model

Answer 30

A generative model used in image generation that works by iteratively refining a noise signal to transform it into a realistic image when prompted.

Question 31

Discriminative model

Answer 31

A type of model (see also machine learning model) used in machine learning that directly maps input features to class labels and analyzes for patterns that can help distinguish between different classes. It is often used for text classification tasks, like identifying the language of a piece of text. Examples are traditional neural networks, decision trees and random forest.

Question 32

Disinformation

Answer 32

Audio or visual content, information and synthetic data that is intentionally manipulated or created to cause harm. Disinformation can spread through deepfakes by bad actors.

Question 33

Entropy

Answer 33

The measure of unpredictability or randomness in a set of data used in machine learning. A higher entropy signifies greater uncertainty in predicting outcomes.

Question 34

Expert system

Answer 34

A form of AI that draws inferences from a knowledge base to replicate the decision-making abilities of a human expert within a specific field, like a medical diagnosis. This term is often used to describe the opposite of artificial general intelligence.

Question 35

Explainability

Answer 35

The ability to describe or provide sufficient information about how an AI system generates a specific output or arrives at a decision in a specific context to a predetermined addressee. Explainability is important maintaining transparency and trust in AI. → Acronym: XAI

Question 36

Exploratory data analysis

Answer 36

Data discovery process techniques that take place before training a machine learning model to gain preliminary insights into a dataset, such as identifying patterns, outliers and anomalies and finding relationships among variables.

Question 37

Fail-safe plans

Answer 37

A back-up plan to be executed in case an AI system behaves in unexpected or dangerous ways. A fail-safe plan can be part of a technical redundancy solution to boost robustness.

Question 38

Fairness

Answer 38

An attribute of an AI system that prioritizes relatively equal treatment of individuals or groups in its decisions and actions in a consistent, accurate and measurable manner. Every model must identify the appropriate standard of fairness that best applies, but most often it means the AI system's decisions should not adversely impact, whether directly or disparately, sensitive attributes like race, gender or religion.

Question 39

Federated learning

Answer 39

A machine learning method that allows models (see also machine learning model) to be trained on the local data of multiple edge devices or servers. Only the updates of the local model, not the training data itself, are sent to a central location where they are aggregated into a global model — a process that is iterated until the global model is fully trained. This process enables better privacy and security controls for individual user data.

Question 40

Fine-tuning

Answer 40

The process of taking a foundation model and training it further for a specialized task through supervised learning. It involves taking a foundation model that has already learned general patterns from a large dataset and training it for a specific task using a much smaller and labeled dataset. → See foundation model.

Question 41

Foundation model

Answer 41

A large-scale model that has been trained on extensive and diverse datasets to enable broad capabilities, such as language (see also large language model), vision, robotics, reasoning, search or human interaction, that can function as the base for use-specific applications. → Also called general purpose AI model and frontier AI.

Question 42

Generalization

Answer 42

The ability of a model (see also machine learning model) to understand the underlying patterns and trends in its training data and apply what it has learned to make predictions or decisions about new, unseen data.

Question 43

Generative AI

Answer 43

A field of AI that uses deep learning trained on large datasets to create new content, such as written text, code, images, music, simulations and videos. Unlike discriminative models, generative AI makes predictions on existing data rather than new data.

Question 44

Greedy algorithm

Answer 44

A type of algorithm that makes the optimal choice to achieve an immediate objective at a particular step or decision point based on the available information and without regard for the long-term optimal solution.

Question 45

Ground truth

Answer 45

The absolute or objectively known state of a dataset against which the quality of an AI system can be evaluated. It serves as the real-world reference against which the outputs are measured for accuracy and reliability.

Question 46

Hallucinations

Answer 46

Instances in which generative AI models create seemingly plausible but factually incorrect outputs under the appearance of fact. → Also known as confabulations.

Question 47

Human-centric AI

Answer 47

An approach to AI design, development and deployment that prioritizes human well-being, autonomy, values and needs. The goal is to develop AI systems that amplify and augment human abilities rather than undermine them.

Question 48

Human-in-the-loop

Answer 48

A design paradigm that incorporates human oversight, intervention, interaction or control over the operation and decision-making processes of an AI system. → Acronym: HITL

Question 49

Impact assessment

Answer 49

An evaluation process designed to identify, understand and mitigate the potential ethical, legal, economic and societal implications of an AI system. → Also referred to as a risk assessment.

Question 50

Inference

Answer 50

A type of machine learning process in which a trained model (see also machine learning model) is used to make predictions or decisions based on input data.

