Skip to content

Artificial intelligence (AI) governance glossary

Organizations are under intense pressure to leverage AI as a competitive advantage. However, using AI without proper governance or visibility and control of the underlying data can lead to model bias, inaccuracies, legal and ethical implications, trust issues and more. This glossary will help you understand words, terms, and phrases you’ll come across as you explore AI and AI governance.


AI governance is the application of rules, processes, and responsibilities to drive maximum value from your automated data products by ensuring applicable, streamlined, and ethical AI practices that mitigate risk, adhere to legal requirements, and protect privacy.

The simulation of human intelligence in machines that are programmed to think and learn like humans. AI can take on tasks that were previously done by people including problem solving repetitive tasks.

How AI models will be used to solve a specific problem. Examples include chatbots, fraud detection, and personalized product recommendations.

A member-based foundation that aims to harness the collective power and contributions of the global open-source community to develop AI testing tools to enable responsible AI. The Foundation promotes best practices and standards for AI.

Learn more about AI Verify Foundation

Refers to AI systems with human-level cognitive abilities, capable of understanding, learning, and applying knowledge across various domains autonomously.

Clearly defined roles and responsibilities for all stakeholders involved in the AI lifecycle, including developers, data scientists, business users, legal and privacy professionals, to report, explain, or justify AI model output.

A methodology or collection of directives and regulations crafted to execute a defined task or address a specific issue, employing computational resources.

An official, independent examination and verification to ensure that AI and the data that drives it meet specific criteria set forth by governing or regulatory entities.


Bias in AI is the presence of discriminatory outcomes/outputs. Bias can arise from various sources including incomplete data, improper use-case, or flaws in the AI algorithm.


A program designed to simulate human conversation through text or voice interactions, often in use cases involving customer service.

A field of AI where AI models utilize supervised learning to bucket data, images, or other data inputs into a predefined number of classes/categories.

A field of AI focused on enabling computers to interpret visual data from images or videos.

A vast dataset utilized by AI models to discern patterns, forecast outcomes, or generate specific results. It may comprise some combination of structured and unstructured data, addressing singular or diverse subjects.


Data quality determines if data is fit for use to deliver high quality AI model outputs. Data is considered high quality based on consistency, uniqueness, completeness and validity. High quality data must have standardized data entries, data sets that do not contain repeated or irrelevant entries, be representative of real-world conditions, and conform to the formatting required by the business.

A not for profit member organization that brings together leading businesses and institutions across multiple industries to learn, develop, and adopt responsible data and AI practices.

Learn more about Data & Trust Alliance

The practice of managing, organizing, and manipulating data to extract meaningful insights and support decision-making. Within AI, DataOps is used to manage and support the full lifecycle of AI development and deployment.

Synthetic media created using AI, manipulating videos or images to replace one person's likeness with another, often raising concerns about misinformation and deception.


The guideline and principles that govern the ethical and moral implications of AI development and deployment. Ethical AI ensures AI benefits while minimizing potential risks and ethical dilemmas.

Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence: An Executive order that aims to strengthen AI practices and mitigate risks across the US federal government and private industry, establish standard practices, and enhance information sharing.

Learn more about Executive Order 14110 (US White House)

The ability to detail outputs of AI Models, ensuring they are understood, to the best of their ability, by technical and non-technical audiences alike.

European Union law that provides AI developers with clear requirements and obligations regarding specific uses of AI, helping them build trustworthy AI use cases while mitigating risks to EU citizens and organizations. The Act was passed on March 13, 2024.

Learn more about EU AI Act


Ensuring unbiased, equitable, and accurate outcomes across diverse groups and individuals, mitigating bias and discrimination.

Process for determining the most informative attributes of a dataset for use in a machine learning model.

Fine-tuning is the process of adjusting and optimizing a pre-trained large language model on a specific dataset, set of questions & answers, or tasking to improve its performance. It involves tuning the model outputs with the new data, allowing it to adapt and specialize for the desired task or domain.

A core large language model (LLM) framework, trained on large and diverse data sets, used as the basis for developing more complex or specialized models in machine learning (ML) and artificial intelligence (AI).


Artificial intelligence capable of producing previously unseen text, images, or other media in response to user prompts.


