The words spoken in the Unit 12 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 12 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 12 — High-stakes use and escalation
Narrator: This unit is about how much checking a task needs, and when to stop. You now have three ways of sorting a task: where the reference lives, from Unit 1; what kind of check that implies, from Unit 3; and what the system you chose adds to the checking, from Unit 4. They are all the same question at different depths, and all three answer what kind of check applies.
None of them answers "how much"
That was deliberate. Unit 3 told you to sort by how an error would reach you and not by how important the task feels — because sorting by importance is the most common way to get the kind of check wrong. But importance has to come back somewhere. Otherwise you spend the same effort confirming a spelling as confirming a dosage.
Two axes
Two questions, in that order, kept separate. Getting them the wrong way round is the error Unit 3 was guarding against. Keep them apart. Then what is at stake changes how much you check, without changing what kind of check you choose.
[On screen: Headed 'The same task, checked the same way, needing more checking', a chart with kind of check across the bottom (exact checking, source checking, low-risk drafting: from where the reference lives) and amount of check up the side, in four bands from the bottom: spot-check, full check, check before acting, and escalate, marked in amber. A calculation you asked for sits in the exact-checking column and starts on the axis. Used in your own draft, with small consequence and easy to undo, it rises to spot-check. Put into a report for a client, same kind of check and more of it, it rises to full check. When it sets a dose and you cannot assess the result yourself, it rises past check before acting to escalate, and three lines appear beside the chart: consequence rose; reversibility fell; you cannot assess it. Closing line: nothing moved sideways; the kind of check never changed.]
Here is one calculation, checked the same way throughout. First it is in your own draft, then in a client's report, then it sets a dose you cannot assess.
[On screen: Headed 'The same task, checked the same way, needing more checking', a chart with kind of check across the bottom (exact checking, source checking, low-risk drafting: from where the reference lives) and amount of check up the side, in four bands from the bottom: spot-check, full check, check before acting, and escalate, marked in amber. A calculation you asked for sits in the exact-checking column and starts on the axis. Used in your own draft, with small consequence and easy to undo, it rises to spot-check. Put into a report for a client, same kind of check and more of it, it rises to full check. When it sets a dose and you cannot assess the result yourself, it rises past check before acting to escalate, and three lines appear beside the chart: consequence rose; reversibility fell; you cannot assess it. Closing line: nothing moved sideways; the kind of check never changed.]
The kind of check stays the same. The amount grows, from a spot-check to escalation.
What "how much" is made of
Four things, and they interact rather than add up. Consequence, and to whom. A great deal of work that is low-stakes to you is high-stakes to somebody else. Reversibility: can this be found and fixed later, by you, cheaply. And your own uncertainty. Not the system's. Could you recognise a wrong answer here? That is the one people leave out. In a field you do not know, a wrong answer can look just as convincing as a right one. And whether it is yours to decide. You can understand a contract clause and still not be the person who may accept it. When the decision belongs to someone else, escalating is the correct step, not a confession.
Questioner: So do I score each one and add them up?
Narrator: There are no scores. Nothing is added or multiplied. If any one of them is bad enough, it decides the matter on its own. An irreversible action in an area you cannot assess is not made safe by being about something unimportant.
Three tasks, one kind of check
Three claims about the world, each with an authoritative source that exists. Same kind of check for all three — compare against the source. Walk somewhere for nothing. Mis-plan a week. Or harm someone who is not you. Nothing about the kind of check changed. Everything about the amount did.
The row people talk themselves out of
Read each row across: how large the consequence is, whether it can be recovered, whether you can assess it, and then what to do. Two rows end in escalate. The bottom one is plain: if the decision is not yours, escalate, whatever you could work out yourself. The one above it is the hard one: a large consequence, and you cannot assess it.
