Glossary

Words that do work

Plain-English definitions — the AI vocabulary you need and the funder-comms vocabulary the Lab uses.

AI Fluency
Module 0

The practiced ability to use AI well enough to keep human judgement, dignity and voice at the centre of the work.

Answer Engine Optimization (AEO)
Module 1

Structuring your website and materials so AI assistants, not just search engines, quote you accurately when funders ask about your work.

A person actively agreeing to a specific use of their story, image or data — informed, revocable, and re-confirmed when the use changes.

Data Colonialism
Module 4

The extraction of stories, images and behaviour from African communities into systems they don't own, govern or benefit from.

Data Sovereignty
Module 4

The principle that communities have authority over how data about them is collected, stored, used and destroyed.

Deficit Framing
Module 3

Describing a community by what it lacks rather than by what it holds and does. A dignity violation.

Delegation
Module 1

The first D. Deciding what to hand to AI and what to keep human — always keeping decisions, positions and consent with people.

Description
Module 2

The second D. Giving AI enough context — voice, values, audience — that its output can honestly be called yours.

Diligence
Module 4

The fourth D. Consent, POPIA, disclosure, human sign-off — the non-negotiables that Ubuntu makes visible.

Discernment
Module 3

The third D. Reading AI output through the three-question filter: Is it true? Is it ours? Does it dignify?

Disclosure
Module 4

Openly telling audiences when AI meaningfully shaped the words, images or analysis they are reading.

Extraction with Wi-Fi
Module 4

A working name for AI use that repeats extractive practice — taking stories, images and labour from communities without consent, credit or benefit.

An organisation or collective a foundation funds. In this Lab, always an agent, never a recipient.

A Game With A Purpose — a short game whose real function is to teach a judgement or reveal a habit.

Hallucination
Module 3

When a language model confidently invents a fact, quote, statistic or source that does not exist.

Large Language Model
Module 0

A system trained on huge amounts of text to predict the next word. It is a very fast pattern-matcher, not a knower of truth.

Monitoring and Evaluation — how funders and grantees track what changed and why. Often the most quietly extractive part of the work.

South Africa's Protection of Personal Information Act. The legal floor for handling personal data — Ubuntu asks for more.

Poverty Porn
Module 3

Imagery or storytelling that dwells on suffering to move donors. Fails the dignity test every time.

The instructions and context you give an AI. In this Lab, the prompt IS the work.

Relational Web
Module 2

The idea that a communication only means anything inside relationships — with grantees, communities, funders, and the wider public.

Search Engine Optimization — structuring content so search engines can find and rank it. AEO's older sibling.

System Prompt
Module 5

The behind-the-scenes instruction that sets how an AI behaves in a conversation. Your voice brief is a system prompt.

Theory of Change
Module 5

A written account of how a funder believes its money and relationships lead to the change it wants to see.

Training Data
Module 0

The text an AI learned from. Mostly English, mostly northern. That's why it flattens South African context by default.

Southern African philosophy: a person is a person through other people. I am because we are.

Ubuntu AI
Module 4

Using AI in a way that honours the relational web — consent, dignity, community authorship, human decisions.

Voice Brief
Module 5

A one-page description of your foundation's voice, values, audiences and no-go language. The reusable context you paste into every AI conversation.