Misleading content created by generative AI that conflicts with source material or generates factually inaccurate output while maintaining the appearance of accuracy.


The process of developing insights or conclusions from available information by analyzing patterns and relationships within data.


A prompt intentionally designed to avoid ethical guidelines or restrictions within a large language model (LLM).


A specific type of artificial AI model that has been trained on large amounts of text to interpret, understand, and generate human-like language outputs. Examples include OpenAI ChatGPT and Google Gemini.


A subset of AI where algorithms learn patterns and make predictions from data without explicit programming, improving performance over time.

A collection of tools, technologies, and best practices to build, deploy, monitor and manage machine learning and AI models. It is the key capability for scaling and governing AI at the enterprise level.

MLOps stands for Machine Learning Operations. MLOps is a core function of Machine Learning Engineering, focused on streamlining the process of taking machine learning and AI models from ideation to production, and then maintaining and monitoring them. End-to-end MLOps systems also typically incorporate data layer connectivity and feature curation. MLOps is a collaborative function, often comprising data scientists, devops engineers, and IT.


A field of AI where models understand, interpret, and generate human language.

A model inspired by the human brain, comprising interconnected nodes (neurons) that process information to perform tasks like pattern recognition, text generation, or prediction. Not included in this glossary are individual components of a neural network, such as activation functions, layer types, and cost functions.

US Government agency dedicated to promoting U.S. innovation and industrial competitiveness by advancing measurement science, standards, and technology in ways that enhance economic security and improve our quality of life. NIST has developed an AI Risk Framework to help organizations better manage AI risks.

Learn more about National Institute of Standards and Technology (NIST)


Ensuring data used in AI is handled in accordance with all applicable data privacy laws and regulations to prevent unauthorized access, use, or disclosure of sensitive information.

The input provided to a model or system to initiate a specific task or query. It can take various forms, such as a question, a statement, or a command, and is designed to elicit a desired response from the AI system.

A technique used in artificial intelligence, especially in natural language processing, where the user optimizes a prompt to a Generative AI model via commands, examples, or other techniques, to elicit a desired response


A field of AI where models utilize supervised learning to predict a continuous outcome from data, images, or other data inputs.

Ensuring that AI models adhere to relevant laws and regulations including data protection laws, anti-discrimination laws, and industry-specific regulations like HIPAA.

A global and member-driven non-profit dedicated to enabling successful responsible AI efforts in organizations.

Learn more about Responsible AI Institute

The combination of a retrieval based method with an LLM, where supporting documents are embedded into a vector store, retrieved when a user queries the model, and included as additional context during prompting to an LLM

The ability of AI models to consistently produce accurate and trustworthy results, without unexpected errors or biases, across various tasks and conditions.


Implementing measures to protect AI models from internal and external threats as well as ensuring the confidentiality of the data used by AI models.

A Canadian entity focused on the standardization of guidelines, processes, and characteristics in a wide range of fields, including AI. In 2023, the SCC developed the AI and Data Governance Standardization Collaborative (AIDG) to expand on data governance standardization and include new AI issues.

Learn more about Standards Council of Canada (SCC)

Artificial data that mimics real data which is often used for training machine learning (ML) models when there isn’t sufficient real data or it is highly sensitive.


A portion of a dataset that is used to train an AI model until it can proficiently predict results or uncover patterns.

A portion of a dataset that is not used to train an AI model but rather measure performance of an AI model after training is completed.

Ensuring that AI models, including algorithms and decision-making processes, are open and understandable to relevant stakeholders, such as developers, data scientists, business users, legal and privacy professionals.

Data that is high-quality, secure, understood, accurate, and easily accessible by the proper individuals to confidently create and deploy valuable, ethical, responsible, and compliant AI models.

Often used interchangeably with Ethical AI, Trustworthy AI is the development of AI models that are reliable, ethical, and can be trusted by users.


A portion of a dataset that is not used to train an AI model but rather measure performance of an AI model during training.

A vector database is a database designed to store and organize unstructured data, like text, images, or audio, using vector embeddings (high-dimensional vectors). This structure enables rapid retrieval of similar items, streamlining search and retrieval processes.

A Canadian government backed AI Institute that empowers researchers, businesses and governments, to develop and adopt AI responsibly.

Learn more about Vector Institute
Back to top