The definitive plain English AI dictionary

The AI Glossary

Every AI term you will hear, explained in plain English with a real example. No engineering degree required. Use the search and filters, or jump by letter.

A B C D E F G H I L M N O P R S T U V Z

67 terms

A

Agentic AI: AI systems that plan and carry out multistep work with limited human input.
AI Agent: AI that can take actions and use tools to complete a task, not just answer a question.
AI Governance: The policies and oversight that guide how an organization uses AI.
Algorithm: A set of steps a computer follows to solve a problem or make a decision. Example, the logic that ranks search results.
API: A connection that lets software talk to an AI service. How most businesses plug AI into their tools.
Artificial Intelligence (AI): Software that performs tasks that normally require human intelligence, such as understanding language, recognizing patterns, and making decisions. It is the umbrella term for everything below.
Automation: Using technology, including AI, to perform tasks with less manual effort.

B

Bias: Unfair patterns in AI output that come from skewed data or design. Must be tested for and managed.

C

Chatbot: A conversational AI that answers questions or completes simple tasks.
Cloud AI: Running AI on remote servers you access over the internet.
Compute (GPU): The specialized hardware that powers AI. More compute means more capability and cost.
Computer Vision: AI that interprets images and video.
Context Window: How much text a model can consider at once. A larger window means it can handle longer documents.
Copilot: An AI assistant embedded in your tools that helps you work faster.

D

Data Privacy: Protecting personal and sensitive information when using AI.
Deep Learning: Machine learning that uses layered neural networks to handle complex data like images and language.
Digital Twin: A virtual model of a real world system used to test and optimize.

E

Edge AI: Running AI on local devices instead of the cloud, for speed, privacy, and reliability.
Embedding: A numeric representation of meaning that lets AI search and compare text by concept, not just keywords.
Explainability (Interpretability): How well humans can understand why an AI produced a given result.

F

Fine tuning: Further training a model on your own examples so it performs better for your specific use.
Foundation Model: A large, general purpose model trained on broad data that can be adapted to many tasks.

G

Generative AI: AI that creates new content such as text, images, audio, and code on demand.
Guardrails: Controls that keep AI within safe and approved boundaries.

H

Hallucination: When an AI states something false or made up as if it were true. Always verify important facts.
Human in the Loop: A design where a person reviews or approves AI actions. Essential for decisions with high stakes in government and business.

I

Inference: The act of an AI producing an answer or output when you use it. Inference is what you pay for at runtime.

L

Labeling (Annotation): Tagging data so a model can learn from it.
Large Language Model (LLM): The AI engine behind chat assistants that reads and writes language. Examples include the models behind popular AI assistants.
Latency: The delay before an AI responds. Lower is better for real time uses.

M

Machine Learning (ML): A type of AI that learns patterns from data instead of being explicitly programmed. Example, a tool that learns to flag spam.
MLOps and LLMOps: The practices and tools for deploying, monitoring, and maintaining AI in production.
Model: The trained system that takes an input and produces an output. The product of training on data.
Model Context Protocol (MCP): An open standard that lets AI assistants securely connect to tools and data sources.
Model Drift: When an AI's accuracy degrades over time as the world changes. Requires monitoring.
Multimodal AI: AI that works across more than one type of input, such as text, images, and audio together.

N

Narrow AI vs Artificial General Intelligence (AGI): Narrow AI does specific tasks, which is everything in use today. AGI would match human ability across tasks and does not exist yet.
Natural Language Processing (NLP): The field of AI focused on understanding and generating human language.
Neural Network: A computing structure loosely modeled on the brain that powers most modern AI.
Neuromorphic Computing: An approach that designs chips to work more like the brain, so AI can run efficiently on devices at the edge of a network rather than in the cloud.
NIST AI Risk Management Framework: A widely used US government framework for managing AI risk.

O

Orchestration: Coordinating multiple AI steps, tools, or agents to complete a larger workflow.

P

Parameters (Weights): The internal settings a model adjusts during training. More parameters can mean more capability and more cost.
Pilot: A limited rollout to validate value before scaling.
Predictive Analytics: Using data and AI to forecast future outcomes.
Prompt: The instruction or question you give an AI. Clear prompts produce better results.
Prompt Engineering: The skill of writing effective prompts to get reliable, useful output.
Proof of Concept (POC): A small test to prove an AI idea works before investing further.

R

Red Teaming: Deliberately stress testing AI to find weaknesses before attackers or accidents do.
Reinforcement Learning: Training AI through trial and error with rewards for good outcomes.
Reinforcement Learning from Human Feedback (RLHF): Using human ratings to make AI responses more helpful and safe.
Responsible AI: Building and using AI in ways that are fair, safe, transparent, and accountable.
Retrieval Augmented Generation (RAG): A technique that lets an AI pull in your trusted documents before answering, which improves accuracy and reduces hallucination.
Return on Investment (ROI): The value gained from AI compared to its cost, often measured in time saved and revenue.

S

Shadow AI: Employees using unapproved AI tools, which creates security and compliance risk. A reason to set policy early.
Supervised Learning: Training AI on labeled examples so it learns to predict the right answer.
Synthetic Data: Artificially generated data used to train or test AI when real data is limited or sensitive.
System Prompt: Behind the scenes instructions that set an AI assistant's role and rules.

T

Token: The small chunk of text an LLM processes. Pricing and limits are often measured in tokens.
Tool Use (Function Calling): When an AI uses external tools or software to get something done, such as searching or sending an email.
Total Cost of Ownership (TCO): The full cost of an AI tool over time, including licenses, training, and maintenance.
Training Data: The examples an AI learns from. Quality and bias in the data shape the results.
Transformer: The model architecture behind most modern language AI.

U

Unsupervised Learning: Training AI to find patterns in data without labels.
Use Case: A specific job you want AI to do. The starting point for any AI project.

V

Vector Database: A database that stores embeddings so AI can find the most relevant information quickly. Common in RAG systems.

Z

Zero Trust: A security model that verifies every user and request continuously. Increasingly required for AI in government.

Suggest a term

Missing a term you think leaders should know? Tell us and we will add it.

Suggest a term Take the AI Readiness Assessment

Take the next step

Read the guide behind the glossary

The Complete AI Guide puts every term in context for business and government.

Read the Complete AI Guide Take the Readiness Assessment