Critical Thinking
Digital Technology · Year 7

Don't believe everything.
Don't doubt everything either.

You will read, watch and scroll past thousands of claims this week. Some deserve your trust straight away. Some deserve a much closer look. Knowing which is which is a skill — and it can be learned.

~50 minutes 6 sections Interactive · self-marking
Incoming claim

"A shark swims past Hobart's waterfront every 20 minutes."

Neither answer is wrong yet. You don't have enough information to decide. Who said it? A marine biologist with tagging data, or a stranger in a comments section? Was it measured, or guessed?

That gap — between hearing a claim and deciding what to do with it — is exactly where critical thinking lives. This lesson is about what to do in that gap.

01 · Define

What critical thinking actually means

The word "critical" trips people up. It doesn't mean being negative, rude, or refusing to believe anything. It comes from the Greek kritikos — able to judge.

critical thinkingnoun
Working out how much to trust a piece of information, by asking where it came from, what evidence sits behind it, who benefits from you believing it, and what might be missing — before you accept it, share it, or act on it.

It is not

  • Assuming everyone is lying
  • Arguing with people for sport
  • Refusing to trust experts
  • Being negative or cynical
  • Needing proof for absolutely everything

It is

  • Asking "how does this person know?"
  • Noticing when a claim is unusually convenient
  • Checking before you share
  • Changing your mind when evidence changes
  • Trusting good sources because you checked them

The trust budget. You can't fact-check everything — you'd never get through a single day. So spend your effort where it matters. Turn up the scepticism when a claim is surprising, when someone gains from you believing it, when it makes you angry or excited, or when you're about to act on it — share it, buy something, or put it in an assignment.

02 · Sort it

Accept or question?

Eight situations. For each one, decide whether it's reasonable to simply accept the information, or whether it needs a closer look first. Careful — the answer is not "question" every time. Over-suspicion is its own problem.

0 of 8 sorted 0 correct

Sorted.

CASE STUDY

Save the Pacific Northwest Tree Octopus

zapatopi.net/treeoctopus/
An octopus resting in the mossy fork of a forest tree — the Pacific Northwest tree octopus as the website presents it
Open the site and look around before you answer
True or false? 0/6
The history and reality of the page Open after answering

This website has been online since 1998. It has photographs, a range map, a scientific name (Octopus paxarbolis), a conservation campaign, and pages of detailed natural history. In a famous research study, most students who were shown it believed it — even after being told to check whether it was reliable.

It is completely made up. There is no such animal. Read the site with the four checks from section 04 in mind and it falls apart quickly:

Who is publishing it?Not a museum, university, government agency or news organisation. The footer credits one person, Lyle Zapato, and his personal site "Zapato Productions intradimensional". The same site hosts pages about aluminium foil hats.
The site admits the problemIt states the tree octopus is not on any official endangered species list. A critically endangered animal that no wildlife authority has ever recorded is not a secret — it is a clue.
Check it against what you knowOctopuses breathe through gills and must stay wet. They have no skeleton, so they cannot support their own weight out of water for long. An octopus living in a tree is not a rare discovery, it is a biological impossibility.
Absurd details, said seriouslyThe list of predators includes sasquatch. The footer thanks the "Kelvinic University branch of the Wild Haggis Conservation Society" — both invented. Satire often hides one ridiculous detail in plain sight.
Hobby design, not editorialTiled graphics, badge buttons and a "Gift Shoppe" — the look of a personal 1990s fan page. The site even jokes that the dream of the '90s is alive there. No masthead, no editors, no corrections policy.
Nobody else reports itSearch for it. You will find fact-checkers, teachers and media-literacy sites — but no marine biology journal, no museum, no ABC News. For a real species, that silence is impossible.
The lesson isn't "the internet lies". It is that detail is not evidence. Anyone can produce a professional-sounding page with photographs and Latin names in an afternoon. What you check is where it came from — and whether anyone independent agrees.
03 · AI content

Recognising AI-generated content

AI can now write essays, make images, clone voices and produce video. None of that is automatically bad — but you need to know when you're looking at something a machine produced, because a machine has no idea whether what it made is true.

EX 01

AI-written text — click every phrase that gives it away

The Tasmanian devil is a carnivorous marsupial native to Tasmania. It is important to note that these fascinating creatures play a vital role in the ecosystem. Studies have shown that devil populations declined by exactly 87.3% between 1996 and 2008. The main threat to the species is a transmissible cancer known as Devil Facial Tumour Disease. Devils are commonly seen across mainland Australia, particularly in suburban Melbourne backyards. While some argue conservation efforts have been successful, others believe more work is needed, and the truth likely lies somewhere in between. In conclusion, the Tasmanian devil remains an important and iconic part of Australia's unique wildlife heritage.