Question 51

Input data

Answer 51

Data provided to or directly acquired by a learning algorithm or model (see also machine learning model) for the purpose of producing an output. It forms the basis for machine learning models to learn, make predictions and carry out tasks.

Question 52

Interpretability

Answer 52

The ability to explain or present a model's reasoning in human-understandable terms. Unlike explainability, which provides an explanation after a decision is made, interpretability emphasizes designing models that inherently facilitate understanding through their structure, features or algorithms. Interpretable models are domain-specific and require significant domain expertise to develop.

Question 53

Large language model

Answer 53

A form of AI that utilizes deep learning algorithms to create models (see also machine learning model, foundation model and fine-tuning) pretrained on massive text datasets for the general purpose of analyzing and learning patterns and relationships among characters, words and phrases to perform text-based tasks. There are generally two types of LLMs: generative models that make text predictions based on the probabilities of word sequences learned from its training data (see also generative AI) and discriminative models that make classification predictions based on probabilities of data features and weights learned from its training data (see also discriminative model). The word large generally refers to the model's capacity measured by the number of parameters and to the enormous datasets it is trained on. → Acronym: LLM

Question 54

Machine learning

Answer 54

A subfield of AI involving algorithms that iteratively learn from and then make decisions, recommendations, inferences or predictions based on input data. These algorithms build a model from training data to perform a specific task on new data without being explicitly programmed to do so. Machine learning implements various algorithms that learn and improve by experience in a problem-solving process that includes data collection and preparation, feature engineering, training, testing and validation. Companies and government agencies deploy machine learning algorithms for tasks such as fraud detection, recommender systems, customer inquiries, health care, and transportation and logistics. → Acronym: ML

Question 55

Machine learning model

Answer 55

A learned representation of underlying patterns and relationships in data, created by applying an AI algorithm to a training dataset. The model can then be used to make predictions or perform tasks on new, unseen data.

Question 56

Misinformation

Answer 56

False audio or visual content that is unintentionally misleading. It can be spread through deepfakes created by those who lack intent to cause harm.

Question 57

Model card

Answer 57

A brief document that discloses information about an AI model, like explanations about intended use, performance metrics and benchmarked evaluation in various conditions, such as across different cultures, demographics or race (see also system cards).

Question 58

Multimodal models

Answer 58

A type of model used in machine learning (see also machine learning model) that can process more than one type of input or output data, or "modality," at the same time. For example, a multimodal model can take both an image and text caption as input and then produce a unimodal output in the form of a score indicating how well the text caption describes the image. These models are highly versatile and useful in a variety of tasks, like image captioning and speech recognition.

Question 59

Natural language processing

Answer 59

A subfield of AI that helps computers understand, interpret and apply human language by transforming information into content. It enables machines to translate languages, read text or spoken language, interpret its meaning, measure sentiment, and determine which parts are important for understanding.

Question 60

Neural networks

Answer 60

A type of model (see also machine learning model) used in deep learning that mimics the way neurons in the human brain interact with multiple processing layers, including at least one hidden layer. This layered approach enables machine-based neural networks to model complex nonlinear relationships and patterns within data. Artificial neural networks have a range of applications, such as image recognition and medical diagnoses.

Question 61

Open-source software

Answer 61

A development model that provides free and open access to source code to the public, which can then be viewed, modified and redistributed according to the terms of its respective license. The goal is to promote innovation, transparency, shared collaboration and learning among researchers and technical experts. Open source models are often more lightly regulated.

Question 62

Overfitting

Answer 62

A concept in machine learning in which a model (see also machine learning model) becomes too specific to the training data and cannot generalize to unseen data, which means it can fail to make accurate predictions on new datasets.

Question 63

Oversight

Answer 63

The process of effectively monitoring and supervising an AI system to minimize risks, ensure regulatory compliance and uphold responsible practices. Oversight is important for effective AI governance, and mechanisms may include certification processes, conformity assessments and regulatory authorities responsible for enforcement.

Question 64

Parameters

Answer 64

The internal variables an algorithmic model learns from the training data. They are values the model adjusts to during the training process so it can make predictions on new data. Parameters are specific to the architecture of the model. For example, in neural networks, parameters are the weights of each neuron in the network.

Question 65

Post processing

Answer 65

Steps performed after a machine learning model has been run to adjust its output. This can include adjusting a model's outputs or using a holdout dataset — data not used in the training of the model — to create a function run on the model's predictions to improve fairness or meet business requirements.