• Any artificial system that performs tasks under varying and unpredictable circumstances without significant human oversight, or that can learn from experience and improve performance when exposed to data sets.
• An artificial system developed in computer software, physical hardware, or other context that solves tasks requiring human-like perception, cognition, planning, learning, communication, or physical action.
• An artificial system designed to think or act like a human, including cognitive architectures and neural networks.
• A set of techniques, including machine learning, that is designed to approximate a cognitive task.
• An artificial system designed to act rationally, including an intelligent software agent or embodied robot that achieves goals using perception, planning, reasoning, learning, communicating, decision making, and acting.

Allocated responsibility based on regulation, agreement, assignment. In an AI governance context, the obligation of an individual or organization to account for its activities, for completion of a deliverable or task, accept the responsibility for those activities, deliverables or tasks, and to disclose the results in a transparent manner.

Product of a set of informational and evaluative practices that can provide justified confidence that an AI system operates in context in a trustworthy fashion and as claimed. This definition draws from MITRE's use of the term "justified confidence" (from international software assurance standards) and the UK Centre for Data Ethics and Innovation usage in its "roadmap to an effective AI assurance ecosystem."

Set of computational rules to be followed to solve a mathematical problem or analytical process.

Association is used to discover interesting relationships between variables in a dataset.

Systematic error in an AI model due to issues with the training data or flawed assumptions.

Conversational agent that dialogues with its user (for example: empathic robots available to patients, or automated conversation services in customer relations)

Data mining technique which groups unlabeled data based on their similarities or differences

A field of AI focused on enabling machines to interpret and understand visual information from the world.

Type of supervised ML application that is used to extract and classify important pieces of information from large volumes of data—including context, emotion, and intent. It can be useful for gaining an understanding of customer interactions and can be used to improve customer experience.

Process of applying graphical, statistical, or quantitative techniques to a set of observations or measurements in order to summarize it or to find general patterns.

Set of processes that ensures that data assets are formally managed throughout the enterprise. Establishes authority and management and decision making parameters related to the data produced or managed by an organization.

Cross-functional discipline that combines elements of computer science, mathematics, statistics, and subject-matter expertise in order to produce data-driven insights and processes that can help solve business, operational, and strategic problems for different kinds of organizations

Broad set of techniques for machine learning in which hypotheses take the form of complex algebraic circuits with tunable connection strengths. The word "deep" refers to the fact that the circuits are typically organized into many layers, which means that computation paths from inputs to outputs have many steps. Deep learning is currently the most widely used approach for applications such as visual object recognition, machine translation, speech recognition, speech synthesis, and image synthesis; it also plays a significant role in reinforcement learning applications.

Modelling [that] produces consistent outcomes for a given set of inputs, regardless of how many times the model is recalculated. The mathematical characteristics are known in this case. None of them is random, and each problem has just one set of specified values as well as one answer or solution. The unknown components in a deterministic model are external to the model. It deals with the definitive outcomes as opposed to random results and doesn't make allowances for error.

Dimensionality reduction is used to reduce the number of dimensions while still maintaining meaningful properties close to the original data.

Representation of the mechanisms underlying AI systems' operation.

Class of models, often transformers trained by self-supervision on large-scale broad data, that can be easily adapted to perform a wide range of downstream tasks. Best-known examples are Large Language Models (LLMs), which focus on language-specific systems, but the term extends to models for all modalities of data and knowledge (such as images, videos, and audios).

Production of confidently stated but erroneous or false content by AI systems, which can mislead users. May also be called confabulation.

Type of supervised ML that applies computer vision techniques that are used to detect instances of objects of a certain type of classification such as a car or pedestrian

Class of language models that use deep-learning algorithms and are trained on extremely large textual datasets that can be multiple terabytes in size. LLMs can be classed into two types: generative or discriminatory. Generative LLMs are models that output text, such as the answer to a question or even writing an essay on a specific topic. They are typically unsupervised or semi-supervised learning models that predict what the response is for a given task. Discriminatory LLMs are supervised learning models that usually focus on classifying text, such as determining whether a text was made by a human or AI.

Field and practice of using algorithms to "learn" by extracting patterns from a large body of data. Building a machine learning model is an iterative approach to problem solving and has an adaptive approach that looks over a large body of all possible outcomes and chooses the result that best satisfies its objective function.

A program or mathematical structure trained to make predictions or decisions based on data.

Phase in the data science development lifecycle where practitioners try to fit the best combination of weights and bias to a machine learning algorithm to minimize a loss function over the prediction range

AI for particular tasks (e.g., speech or facial recognition). Human-level AI seeks broadly intelligent, context-aware machines. It is needed for effective, adaptable social chatbots or human-robot interaction.

Field concerned with machines capable of processing, analyzing, and generating human language, either spoken, written or signed.

A model that, taking inspiration from the brain, is composed of layers (at least one of which is hidden) consisting of simple connected units or neurons followed by nonlinearities

When an AI model learns noise in its training data too well and performs poorly on new data.

Numbers inside an AI model that determine how an input (e.g. a chunk of prompt text) is converted into an output (e.g. the next word after the prompt). Process of "training" an AI model consists of using mathematical optimization techniques to tweak the model's parameter values over and over again until the model is very good at converting inputs to outputs.

Type of supervised ML application that is used to provide deep insights into various data points and allows for the anticipation of results based on given output variables. Examples of predictive analytics include credit scoring to predict likelihood of paying on time based on factors including customer data and credit history.

Practice of crafting effective inputs (prompts) to guide the outputs of generative AI systems.

Behavioral machine learning model that is similar to supervised learning, but the algorithm isn't trained using sample data. This model learns as it goes by using trial and error. A sequence of successful outcomes will be reinforced to develop the best recommendation for a given problem.

Composite measure of an event's probability of occurring and the magnitude or degree of the consequences of the corresponding event. The impacts, or consequences, of AI systems can be positive, negative, or both and can result in opportunities or threats

An organization's readiness to bear the risk in order to achieve its objectives. Risk tolerance can be influenced by legal, regulatory, budget and other factors.

Preconfigured software instance that uses business rules and predefined activity choreography to complete the autonomous execution of a combination of processes, activities, transactions, and tasks in one or more unrelated software systems to deliver a result or service with human exception management.

Also known as supervised machine learning, this type of machine learning uses labeled datasets to train algorithms to classify data or predict outcomes accurately.

• Input data is fed into the model.
• Weights are adjusted until the model has been appropriately fitted, i.e. generalizes and adequately represents the pattern.
• A training dataset is used to teach models to yield the desired output and includes inputs and outputs that are correctly categorized or "labeled", which allow the model to learn over time. The algorithm measures its accuracy through the loss function, adjusting until the error has been sufficiently minimized.

Dataset(s) used to teach an AI model to make predictions or decisions

Reflects the extent to which information about an AI system and its outputs is available to individuals interacting with such a system – regardless of whether they are even aware that they are doing so.

Often used in data exploration before a learning goal is established. Unsupervised machine learning uses unlabeled data. From that data, it discovers patterns that help solve clustering or association problems. It's useful when subject matter experts are unsure of common properties of a data set. Unsupervised learning models are utilized for three main tasks—clustering, association, and dimensionality reduction.