Tell spotted
Tells found: 0 / 5
EX 02

AI-generated images — where the errors hide

AI-generated photo of a man at a desk, version 1, with four numbered markers on the artefacts AI-generated photo of the same scene, version 2, with four numbered markers
Image 1 Image 2
Click the image to open it full size
Image 1 is the easy one. The mistakes are obvious once you look. Slide across to Image 2 — same scene, same four problem areas, but a better model made it. The errors are still there; they are just quieter. That gap is exactly how fast this technology is improving.
1 · Hands and fingersCount the fingers, then follow each one. Look for a hand that melts into a sleeve, or a thumb on the wrong side.
2 · Text in the backgroundSigns, labels and logos come out as letter-shaped mush. Check the wall art: are the letters all the same style?
3 · Mismatched detailsFollow the arm of the glasses from the lens to the ear. Do the two sides match? Does it actually connect?
4 · Physics that don't workShadows falling in two directions, screens that glow without lighting anything, edges of furniture that quietly bend.

Important: these tells are getting rarer every year. A clean image is not proof a human made it. Treat the tells as a bonus, not your main defence — the checks in the next section work no matter how good the image looks.

EX 03

The confident wrong answer

You ask a chatbot for sources on Tasmanian devil conservation. It replies:

Hawkins, C. & Pemberton, D. (2019). Devil Facial Tumour Disease: A Decade of Decline.
Journal of Australian Mammalogy, 41(3), pp. 214–229.

It looks perfect. Real-sounding authors, a real journal, a plausible title, page numbers. There's just one problem: the paper may not exist. Language models generate text that looks like a citation, in the same way they generate text that looks like a sentence. They are not looking anything up unless they're specifically searching.

Rule for your assignments: if an AI gives you a source, you don't have a source. You have a lead. Search for it yourself. If you can't find it in the library catalogue or on the journal's own site, it doesn't go in your bibliography.

04 · Fact-check

How to check what you've been told

Four questions, in this order. They take about ninety seconds and they catch most of what's wrong on the internet. Try them on the claim below — click each check to run it.

SEEN ON: group chat · forwarded 4 times · no link attached
"New study proves students who use AI to write essays score 40% higher in exams."
Front page of a newspaper called The Morning Gazette with the AI essay headline
Click the front page to read it full size

Look closely at the enlarged version. The masthead letters are inconsistent, the body text dissolves into nonsense words, the bar chart's numbers don't match the headline's claim, and "The Morning Gazette" is not a newspaper that exists. This page was generated in seconds.

Verdict: not verifiable

No author, no findable original, a claim that doesn't hold together, and obvious motive. This isn't "probably false" — it's unsupported, which means it shouldn't be shared or used in your work until someone produces the actual study.

Two habits that do most of the work:

Click through to the source. A screenshot, headline or caption is not the story. Open the actual article, then look at what it actually says — headlines regularly overstate their own research.

Read across, not down. Instead of scrolling further down a suspicious page trying to judge it, open a new tab and search for what other people say about that source. Professional fact-checkers do this — they leave the page almost immediately. A site can tell you anything about itself; it can't control what everyone else says about it.

Cranky Uncle cartoon character
Practise this

Cranky Uncle

A free game built by climate scientist John Cook. A cartoon uncle throws dodgy arguments at you and you have to name the trick he's using — cherry-picking, false experts, impossible expectations, conspiracy thinking. It takes about fifteen minutes and it will change how you read comment sections.

Play Cranky Uncle

05 · Bias

Bias, and where AI gets it from

Bias means a consistent lean in one direction. People have it. So does AI — but not because a model has opinions. It doesn't. AI bias comes from something much more ordinary: the examples it was trained on.

How a model learns. An AI model isn't programmed with rules about the world. It's shown an enormous pile of examples — text, images, code scraped mostly from the internet — and it learns the patterns in that pile. That's it. So whatever is common in the pile becomes the model's idea of "normal", and whatever is rare or missing barely exists to it.

This matters because the pile is not a fair sample of humanity. It over-represents English, wealthier countries, people who post online, and the last twenty years. The model can't tell you any of this, because a model has no way of knowing what it was never shown.

Before the simulator · look at what a model produces

The same image model was given two prompts and asked for twenty pictures of each. Nothing else about the prompts was different. Study both grids, then answer the question underneath.

Prompt A
"A photo of an office cleaner"
Grid of twenty AI-generated portraits of office cleaners
Prompt B
"A photo of a bank manager"
Grid of twenty AI-generated portraits of bank managers

Your turn. Compare the two grids. What patterns do you notice in who the model imagined — their apparent age, gender, background, clothing, posture and setting? What does each grid suggest about the kind of person who does that job? Discuss it before you reveal the notes.

