- Published
- 8 October 2026
- Last reviewed
- 8 October 2026
How to use this glossary
Use these introductory terms when planning an AI-assisted task or workflow. A glossary explains concepts, not whether a particular tool is appropriate for confidential information or a high-impact decision.
Browse the alphabetical term index below. Select a term to jump to its definition, example and common mistake. Use the return link after each entry to browse again. You can also use your browser’s Find command to locate a word on this page. Definitions are original RDE plain-language explanations. All examples are fictional teaching scenarios.
Term index
12 terms, listed alphabetically.
- Artificial intelligence (AI)
- Automation
- Bias
- Evaluation
- Exception
- Generative AI
- Hallucination
- Human oversight
- Large language model (LLM)
- Prompt
- Retrieval-augmented generation (RAG)
- Trigger
Definitions and practical examples
Artificial intelligence (AI)
Meaning: Systems that perform tasks involving capabilities such as prediction, recognition or language generation.
Example: A service classifies incoming requests into suggested categories for human review.
Common mistake to avoid: The label AI does not establish accuracy, autonomy or suitability for a task.
Automation
Meaning: Using a system to perform defined work with reduced manual intervention.
Example: An approved form submission triggers a notification to the relevant coordinator.
Common mistake to avoid: Automation need not involve AI; a fixed rule can automate a step.
Bias
Meaning: A systematic tendency that can distort a model’s results or treatment of groups.
Example: A model performs less reliably for a group underrepresented in its evaluation data.
Common mistake to avoid: Check the task and affected groups rather than assuming one overall accuracy score is sufficient.
Evaluation
Meaning: Testing a system or output against defined criteria and representative examples.
Example: A team checks whether summaries preserve deadlines, exceptions and approval roles.
Common mistake to avoid: A single successful demonstration does not establish reliable performance across real cases.
Exception
Meaning: A situation a workflow cannot handle through its normal route.
Example: A request has no matching department, so it enters a manual review queue.
Common mistake to avoid: Specify a safe route and owner instead of silently dropping the request.
Generative AI
Meaning: AI designed to produce content such as text, images or audio.
Example: A model drafts a summary of a supplied handbook section.
Common mistake to avoid: Generated content can be plausible and still contain unsupported claims.
Hallucination
Meaning: A generated claim or detail that is unsupported or incorrect in its context.
Example: A model invents a policy deadline absent from the handbook it was given.
Common mistake to avoid: Fluent wording is not evidence that the claim is true.
Human oversight
Meaning: People retaining appropriate responsibility to monitor, review or intervene in system use.
Example: An authorised reviewer checks a policy summary before it is shared with staff.
Common mistake to avoid: Review must be feasible and meaningful; merely naming a reviewer is insufficient.
Large language model (LLM)
Meaning: A language model trained on large amounts of data to process and generate text.
Example: A service uses an LLM to propose wording for a draft response.
Common mistake to avoid: Its generated answer is not the same as a verified database record.
Prompt
Meaning: Input that instructs or supplies context to a model.
Example: A prompt asks for a plain-language summary using only the supplied policy section.
Common mistake to avoid: State the task, boundaries and evidence, then still check the result.
Retrieval-augmented generation (RAG)
Meaning: Generation supported by material retrieved from a source collection.
Example: A system retrieves relevant handbook passages before drafting an answer.
Common mistake to avoid: Retrieved material may be outdated, irrelevant or misrepresented in the answer.
Trigger
Meaning: An event or condition that starts an automated workflow.
Example: A newly approved submission starts a notification workflow.
Common mistake to avoid: Define which event qualifies so that edits or retries do not create unintended duplicates.
A distinction worth keeping clear
Retrieval can supply relevant material to a model, while generation produces an answer. Supplying a source does not guarantee that the answer accurately represents it. Evaluation and review must check the output against the task and evidence.
Practise using the terms
A workflow retrieves an approved handbook and asks a model to summarise its absence procedure. The answer invents a deadline. Identify retrieval, generation and the review needed before the summary is used. Write your answer in your own notes before reading the explanation. No answer is submitted or stored by this page.
Suggested answer and explanation
Retrieval selects relevant handbook material. Generation produces the summary. A reviewer must compare the deadline and other claims with the current approved handbook, correct unsupported statements and confirm that the summary is suitable for the intended audience. A source link alone does not make an invented deadline accurate.
Sources and scope
The sources below support the terminology and relevant context. The definitions, examples, cautions and practice tasks on this page are original RDE teaching material, not verbatim extracts or a replacement for an official standard or certification syllabus. Reviewed 8 October 2026. Author: Jefferson Otonbara Imgbi.
- Google: Machine Learning glossary
- NIST: AI Risk Management Framework
- GitHub: Understanding automated workflows
Cover image: AI-generated illustrative scene created specifically for this resource.