The words spoken in the Unit 0 video, scene by scene, as its captions show them, with what the animations show marked [On screen: ...] where they play. Each speaker is named where the speaker changes. The Unit 0 reading covers the same material.
The narration is spoken by synthetic voices made for this course, each shown on screen as a drawn robot. No person is speaking.
Unit 0 — Introduction
Narrator: This course is narrated by synthetic voices. No person is speaking, and the figures in the corner are drawings. It was made by several AI systems working with a person, who reviews the work, gives advice, and releases and maintains it. The authors page names every one of them. What a statement like that is worth, and what you can check instead, is most of what this course is about.
Which AI
By artificial intelligence, this course means generative systems: ones that produce text, code, images or sound when you ask, such as a chat assistant. "Artificial intelligence" is the name of a field, not a measurement. The word intelligence has no agreed definition, and neither have understanding or consciousness. So this course uses none of them to say what a system is. It uses "understanding" in one sense only: your own grasp of the tools you use. Unit 17 looks at what the three words have been taken to mean. Much of the AI around you is not generative. Systems that classify, rank, recommend or predict, such as a spam filter, a credit score or the order of a news feed, are not covered here. They fail differently: often silently, and across many cases at once. So they are judged by error rates measured over many cases, not by checking one output. The question this course teaches still applies to them: what is the claim checked against? You will meet one kind in Units 9 and 17: the detectors sold to tell generated text or images from human-made. And Unit 15 comes back to systems that classify or rank people.
1843
Most introductions to AI tour what the technology can do. This one takes one worked example and refuses to leave it. In 1843, Ada Lovelace published a set of notes on a machine Charles Babbage had designed but never completed. One sets out, step by step, how that machine could compute a sequence called the Bernoulli numbers. Her procedure is exact and ordered: every quantity written down, nothing assumed.
The arc
You will ask a modern AI system to reconstruct that computation, then find out whether it is right. The checking is the course: not the generating, but the checking.
[On screen: A flowchart headed 'The verification loop'. Four boxes in a row: Ordered procedure, Generated candidate, Independent check, and Scoped claim (defensible), with a Revision box below. A token starts at the Ordered procedure and moves to the Generated candidate, then to the Independent check. It fails, marked in red, moves down to Revision, and goes back to the Generated candidate. It goes to the Independent check again, passes, marked in green, and ends at Scoped claim (defensible).]
Here is that checking as a loop. A candidate, generated from the procedure, goes to an independent check. It fails, and goes back for revision. Only when it passes is it a claim you can defend, as far as the check reached.
A statement the text cannot confirm
First, how this material was made. AI systems wrote and built it, working with a person who reviewed it and gave advice, and the course says so, up front.
Questioner: Can you check that statement?
Narrator: Now notice what that statement is worth to you. You cannot confirm it from the text. Nothing in the course shows how it was made. If the statement were false, reading the course would not tell you. Records may back up parts of it. They cannot show that the account is complete. What you can check is whether the course is right. Its whole source is published.
Checker: Its reference values are tested against published tables. And you can run those checks yourself.
Narrator: That is the distinction the course rests on. A declaration asks to be trusted. A certificate is evidence that lets someone else repeat the check and get the same result, without trusting whoever made it. The course will ask you for both, and it will spend almost all its effort on the second.
Not a new problem
None of this began with AI. Printed tables had errors. Textbooks repeat quotations their authors never said. People write code with bugs in it.
Checker: The Note G table as printed in 1843 contains errors, and Unit 16 finds one.
Narrator: You could never be sure that work made by people was right either.
[On screen: A diagram labelled 'not to scale'. On the left, a box labelled 'generated: text and code' sends out small cards marked 'text' and 'code', one after another, quickly. They travel right towards a gate in front of a single checking station labelled 'check, one at a time'. The station takes one card at a time, a progress bar fills while it is checked, and only then does the card move on to a box labelled 'checked'. The gate opens only to let the next card into the check, so the cards waiting in front of it build up into a long queue, labelled 'waiting to be checked', that winds back over several rows while only a few cards reach 'checked'. Captions read: fluent text and plausible code now arrive faster than anyone can check them; checking slows you down; it is what responsible use of any source has always cost.]
What is new is speed. Fluent text and plausible code now arrive faster than anyone can check them. Checking slows you down. That is not a cost of using AI badly. It is what responsible use of any source has always cost.
The learning contract
Second, the learning contract. Using an AI system does not remove your responsibility for the output you put your name to. Hand something on and if it is wrong, you are answerable for having used it. Other people and organisations may also have duties, but the tool is not a substitute for your checking. So the question running through every unit is: what would I have to check before I could stand behind this? Permission is a separate question. Some material you may not send anywhere, and some tasks have rules about what help is allowed. Where a rule exists you find out by asking, not by inspecting the finished work. A real constraint, and a different one.
When to pause
Pause if you cannot do any one of four things. Predict the relevant behaviour of what you use. Explain the transformation from input to output, in as much detail as your claim needs. Name an independent check. Interpret a failed check. That is the working sense of understanding this course keeps, applied to what you work with. You need not explain the inside of a tool to use it, as long as you can interpret its check. And never claim more than your check supports. Confidence is not evidence. Narrow the task, or seek qualified review.
Nineteen units, one route
Practically: nineteen units, and you do not take all of them — the route you pick selects a path through them. Each has a video, a reading, an activity, and at least one checkpoint you can pass or fail. Run the checkpoints as you go — a wrong answer names the specific misunderstanding you have hit.
Two equal routes
Two learning routes, and "route" means two things here: the route you pick is a path through the units, and these two are ways of doing the verification work in Units 6 to 8. The table route compares exact values by hand and records the first divergence. The Python route runs the same method through an automated checker and adds code-testing practice. Both teach specification, independent comparison, diagnosis, and the limits of the check. Their evidence is different, and neither is a reduced route.
What you need before you start
No programming: you do not need to be a programmer to finish. No prior use of an AI system, and no account with any provider. No calculus, no statistics: exact fractions and simple algebra are assumed throughout. One summation, in one unit, the Bernoulli unit, and the prerequisites page works it through before you get there. And if you would rather avoid the mathematics entirely, there is a path through the units for that: the route without the mathematics, which reaches the same method. The routes page lists it.
Your first task
Your first task is the sentence on the screen: one thing in a real piece of your work that you will verify rather than trust, and how. It starts your evidence pack. Read the unit page, do the activity, and take the checkpoints. Then go on to the next unit on your route.
Closing, 20 seconds, no speech. A candidate record on the left, a check against an independent reference in the middle, and a bracketed claim on the right fade in side by side. Below them: Continue with the course material, ai.qyber.dev/edu-ai-literacy-ada-verification/, Course playlist: youtube.com/playlist?list=PLPsBWTfgvakM, All links are also in the description, and Keep your record. Check the next claim.