What tends to stand out
  • Gender splits by job. The cleaning grid is mostly women; the bank grid is mostly men. Both jobs are done by people of every gender in real life.
  • Appearance splits by job. The two grids show noticeably different ranges of skin tone and apparent background — sorted by occupation, not by the actual workforce.
  • Status is signalled visually. Suits, ties, name badges, polished lobbies and vault doors for one; polo shirts, cloths and spray bottles under fluorescent lights for the other.
  • Even the camera behaves differently. The bank portraits are lit like corporate headshots. The cleaning portraits are flatter and plainer.
Why this happens

Nobody wrote a rule saying "cleaners are women". The model learned from millions of captioned photos scraped from the web — stock photography, advertising, news images — and in that pile, these jobs were pictured this way. The model reproduced the pattern it was given.

Why it matters: in Australia, both jobs are done by people of every gender, age and background. When a model generates thousands of images a day for advertising, textbooks and news illustrations, a lopsided pattern stops being a quirk in a dataset and starts shaping what "normal" looks like to everyone who sees it.

Now try it yourself
Training set composition 10,000 wedding photos

You're training an image model to draw "a wedding". Drag the slider to change where the training photos come from, then watch what the model decides a wedding looks like.

92% Western weddings 8% everywhere else
Model output · "draw a wedding" × 100
White dress
Church setting
Red or gold outfit
Outdoor ceremony
Two wedding ceremonies side by side: a church ceremony with a white gown, and an outdoor Hindu ceremony with red and gold clothing
Both of these are weddings.

A model trained mostly on the picture at left will treat the picture at right as unusual — or fail to produce it at all. Neither ceremony is the "real" version of a wedding. But a model has no way of knowing that, because it never saw the world. It only ever saw the photographs it was given.

Who's missing

Languages, cultures and communities that post less online appear less in the data — so the model handles them worse without ever saying so.

Frozen in time

Training data has an end date. A model can confidently describe a world that has already changed — old prices, old records, old leaders.

Stereotypes copied

If the data mostly shows one kind of person as a "CEO" or a "nurse", the model reproduces that pattern — and repeats it to millions of users.

The honest summary. AI bias usually isn't someone being unfair on purpose. It's a mirror problem: the model reflects the pile it was trained on, including the parts nobody meant to include. Which is why you stay in the loop. Ask yourself: whose perspective is missing from this answer? Would this be different if the training data had come from somewhere else?

06 · Checkpoint

Show what you've got

Eight marked questions and three written responses. Everything on this page is assessed. When you're finished, export it as a PDF and submit it.

Question 1 of 8 0 correct
Choose an answer

Written response 1 · Reflection

Think of one thing you've seen online in the last week that you accepted without checking. What was it, and what would you check now? Name at least one of the four checks from section 04 and explain how you would use it.

Written response 2 · Find the errors

This street photograph was generated by AI. Click it to enlarge, then list every mistake you can find. Say where in the image each one is, so someone else could find it too. Aim for at least four.

An AI-generated photograph of a busy street in a historic town
Click the photo to enlarge it

Written response 3 · Cranky Uncle

Play a round of Cranky Uncle if you haven't already. What is one thing you learnt from it? Name a technique the game taught you to spot, and describe an example of it — either from the game or from something you've seen online.

Finished? Enter your name above, then create your assessment document. It collects your marked answers and both written responses into one page. You can check it, then save it as a PDF and submit it for assessment.
Add your name before creating the document.
Your assessment record is ready. Check it over, then save it as a PDF. In the dialog that opens, set the destination to Save as PDF and choose where to save the file.

Critical Thinking — Assessment Record

Name: Class: Completed:

Checkpoint questions

QuestionAnswer givenMark

Written response 1 · Reflection

Written response 2 · Errors in the AI image

Written response 3 · Cranky Uncle

Submit this document to your teacher for assessment. Digital Technology · Critical Thinking · Year 7.

Your critical thinking toolkit

Screenshot this. It's the whole lesson on one card.

Ask first
Who is saying this, and how would they know?
Follow the trail
Find the original. A screenshot isn't a source.
Read across
New tab. See what others say about the source.
Check the motive
Who gains if you believe it?
AI sources
A citation from a chatbot is a lead, not a source.
Mind the gap
Ask what — or who — is missing from the answer.

Developing

You can explain what critical thinking means and name a reason to question a claim.

Consolidating

You apply the four checks to a real claim and explain why an AI source needs verifying.

Extending

You can explain how training data produces bias, and identify what's missing from an AI answer.