AI Glossary

Accessibility means designing tools, content, services, and learning experiences so people with disabilities can use them. In AI, accessibility includes considering whether AI tools work with assistive technologies, produce accessible outputs, and support users with different needs.

Adaptive learning refers to digital learning systems that adjust content, pacing, difficulty, or feedback based on a learner’s progress or responses. AI can support adaptive learning by identifying patterns and suggesting next steps for practice or support.

An AI agent is a system that can carry out tasks with some level of independence. Agents may be able to use tools, follow instructions, retrieve information, complete steps in a workflow, or take actions on behalf of a user.

Example: An AI agent might help summarize documents, draft a response, and organize information into a report.

Agentic AI refers to AI systems that can take actions or complete tasks with a higher degree of independence. These systems may plan steps, use tools, retrieve information, or complete parts of a workflow based on a user’s goal.

Agentic AI should be used carefully, especially when actions affect people, systems, data, or institutional processes.

AI alignment refers to efforts to ensure that an AI system behaves in ways that are consistent with human goals, values, and expectations. In practice, this means designing, testing, and using AI tools so they support safe, fair, and responsible outcomes.

AI ethics refers to the responsible use, design, and governance of AI. It includes issues such as fairness, privacy, accountability, transparency, accessibility, safety, bias, environmental impact, and social impact.

AI literacy is the ability to understand, evaluate, and use AI tools responsibly. It includes knowing what AI can and cannot do, how to assess outputs, how to protect information, and how to think critically about ethical, social, and practical impacts.

An AI model is a system trained on data to recognize patterns, make predictions, classify information, or generate outputs. Models are the underlying systems that power tools such as chatbots, image generators, and recommendation systems.

AI safety refers to efforts to reduce risks from AI systems and ensure they operate in ways that are reliable, secure, and appropriate for their intended use.

AI-assisted writing refers to using AI to support writing tasks. This may include brainstorming, drafting, revising, summarizing, editing, or changing tone. Users remain responsible for reviewing, revising, and verifying AI-assisted writing.

An algorithm is a set of instructions or rules that a computer follows to complete a task or solve a problem. In AI, algorithms are used to process data, identify patterns, make predictions, or generate outputs.

Algorithmic bias occurs when an AI system produces unfair, inaccurate, or discriminatory outcomes because of the data, assumptions, design choices, or processes used to build it.

Anthropomorphism means treating AI as if it has human qualities, emotions, intentions, or understanding. AI tools may sound conversational or confident, but they do not think, feel, understand, or care in the way people do.

An application programming interface, or API, allows different software systems to communicate with each other. APIs can be used to connect AI tools with other applications, databases, or services.

Artificial general intelligence, or AGI, refers to a theoretical form of AI that could perform a wide range of intellectual tasks at a human level. AGI does not currently exist.

Artificial intelligence, or AI, refers to computer systems designed to perform tasks that typically require human intelligence. These tasks may include learning, recognizing patterns, understanding language, generating content, making predictions, or supporting decision-making.

Automation means using technology to complete tasks with limited human involvement. AI can support automation by helping systems classify information, generate responses, route requests, or complete multi-step workflows.

Bias in AI occurs when a system produces outputs that reflect unfair assumptions, stereotypes, or imbalances in the data or design. Bias can affect how AI tools respond to people, topics, communities, or situations.

A black box is an AI system whose internal decision-making process is difficult to understand. Users can see the input and output, but not always how the system reached its answer.

A chatbot is a tool that simulates conversation with a user through text or voice. AI chatbots can answer questions, draft content, summarize information, and support other tasks.

ChatGPT is an AI chatbot developed by OpenAI. It can generate text, summarize information, analyze files, answer questions, create images, support coding, and assist with a wide range of tasks depending on the version and available features.

Citation is the practice of acknowledging sources used in academic, professional, or creative work. When AI is used, citation or acknowledgement expectations may depend on the course, assignment, publication, style guide, or context.

Computer vision is a field of AI that allows systems to interpret images or video. It can be used for tasks such as recognizing objects, reading visual information, or analyzing images.

A context window is the amount of information an AI model can consider at one time during an interaction. A larger context window allows a tool to process more text, longer conversations, or larger documents.

