The AI Mastery Dictionary: 130 AI Terms Explained in Plain Language
A free plain-language dictionary of the AI, business and web-development vocabulary non-technical builders actually meet — with everyday analogies, the pairs everyone confuses, and where each term shows up in real work. Plus a card game to test yourself.
By William Ifeanyi Moore · 2026-08-02 · 22 min read
The AI Mastery Dictionary: 130 AI Terms Explained in Plain Language
Most AI glossaries are written by people who already understand AI, for people who already understand AI. This one isn't. Every entry here is written for someone with no technical background — plain sentence first, everyday analogy second, and a note on where you'll actually meet the term in real work.
Read the full, searchable version at /dictionary, filterable by topic and skill level. And when you want to find out how much of it has actually stuck, play the card game — five decks, three difficulty levels, no account needed.
Key takeaways
- 130 terms across five areas: AI fundamentals, AI at work, building and shipping, AI video and image, and data privacy and governance.
- The confusable pairs are where the value is. AI vs machine learning, authentication vs authorisation, anon key vs service-role key, prompt injection vs jailbreaking, data residency vs data sovereignty, encryption vs hashing. Getting these wrong is what causes real damage.
- Some definitions are still settling — agentic AI, frontier model, AEO and llms.txt all mean different things to different vendors. We flag those rather than pretending they're settled.
- One important legal correction for Nigerian businesses: the current law is the Nigeria Data Protection Act 2023, not the older 2019 regulation. Teaching the outdated one has compliance consequences.
- Two safety habits recur across every area: always verify AI facts, figures and law; and never expose a secret key or skip row-level security.
How to use this
Terms are tiered. Beginner entries are the ones to learn first — they cover about 80% of what you'll meet in an ordinary week of building or working with AI. Advanced entries are optional depth for when a specific problem sends you looking.
If you're learning to build, read the AI fundamentals and the building sections. If you're bringing AI into a company, read AI fundamentals plus data and governance. If you're making images or film, read AI fundamentals plus the video section. Nobody needs all 130 at once.
The six pairs everyone confuses
These are the highest-value clarifications in the whole dictionary. Each one is a mistake we've watched people make repeatedly, and several of them are expensive.
AI vs machine learning
AI is the goal — machines doing things we'd call intelligent. Machine learning is the main technique used to get there today: software that learns patterns from data instead of being programmed with explicit rules. Deep learning sits inside machine learning, which sits inside AI. They are not synonyms, and people who use them interchangeably usually reveal they've not thought about either.
Authentication vs authorisation
Authentication is proving who you are. Authorisation is what you're allowed to do once you're in. Showing your ID at the door versus whether your ticket admits you to the VIP section. Almost every access-control bug is someone conflating the two.
Anon key vs service-role key
Your database's public key is designed to be safe in browser code — but only because row-level security stands behind it. The service-role key bypasses that security entirely. Putting the second one where the first belongs is the single most common critical security mistake in AI-built apps, and it hands anyone who views your page source full access to everything.
Prompt injection vs jailbreaking
Prompt injection targets your application — usually through a poisoned document your AI reads, which contains instructions telling it to ignore its real ones. Jailbreaking targets the model's own safety rules. If you're building anything that reads untrusted content, injection is your problem and it is not solved.
Data residency vs data sovereignty
Residency is where data physically sits. Sovereignty is whose laws govern it. You can satisfy the first and still fail the second: data held in-country by a foreign provider may remain reachable under that provider's home law. Companies routinely believe they've solved a compliance problem when they've solved only half of it.
Encryption vs hashing
Encryption can be reversed with a key. Hashing cannot be reversed at all — which is exactly why passwords are hashed, not encrypted. If a service can email you your existing password, they've done it wrong.
Terms that are still being argued about
An honest dictionary flags what isn't settled. Four terms here mean different things depending on who's speaking:
- Agentic AI — every vendor defines this differently. Broadly it means a system that plans and takes multi-step actions rather than just answering, but treat any single definition as provisional.
- Frontier model — used loosely, mostly in policy and safety contexts. All frontier models are foundation models; not all foundation models are frontier.
- AEO — answer engine optimisation. Real and worth doing, but metrics and best practices are still forming.
- llms.txt — a proposed file summarising your site for AI models. Not an official standard, and studies of real traffic show almost no bots actually request it. Low risk, low proven benefit — treat it as optional.
Anyone teaching these as settled facts is overselling.
The Nigerian legal correction that matters
If you're building or operating in Nigeria, this one has consequences.
Many courses, blog posts and even consultants still reference the NDPR 2019 as Nigeria's data protection rule. It has been superseded. The Nigeria Data Protection Act 2023 is the current law — it established the Nigeria Data Protection Commission, and the Commission has since confirmed the 2019 regulation and its implementation framework are no longer applied.
The Act has extraterritorial reach and stricter duties for organisations processing data at scale or in economically significant sectors. If your compliance documentation cites the 2019 regulation, it is out of date.
The five decks
The full dictionary at /dictionary is organised into five browsable decks — the same decks you can drill individually in the card game:
AI Fundamentals — how these systems actually work. Context windows, tokens, hallucination, grounding, temperature, prompt engineering, RAG, quantisation, MCP. The vocabulary that makes every other conversation possible.
AI at Work — what AI is actually being used for, function by function: content generation, document processing, ticket triage, knowledge management, forecasting, localisation. Including the rule that governs all of it — automating a mess only produces a faster mess.
Building & Shipping — the web-development vocabulary you meet the moment you build something real. Environment variables, row-level security, webhooks, DNS propagation, service workers, split payments, test keys versus live keys.
AI Video & Image — the directing vocabulary for generative film and stills: identity lock and drift, keyframes, seeds, frame extraction, negative prompting, upscaling.
Data, Privacy & Governance — what every business handling real data needs: data classification, local models, precomputation, total cost of ownership, and the difference between the three ways AI spending accumulates.
Two habits worth more than any single definition
Always verify. These systems will state something false with total confidence — a fabricated statistic, a misremembered law, a plausible name that doesn't exist. Verify facts, figures, names and legal claims every single time. The confidence of the output tells you nothing about its accuracy.
Never expose a secret key, never skip row-level security. Anything named "secret", "private" or "service role" belongs server-side only — never in frontend code, never in a repository, never pasted into a chat window. And row-level security defaults to deny-all when you enable it, which means enabling it and writing the policies is the job. These two mistakes cause more real damage to beginner projects than everything else combined.
Where to go from here
Test yourself first. The AI Mastery card game turns this dictionary into five decks and three difficulty levels — hard mode leads with the confusable pairs above, because those are the ones worth actually knowing. No account, no email, and you can share your score.
Learn the craft, not just the words. Knowing what a term means is not the same as being able to use it. Our live courses teach every one of these hands-on, from the foundation, with no experience assumed — monthly cohorts, first one starting September, early-bird seats from ₦5,000.
Don't learn alone. The PxLabs community is where this gets used in anger — a builders' forum, 80+ copy-paste prompt templates, a project showcase, Thursday Vibe Sessions and meet-ups across Nigeria.
The dictionary is free and stays free. Send it to someone who needs it.