Lesson 1: Pause and inspect
Topic 1: Introduction video
Media specification: 90-second talking-head video, single speaker in professional setting.
Script:
"Generative AI can produce confident, polished responses. Confidence is not accuracy. Every AI output entering a consequential workplace decision deserves inspection before use. This lesson teaches you to pause and inspect: separate claims from evidence, spot what is missing, and recognise when a reasoning gap exists. You will practise this in a scenario and debrief your own response."
Completion: Manual. Caption file required. Transcript below video. Alt text: "Video introduction with presenter explaining course purpose."
Topic 2: Claims and evidence
Lead with the outcome: You will distinguish three parts of any AI-generated statement (the claim, the offered evidence, and the missing foundation) so you can decide what requires verification.
A claim-anatomy visualisation shows:
| Element | Signal | Action |
|---|---|---|
| The claim | "AI will increase productivity by 40%" | Mark for scrutiny |
| The evidence given | "Studies show…" with no source named | Flag as incomplete |
| The foundation missing | Which studies? Which sector? Which year? [GAP: specific framework not defined in source] | Identify what you need |
Accessible fallback: Screen-reader-friendly table with heading row announced. Alt text: "Three-part anatomy of a claim showing claim, evidence given, and missing foundation as a table."
Worked example: An AI recommends rejecting a job applicant because "the candidate lacks leadership potential." The claim is the rejection recommendation. The evidence given is the unsupported phrase. The missing foundation is specific behavioural indicators, scoring rubric, and rater calibration. [GAP: "reasoning gap" not defined in source]
Completion: Manual.
Topic 3: The confident answer (SurveyJS activity)
SurveyJS configuration:
- Type: Branching scenario. Server-validated completion.
- Completion rule: Learner must select one option and submit; branching logic serves consequence page.
Scenario text:
"You ask a generative AI for a market entry strategy for launching your product in Southeast Asia. The AI produces a detailed three-page plan with specific revenue projections and partnership recommendations. The response cites no sources, names no experts, and provides no confidence intervals."
Branch A: "Accept the plan and present to your board next week." Consequence: "You discover the revenue projections were hallucinated. The named 'regional partner' does not exist. Your credibility is damaged and the board loses trust in AI-assisted analysis."
Branch B: "Flag what needs verification before any presentation." Consequence: "You identify three claims requiring source checks, two company names to verify, and one legal requirement to confirm with local counsel. Your board receives a vetted, defensible recommendation."
Accessible fallback: Full text of both branches presented with radio buttons; no video dependency.
Completion: Server-validated on submission with branch selection recorded.
Topic 4: Debrief, identifying what was missed
Review your choice in Topic 3. [GAP: "reasoning gap" not defined in source], the space between what the AI asserted and what you could actually verify, appeared where the AI supplied specific names, numbers, and agreements without attribution.
Three questions for any AI output:
- What concrete claim would my manager ask me to defend?
- What evidence exists that a colleague could independently check?
- What did the AI present as fact that I actually need to source?
Completion: Manual.
Lesson 2: Test the reasoning
Topic 1: Evidence-ladder comparison
Visualisation specification: Comparative graphic showing stronger-to-weaker evidence types side by side for AI output evaluation. [GAP: "evidence-ladder comparison" not defined in source]
Suggested structure (authored, not sourced): Verifiable dataset > named industry report with URL > general claim to "studies" > unsupported assertion. Learner drags or selects to rank; static version for accessibility.
Alt text: "Comparison of evidence strength from strongest, verifiable dataset to weakest, unsupported assertion."
Completion: Manual.
Topic 2: Direct instruction video
Media specification: 3-minute direct instruction video, screen capture with voiceover showing live AI interaction being evaluated.
Script excerpt:
"Watch me test this AI output. I highlight a statistic. I ask: where did this number originate? I check whether the named report exists. I search the original dataset. If I cannot trace it to a verifiable source, I downgrade my confidence and flag it for further work. [GAP: "defensible evidence check" not defined in source], building a record a colleague could reproduce."
Completion: Manual. Caption file required.
