AI conversations can get exclusionary fast. This glossary is for people who want the practical language around artificial intelligence without the hype, math, vendor jargon, or inside-baseball acronyms.
The goal is not to make everyone a machine learning engineer. The goal is to help you ask better questions, spot sloppy claims, and take part in AI decisions that affect your work, culture, communities, and organizations.
How to use this glossary
- Start with the term you keep hearing.
- Read the plain-language definition first.
- Use the Why it matters line to connect the term to a real decision.
- Come back as the language shifts. We revisit these terms as AI changes.
The terms
Agent
An AI system that can take steps toward a goal, often by using tools, checking results, and deciding what to do next.
Why it matters: Agents can be useful for research, coding, operations, and support workflows, but they need clear boundaries because they can act across systems.
AI Ethics
The practice of asking what an AI system should do, who it affects, what harms it may create, and who is accountable when something goes wrong.
Why it matters: Ethics is not a decoration after launch. It belongs in design, data choices, deployment, review, and governance.
AI Literacy
The practical ability to understand what AI can do, what it cannot do, where it might fail, and how to use it responsibly.
Why it matters: AI literacy helps teams move past panic and hype into better decisions.
Alignment
The challenge of making an AI system behave in ways that match human goals, constraints, and values.
Why it matters: A system can be technically impressive and still optimize for the wrong thing.
Bias
A pattern where an AI system produces unfair, skewed, or harmful results for certain people, groups, languages, cultures, or contexts.
Why it matters: Bias can come from data, design choices, measurement, deployment context, or the assumptions of the people building the system.
Chatbot
An interface that lets people interact with software through conversation.
Why it matters: A chatbot may look simple, but the quality depends on the model, instructions, tools, data access, and safety rules behind it.
Context Window
The amount of text, data, or conversation an AI model can consider at one time.
Why it matters: When important context falls outside the window, the model may miss details or answer from incomplete information.
Data Governance
The policies and practices that define how data is collected, stored, accessed, shared, corrected, and deleted.
Why it matters: AI systems often amplify existing data practices. Weak governance becomes a bigger risk when automation enters the room.
Dataset
A collection of examples used to train, test, evaluate, or operate an AI system.
Why it matters: The quality, source, consent, and representativeness of a dataset shape what the system learns and who it may harm.
Evaluation
The process of testing whether an AI system performs well enough for its intended use.
Why it matters: A few impressive demos are not enough. Useful evaluations test real tasks, edge cases, safety, and failure modes.
Fine-Tuning
Training an existing model further on a narrower set of examples so it behaves better for a specific use case.
Why it matters: Fine-tuning can improve consistency, but it is not always the right answer. Better prompts, retrieval, or workflow design may be simpler.
Foundation Model
A large general-purpose AI model that can be adapted to many tasks, such as writing, coding, image understanding, search, or analysis.
Why it matters: Foundation models are powerful building blocks, but they still need context, evaluation, and governance before serious deployment.
Generative AI
AI that creates new text, images, audio, video, code, or other outputs based on patterns learned from data.
Why it matters: Generative AI changes creative, educational, business, and civic workflows because it can produce plausible work very quickly.
Guardrails
Rules, checks, permissions, or design choices that limit what an AI system can do.
Why it matters: Guardrails help reduce risk, but they are not magic. They need testing, monitoring, and human accountability.
Hallucination
When an AI system produces information that sounds confident but is wrong, unsupported, or invented.
Why it matters: Hallucinations are especially risky in journalism, law, health, finance, education, and public-sector work.
Human In The Loop
A workflow where a person reviews, approves, corrects, or supervises an AI system before important actions are taken.
Why it matters: Human review is most useful when the reviewer has enough context, authority, and time to catch problems.
Indigenous Data Sovereignty
The right of Indigenous Peoples to govern the collection, ownership, access, interpretation, and use of data about their communities, lands, cultures, and knowledge.
Why it matters: AI work that touches Indigenous data, knowledge, or communities requires relationship, consent, protocol, and governance beyond generic data policy.
Large Language Model
An AI model trained on large amounts of text that can generate and transform language.
Why it matters: LLMs power many chatbots, writing tools, coding assistants, summarizers, and research workflows.
Model
The part of an AI system that has learned patterns from data and uses those patterns to make predictions or generate outputs.
Why it matters: “The model” is only one part of the system. The surrounding data, prompts, tools, interface, and human process matter too.
Multimodal AI
AI that can work across more than one kind of input or output, such as text, images, audio, video, or code.
Why it matters: Multimodal systems are useful for creative and accessibility workflows, but they also raise new consent, attribution, and interpretation questions.
Prompt
The instruction, question, context, or examples given to an AI system.
Why it matters: Better prompts can improve results, but prompts are not a substitute for good judgment, good data, or proper review.
Retrieval-Augmented Generation
A pattern where an AI system searches approved sources and uses that retrieved context to answer.
Why it matters: RAG can make answers more grounded, but only if the source material is current, relevant, and correctly retrieved.
Synthetic Media
Media created or altered with AI, including images, audio, video, avatars, and generated voices.
Why it matters: Synthetic media can be creative and useful, but it also raises consent, disclosure, provenance, and misinformation concerns.
Token
A small unit of text that an AI model processes. A token can be a word, part of a word, punctuation, or another text fragment.
Why it matters: Tokens affect cost, context limits, and how much information a model can handle at once.
Training Data
The data used to teach an AI model patterns before it is used.
Why it matters: Training data shapes what the model can do, what it misses, and what harms or assumptions it may reproduce.
Workflow Automation
Using software, sometimes with AI, to move tasks through a process with less manual effort.
Why it matters: Automation can save time, but bad workflows scale mistakes faster.
Want this for your team or audience? I run plain-language AI literacy workshops and keynotes that leave people more confident and more discerning, not more hyped. Book an AI keynote or workshop.
Keep exploring: current AI ecosystem work, the blog, or more about Kris Krüg.
This is a first publishable slice of 26 practical terms. AI language changes quickly, so the glossary grows and gets revisited over time.