Ethical AI
Using Artificial Intelligence Ethically

AI is now a colleague.
Learn to work with it responsibly.

This course gives you the knowledge, judgement and practical habits to use AI tools ethically at work, in study, and in everyday life — through short units, hands-on labs and real-world scenarios.

Audio narrationListen to this screen

By the end of this course you will be able to:

LO1Explain, in plain terms, how modern AI systems learn from data and why this creates ethical risk.
LO2Apply the five core principles — fairness, transparency, accountability, privacy and human oversight — to real decisions.
LO3Identify what data is and is not safe to share with AI tools under UK GDPR.
LO4Use AI with integrity: disclose its use, verify its outputs, and remain accountable for the result.
LO5Move beyond asking questions: delegate real work and lifestyle tasks to AI using the Goal–Context–Format–Constraints recipe.
⏱ 45–60 MIN6 LABS8-QUESTION ASSESSMENTCERTIFICATE
Starter unit 1 · For beginners

Beyond questions: put AI to work

Audio narrationListen to this screen

Most people treat AI like a search engine: ask, read, leave. The real value starts when you delegate a task, not just a question.

❓ Asking (where most people stop)

"What makes a good CV?" · "What can I cook with chicken?" · "How do I run a meeting?"

You get generic information — then you still have to do all the work yourself.

🤝 Delegating (where the value is)

"Rewrite my CV for this job advert." · "Plan my meals for the week using what is in my fridge." · "Build me a 30-minute agenda for Monday's project meeting."

You get a finished draft of the actual task — which you then review, refine and own.

The task recipe: four ingredients

🎯 Goal

What do you want produced? An email, a plan, a table, a checklist, a rewrite. Name the deliverable, not the topic.

🧩 Context

What does the AI need to know? Your situation, audience, background facts. AI can't read your mind — context is where most prompts fail.

📐 Format

What should the output look like? Length, structure, tone, bullet points versus prose, table columns.

🚧 Constraints

The rules: word limits, budget, things to avoid, deadlines, dietary needs, level of formality. Constraints turn generic output into your output.

The prompt builder

Pick a task, then switch on each ingredient and watch the prompt — and its quality — transform.

📧 Draft a client email
📋 Build a meeting agenda
🥘 Plan a week of meals
🧳 Plan a weekend trip
Prompt quality
25%
Select a task above to start building.
Waiting for a task…

Steal these ideas

💼 At work

  • Turn messy meeting notes into actions with owners and deadlines
  • Rewrite a difficult message in a calmer, clearer tone
  • Summarise a long report into a one-page brief for your manager
  • Draft job descriptions, rotas, checklists and handover notes
  • Prepare interview questions — or practise answering them

🏡 In life

  • Meal plans around your budget, diet and what is already in the fridge
  • A step-by-step plan for a house move, event or family trip
  • Explain a confusing letter, bill or contract in plain English
  • A personalised study or fitness plan you can actually stick to
  • Draft complaints, appeals and applications — then personalise them

🔁 One more habit: it's a conversation, not a slot machine

The first draft is a starting point. Reply and refine: "Make it shorter." · "More formal." · "Give me three options." · "You've assumed X — actually it's Y, redo it." Iterating is not failure — it's how skilled users work. And everything you produce this way still follows the rules of this course: verify it, disclose it where needed, and own it.

Starter unit 2 · For beginners

The first draft is never the last

Audio narrationListen to this screen

Skilled users treat AI output as a first draft to direct, not a final answer to accept. You don't start over — you reply.

The refinement studio

The AI has drafted an email announcing a new booking system to your team. It's… not great. Apply refinements and watch it transform. Click a command again to undo it.

🗣️ "Use plain English"
✂️ "Make it shorter"
☀️ "Warmer tone"
🎯 "Add a clear ask"
Readability
20%

💡 What this lab teaches

  • Refining beats regenerating. Small, specific instructions ("shorter", "warmer", "plainer") get you to a usable result far faster than re-rolling and hoping.
  • You are the editor-in-chief. Direction, taste and judgement stay with you — the AI just types faster than you do.
  • Name what's wrong. "I don't like it" gets you nowhere. "Too formal, cut the jargon, end with one clear action" gets you exactly what you want.
Starter unit 3 · For beginners

The prompt doctor

Audio narrationListen to this screen

Diagnosing weak prompts is the fastest way to learn to write strong ones. Six patients are waiting. Which ingredient would most improve each prompt?

Consultation room · 0 correct · 6 remaining

🩺 The doctor's rule of thumb

When an AI answer disappoints you, don't blame the AI first — re-read your prompt and ask which ingredient you starved it of. No deliverable named? Goal. Generic answer that could apply to anyone? Context. Right content, wrong shape? Format. Technically fine but useless for your situation? Constraints.

Unit 1 · Foundations

How AI actually works — and why that matters ethically

Audio narrationListen to this screen

Modern AI — including chat assistants and prediction systems — does not "understand" the world. It is a pattern machine: it learns statistical patterns from vast training data, then uses those patterns to generate outputs.

🗄️
Training dataText, images and records gathered from the world — including its flaws
🧠
The modelLearns patterns and correlations — not truth, not values
💬
OutputsFluent, confident answers — right or wrong
DATA IN → PATTERNS LEARNED → PREDICTIONS OUT · the model can only reflect what it was fed

Three consequences you must internalise

🎯 Bias in, bias out

If the training data under-represents or misrepresents a group, the AI's outputs will too — systematically, and at scale.

🎲 Confident errors

AI generates plausible text, not verified fact. It can "hallucinate" — invent details, citations and figures that look real but aren't.

🔍 Opacity

Even developers can't fully explain a specific output. That's why human judgement must stay in the loop for decisions that matter.