A copilot is an AI assistant designed to work alongside a person. It can support tasks such as writing, summarizing, searching, planning, coding, and organizing information. A copilot supports human work but does not replace human judgment.

Copyright is the legal protection given to original works such as text, images, music, videos, code, and other creative materials. AI use can raise copyright questions when users upload, generate, adapt, or share content.

A custom GPT is a tailored version of ChatGPT designed for a specific purpose. It can include instructions, selected capabilities, and reference materials to support a defined task, audience, or workflow.

Example: A team might create a custom GPT to help users navigate a set of approved resources or draft content in a consistent format.

Data is information that can be collected, stored, processed, or analyzed. In AI, data may include text, images, audio, video, numbers, records, or user inputs.

Data privacy refers to how personal or sensitive information is collected, used, stored, shared, and protected. When using AI tools, users should consider what information they are entering and whether the tool is appropriate for that information.

A dataset is a collection of data used for analysis, training, testing, or evaluation. Datasets can include text, images, numbers, audio, video, or other forms of information.

Deep learning is a type of machine learning that uses layered neural networks to identify complex patterns in large amounts of data. It is used in many AI systems, including image recognition, speech recognition, and generative AI.

A deepfake is synthetic or altered audio, image, or video content that makes it appear as though someone said or did something they did not say or do. Deepfakes can be used to mislead, impersonate, or spread misinformation.

Disinformation is false or misleading information that is created or shared intentionally to deceive people.

Explainable AI, or XAI, refers to approaches that make AI outputs easier for people to understand. XAI can help users understand why an AI system produced a recommendation, classification, summary, or decision.

An external AI tool is a tool that is not provided, approved, or supported by George Brown Polytechnic. External tools may have different privacy, security, accessibility, and data protection settings.

Fabricated content is false information created from scratch. In AI, fabricated content may include invented sources, statistics, events, quotes, people, or details.

Fairness in AI means working to ensure that AI systems do not systematically disadvantage people or groups. It includes considering bias, representation, access, equity, and impact.

Fine-tuning is the process of adapting an existing AI model for a more specific task, domain, or audience using additional training or examples.

A foundation model is a large AI model trained on broad data that can be adapted for many different tasks. Large language models are one type of foundation model.

Generative AI is a type of AI that creates new content based on patterns learned from data. It can generate text, images, code, audio, video, summaries, ideas, and other outputs.

GPT stands for Generative Pre-trained Transformer. It refers to a type of large language model that can process and generate text and other content.

A hallucination occurs when an AI tool produces information that is incorrect, unsupported, or invented, even if it sounds confident or convincing.

Example: An AI tool might invent a source, misstate a policy, or provide a false statistic.

Human-in-the-loop means that a person remains actively involved in reviewing, guiding, approving, or making decisions when AI is used. This is important because AI outputs can be incomplete, biased, inaccurate, or inappropriate for the context.

Human-first AI means using AI in ways that keep people, care, judgment, accountability, accessibility, and social impact at the centre. AI should support human work and learning, not replace human responsibility.

An input is the information a user gives to an AI system. Inputs can include prompts, questions, instructions, documents, images, data, or other materials.

An institutionally supported AI tool is an AI tool made available through George Brown Polytechnic or approved for use through institutional processes. These tools generally provide a stronger privacy, security, and support foundation than unsupported public tools.

An intelligent tutoring system is a digital learning system that provides learners with feedback, guidance, or practice based on their responses or progress. Some intelligent tutoring systems use AI to adapt learning activities or identify areas where a learner may need more support.

A large language model, or LLM, is an AI model trained on large amounts of text to process and generate language. LLMs can support tasks such as answering questions, summarizing text, translating language, writing drafts, and generating code.

Learning loss, in the context of AI, can refer to missed learning or skill development when AI is overused or used in place of active thinking, practice, or problem-solving. AI should support learning, not replace it.

Machine learning is a subset of AI where systems learn patterns from data and improve their performance over time without being explicitly programmed for every step.

Misinformation is false or misleading information that is shared without the intent to deceive. AI tools can contribute to misinformation if outputs are not checked before being shared.

A model card is a document that explains key information about an AI model. It may include what the model is designed to do, how it was trained, its intended uses, limitations, risks, performance, and evaluation results.