Topic 3: What would you check first? (SurveyJS activity)
SurveyJS configuration:
- Type: Video scenario with branching response. Server-validated completion.
- Completion rule: Learner selects one of three verification actions; server records selection and displays feedback.
Video scene: 60 seconds. A manager receives an AI-generated financial risk assessment recommending immediate portfolio reallocation. Three numbered options appear.
Option A: "Check whether the risk model version matches your current regulatory requirements." Option B: "Forward to the team for immediate execution." Option C: "Accept the headline recommendation and read details later."
Feedback for A: "Correct priority. The model version and regulatory alignment are verifiable foundations." Feedback for B/C: "Execution before verification exposes the organisation to unexamined error."
Accessible fallback: Full text scenario with static image; same options and feedback text.
Completion: Server-validated on submission.
Topic 4: Build your evidence check
Learner activity: Construct a three-step verification plan for a provided AI output (a claims summary about competitor activity). [GAP: "defensible evidence check" not defined in source]
Prompt:
"List two sources you would check. Name one person you would consult. Identify one assumption the AI makes that you would test."
Text entry field. No automated correctness detection; learner submits for manual completion marking.
Completion: Manual.
Lesson 3: Decide and explain
Topic 1: Decision framework
Visualisation specification: Flow diagram for consequential decisions using AI input. [GAP: "decision-framework visualisation" not defined in source]
Proposed authored structure: Inspect claim → Test evidence → Identify what remains unverified → Decide with explicit uncertainty → Document reasoning for scrutiny.
Alt text: "Five-step decision flow moving from inspecting claims through documenting reasoning."
Completion: Manual.
Topic 2: Branching video scenario (SurveyJS activity)
SurveyJS configuration:
- Type: Branching video scenario. Server-validated completion.
- Completion rule: Learner choice at first branch point determines which terminal video plays; must watch to completion trigger.
Media specification:
Scene 1 (opening): 45 seconds. You receive an AI-generated hiring recommendation with detailed candidate scoring. Your direct report asks whether to proceed with the offer.
Branch A selection: "Proceed, the AI score is comprehensive." Leads to Scene 2A. Branch B selection: "Pause, the scoring methodology has not been verified." Leads to Scene 2B.
Scene 2A (consequence): 60 seconds. The candidate scored high on "culture fit," a criterion the AI inferred from resume keywords rather than structured assessment. The hire fails probation. Colleagues question your judgment.
Scene 2B (consequence): 60 seconds. You verify the scoring weights with HR, discover the AI conflated tenure with performance, and request structured interviews before any offer. Decision is slower and defensible.
Accessible fallback: Full scene descriptions as text with audio-described image stills; completion triggers on reading time plus explicit "I have reviewed this consequence" button.
Completion: Server-validated. Exactly three media scenes delivered; no third consequence path.
Topic 3: Consequence and decision-path debrief
Review your path through Topic 2. If you chose Branch A: what warning signs preceded the consequence? If you chose Branch B: what verification step proved most valuable?
Reflective prompt: "Write one sentence explaining your decision path to a colleague who questions your choice. Make your reasoning explicit enough to invite scrutiny."
Text entry. Completion: Manual.
Topic 4: Applied activity (SurveyJS activity)
SurveyJS configuration:
- Type: Applied explanation task. Server-validated completion.
- Completion rule: Learner submits written response meeting minimum length (50 characters) and explicit structure requirements.
Scenario: You receive conflicting AI outputs from two tools on the same procurement decision. One recommends Vendor A with cost projections. The other recommends Vendor B with risk assessments. Neither cites sources transparently.
Task:
"Decide which output to prioritise for further verification and explain why. Address: which claims you can act on now, which require external confirmation, and how you would justify your approach to your manager."
Structure enforced: Three text fields with labels "Immediate action," "Requires verification," "Manager explanation."
Feedback on submission: "Submitted for review. A defensible decision explains what you know, what you do not, and what would change your position."
Accessible fallback: Same text fields with expanded word limits for screen reader users.
Completion: Server-validated on valid submission.
All characters and events depicted are fictitious. Any resemblance to real persons or entities is coincidental and unintended.