Unit 2 · The ethical framework

The five principles of ethical AI use

Audio narrationListen to this screen

Global frameworks — from the UK Government's AI regulation principles to the EU AI Act and UNESCO's recommendations — converge on five ideas. Click each card to reveal what it means in practice.

⚖️

Fairness

CLICK TO REVEAL

AI must not create or amplify discrimination. In practice: question whether an AI decision would treat different groups differently, and never use AI outputs about people without checking for bias.

🪟

Transparency

CLICK TO REVEAL

People deserve to know when AI is involved. In practice: disclose AI use in your work where it materially shaped the output, and never pass off AI-generated work as purely your own where honesty is expected.

🧾

Accountability

CLICK TO REVEAL

"The AI did it" is never a defence. You remain responsible for anything you submit, publish or act on. In practice: verify before you rely, and own the outcome.

🔒

Privacy

CLICK TO REVEAL

Data you paste into an AI tool may leave your control. In practice: never enter personal data, client information or confidential material into tools that aren't approved for it.

🧑‍⚖️

Human oversight

CLICK TO REVEAL

AI advises; humans decide. In practice: keep a competent human in the loop for any decision affecting people's rights, money, health, safety or opportunities.

0 / 5 principles explored

Lab 1 · Fairness in action

The bias simulator

Audio narrationListen to this screen

You are training a fictional hiring AI. The dots below are the historic applications in your training data — Group A and Group B. Historically, the company mostly hired from Group A. Move the slider and watch what the model learns.

Meridian Logistics · hiring model trainer

90% Group A
Group A approval
78%
Group B approval
31%

💡 What this lab teaches

  • Bias is structural, not intentional. No one programmed the model to discriminate — it inherited the pattern from history.
  • Balanced data helps but isn't a cure. Real fairness work also needs testing, monitoring and human review of outcomes.
  • This is why UK equality law still applies. Using a biased AI in hiring may result in unlawful discrimination under the Equality Act 2010. An employer cannot avoid its responsibilities simply by relying on an AI tool or vendor.
Unit 3 · Privacy & data protection

What you type in may not stay yours

Audio narrationListen to this screen

Every prompt is data leaving your device. It may be stored, reviewed, or used to train future models — depending on the tool and its settings.

Your legal footing: UK GDPR in one minute

👤 Personal data

Information relating to an identified or identifiable living person, such as names, contact details, photos, IDs or opinions about them. When handled for work, sharing it with an AI tool is processing and must comply with data-protection law.

🚨 Special category data

Health, ethnicity, religion, sexuality, political opinions, trade union membership, genetic and biometric data. Extra protection applies. Pasting this into an unapproved AI tool is a serious breach risk.

🏢 Confidential business data

Client lists, contracts, unreleased plans, credentials, safeguarding records. Even if not "personal data", contracts and duties of confidence still apply.

✅ The golden rule

Before pasting, ask: "Would I be comfortable if this appeared publicly with my name attached?" If not — anonymise it, or use an approved tool.

Safer habits

  • Strip names, addresses and identifiers before using AI on real documents.
  • Check your organisation's AI policy and the tool's data settings before first use.
  • Prefer enterprise or approved tools for work data — consumer tools for generic questions only.
  • If you cause or spot a data slip, report it early — under UK GDPR, organisations may have as little as 72 hours to report a notifiable breach to the ICO.
Lab 2 · Privacy in practice

Prompt triage

Audio narrationListen to this screen

A colleague is about to paste each of these into a public, general-purpose chatbot. Click an item, then click the correct bin.

Triage queue · 0 correct · 8 remaining

✅ Safe to paste

⛔ Stop — don't paste

Unit 4 · Integrity, honesty & verification

Use AI without losing your integrity

Audio narrationListen to this screen

Ethical use comes down to honesty in three directions:

🗣️ With others

Disclose AI use where it materially shaped the work — assignments, reports, published content. Follow your institution's or employer's rules; when unclear, ask.

📚 About facts

Verify before you rely. Check names, numbers, quotes, citations, and any legal, medical or financial claims against trustworthy sources.

🪞 With yourself

Learn, don't outsource. If you couldn't explain or defend the output in your own words, you're not ready to submit it.

The verification habit

Before acting on any AI output, run the 4-check:

  • Source check — can I find this claim in a reliable, independent source?
  • Stakes check — what happens if this is wrong? Higher stakes = deeper checking.
  • Freshness check — could this have changed recently? AI outputs may be outdated or lack current information.
  • Sense check — does this actually fit my context, or is it generic advice dressed up?
Lab 3 · Judgement under pressure

Scenario studio

Audio narrationListen to this screen

Scenario 1 · The deadline

It's 22:40. Your report for tomorrow's 9am client meeting is half-finished. An AI assistant has just produced a polished full draft — including three impressive statistics about your client's industry that you don't recognise. What do you do?

Scenario 2 · The shortcut

A teammate has built a brilliant AI workflow: they paste the full text of customer complaint emails — names, account numbers, everything — into a free public chatbot, which drafts empathetic replies in seconds. They offer to set it up for you.
Final assessment

Show what you know

Audio narrationListen to this screen

Eight scenario questions. Pass mark: 6/8 (75%). Feedback is shown after each answer.

Completion

Congratulations — course complete

Audio narrationListen to this screen

This is a certificate of completion for a non-regulated awareness course. It does not award an Ofqual-regulated qualification or professional accreditation.

Your ethical AI toolkit — the one-screen summary

  • Remember the pipeline: data in, patterns learned, predictions out — flaws included.
  • Run the five principles: fairness · transparency · accountability · privacy · human oversight.
  • Apply the golden rule before pasting anything: "comfortable if public with my name on it?"
  • Verify with the 4-check: source, stakes, freshness, sense.
  • Own the outcome: AI drafts — you decide, you disclose, you're accountable.