Model cards can support transparency and informed decision-making when selecting or assessing AI tools.

Multimodal AI refers to AI systems that can work with more than one type of input or output, such as text, images, audio, video, or code.

Natural language processing, or NLP, is a field of AI focused on enabling computers to understand, interpret, and generate human language.

A neural network is a type of machine learning model made up of connected layers that process information and identify patterns. Neural networks are inspired by the structure of the brain, but they do not think like humans.

An output is the response or result produced by an AI system. Outputs may include text, images, code, summaries, tables, classifications, recommendations, or other generated content.

Overfitting happens when an AI model learns the training data too closely and does not perform well on new or different data. It is like memorizing answers instead of understanding the pattern.

Parameters are internal values that an AI model learns during training. They influence how the model processes information and generates outputs.

Personal information is recorded information about an identifiable person. Users should take care before entering personal information into AI tools and should follow George Brown guidance, privacy practices, and institutional policies.

Plagiarism is presenting someone else’s work, words, ideas, or outputs as your own without proper acknowledgement. Expectations for AI use, citation, and acknowledgement may vary by course, assignment, publication, or context.

Predictive AI uses data to estimate what may happen in the future. It may be used to identify patterns, forecast outcomes, or support planning.

A prompt is the instruction, question, or information a user gives to an AI tool. Clear prompts can improve the usefulness of AI outputs.

Prompt engineering is the practice of writing clear and effective prompts to guide an AI tool’s response. This may include giving context, specifying a role, setting a format, adding examples, or defining constraints.

A public AI tool is an AI tool available online that is not provided or supported by George Brown Polytechnic. Users should not enter George Brown work information, institutional data, personal information, or confidential content into unsupported public tools.

A reasoning model is an AI model designed to work through complex tasks more deliberately. It may be useful for multi-step analysis, logic, planning, coding, or problem-solving, but its outputs still require human review.

Reinforcement learning is a type of machine learning where a system improves by receiving feedback on its actions. Some AI models are also refined using human feedback.

Responsible AI refers to the design, use, and governance of AI in ways that are ethical, safe, fair, transparent, accountable, inclusive, and aligned with human needs.

Retrieval-augmented generation, or RAG, is a method that allows an AI system to use selected documents, databases, or sources to produce more grounded responses. RAG can help improve accuracy when the source material is reliable and relevant.

Sensitive information is information that could cause harm, risk, or privacy concerns if it is exposed, misused, or shared inappropriately. This may include health, accommodation, legal, labour relations, financial, security, personal, or confidential institutional information.

Supervised learning is a type of machine learning where a model learns from labelled examples. The system is shown inputs and correct outputs so it can learn patterns and make predictions on new information.

Example: A model trained with labelled images of cats and dogs learns to classify new images as either a cat or a dog.

Synthetic media is content created or altered by AI, such as AI-generated images, audio, video, voices, or avatars.

Temperature is a setting in some AI tools that affects how predictable or creative the output may be. A lower temperature usually produces more focused and consistent responses, while a higher temperature may produce more varied or creative responses.

A token is a small unit of text that an AI model processes. Tokens can be words, parts of words, punctuation, or characters. Token limits affect how much information a model can process at one time.

Training data is the information used to train an AI model. The quality, diversity, and limitations of training data can affect the accuracy, fairness, and reliability of AI outputs.

A transformer is a type of AI model architecture used in many modern language and generative AI systems. Transformers are effective at identifying relationships in language and other data, which allows them to support tasks such as writing, summarizing, translation, and chat.

Transparency means being clear about when and how AI is used. This may include acknowledging AI assistance, explaining how outputs were created, or making sure users understand the role AI played in a task or decision.

An unsupported public tool is a tool that has not been reviewed, approved, or supported by George Brown Polytechnic. These tools may not meet institutional expectations for privacy, security, accessibility, procurement, or data protection.

Unsupervised learning is a type of machine learning where a model looks for patterns in data without being given labelled examples. It can be used to group similar information, identify patterns, or discover relationships in data.

Validation is the process of testing or checking how well an AI model performs on data it has not already seen. Validation helps determine whether the model is likely to work reliably beyond the examples it was trained on.

See Explainable AI (XAI).