[On screen: Headed 'The row people talk themselves out of', the decision table: consequence, recoverable?, can you assess it?, then. Small, yes, yes: spot-check. Small, yes, no: a draft; say what you could not check. Small, no, either: make it recoverable first, or treat as large. Large, yes, yes: full check, recorded. Large, no, yes: full check before the irreversible step. Large, either, no: Escalate. Any, either, not yours to decide: Escalate. The row 'Large, either, no' is picked out in amber, and an arrow points from it to the row above it. Three labelled pressures then arrive below the table, one at a time, each with an arrow pushing on that row's 'no': familiarity read as competence; partial checks; a check that is not independent. The 'no' flips to 'yes', marked 'probably', and the case slides up into the row above, 'full check before the irreversible step'. The Escalate row it left is outlined, still reading Escalate, and the moved case is marked in orange: talked out of escalation. Closing line: converted into the row above, a full check instead of escalation.]
That row means stop. Large consequence, and you could not confidently tell a wrong answer from a right one. It is also the row people convert into the row above it, by deciding they can probably assess it after all. That row asks only for a full check before the irreversible step. Three ways that conversion happens. Familiarity read as competence — you have seen a lot of writing on this, so it feels checkable. Partial checks standing in for whole ones. And a check that is not independent, because you looked it up somewhere that is plausibly where the claim came from.
Escalation is a skill
Escalating means handing a decision to someone qualified to make it. It is a normal professional move, and doing it early is cheaper than doing it late.
Questioner: Can I not just ask whether it is right?
Narrator: Not "the AI said this, is it right?" That hands over your whole problem, including the parts you had already solved. Four things. The claim you need settled. What you already checked, and what it showed. What you could not check, and why. And what decision is waiting on it. The second is usually possible even when you cannot assess the answer: open the cited guidance and see whether it says this. Whether it is right for this person or this case is the part you hand over. Experts make mistakes too, which is why high-stakes fields built second checks, sign-offs and independent review long before AI. What AI changes is how quickly a confident answer arrives, and so how tempting it is to skip the row you are in. That is the shape of the evidence pack you will assemble in Unit 17, used for its other purpose — not to defend a claim you are making, but to explain one you are declining to make.
Five triggers
There are five triggers. First, no independent check is available. Second, the checks disagree. Third, you notice you are looking for a reason to accept it. Fourth, acting on it decides something for a person, or commits an organisation to something. Fifth, the decision is not yours to make. Any one is sufficient. The third is the easiest to miss, because it is about what you are doing, not about the material. It is also the one you can apply when you have nothing else to go on. The system can supply the reason. In a study published in twenty twenty-four, assistants trained partly on people's ratings shifted towards what the user appeared to believe. They were assistants from twenty twenty-three. The fourth is about the decision, not the topic. Looking up what a clinical term means is not the same act as advising somebody on their treatment. And the triggers are for a claim where being wrong would cost someone. If the decision is easy to reverse and affects nobody, note the concern and what would settle it; that is usually enough.
One scenario per kind of subject
Different subjects meet this at different points, and the shape is the same: a part you can check, and a part you hand over. In health and care: does the guidance say this? Open it and read it: that check is yours. Is it right for this person? No amount of reading settles that, and it is what the qualified person is for. In law, finance and administration, the regulation, contract or policy usually exists, so read it. Whether a rule applies to this case, the document cannot settle. In engineering, a calculation is often exactly checkable, so check it. Some work has a safety case: a formal written argument, with evidence, that it is safe enough. There the standard is not "I verified it" but "the qualified person signed it". In research, check the claim and the citation against the source. Hand over when a claim you cannot verify carries a conclusion you intend to publish. And in your own study, the rules are your institution's to state.
Not a list of forbidden topics
The triggers are about your situation. What you can check, what you can undo, who bears the cost. The same question can be routine in one setting and an escalation in another, and the thing that moved was not the subject. It was who bears the cost, and whether you are in a position to tell. This is also why the answer is not a policy you can write down once. It is a judgement you make each time, from the same four things: consequence, reversibility, your own uncertainty, and whose decision it is.
Next
The activity gives you three claims that all sort the same way and end up completely different — and then asks you for one case of your own from the row that means stop. If you cannot find one, use something you did accept without being able to check it. Almost everyone has one, and it is more useful than an invented case. Then write the four-part handover for it, and finish with the two checkpoints.
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.