Media Studies · Media Studies: Critical Lenses, Platforms, AI and Responsible ProductionGenerative AI and Synthetic Media: How It Works, What It Carries and How to Evaluate It
1.3 1.5 1.6 2.1About 75 minutes
Learning goals- I can explain in plain terms how language models and image generators produce media.
- I can evaluate AI-generated texts through accountability, disclosure, verification and provenance.
- I can analyze the biases AI media can carry and the industry and legal issues around it.
Words to know
generative AI
Software that produces new text, images, audio or video from patterns learned in training data.
large language model (LLM)
A generative AI model trained on huge amounts of text that predicts likely next tokens.
hallucination
Confident but false output from an AI tool.
diffusion model
An image generator that turns random noise into an image step by step, guided by a prompt.
provenance
The documented origin and history of a media file.
training data
The large collection of examples a model learns its patterns from.
Remember from Grade 11
In Grade 11 you sorted synthetic media into fabricated, manipulated, deepfake and false-context content, and learned that tracing the source beats hunting for glitches. You met content credentials briefly.
By 2026, generative AI is part of everyday media: search summaries, chat assistants, AI-written product descriptions, AI-generated images and music, voice cloning and AI-assisted editing in ordinary apps. Grade 12 asks you to understand how these tools produce media, so you can evaluate AI-generated texts (1.3), analyze the perspectives and biases they carry (1.5), understand the industry and legal issues around them (1.6), and treat AI generation as a form with its own characteristics (2.1).
Hook: the library that didn't close
A website called Lakeshore Daily Wire (invented) publishes 40 local articles a day. One reports that a branch of the public library will close next month, with a quote from a library spokesperson and a photo of a sad child holding a book. Parents share it widely. The library has no plans to close. The site has no named staff, no address and no corrections page, and its articles read smoothly but say little.
No human may have written the article, taken the photo or checked the facts. How do you evaluate a text like this?
How generative AI produces media (in plain terms)
Large language models (LLMs) are trained on enormous amounts of text. They learn statistical patterns in language and generate text by repeatedly predicting a likely next piece of text (a token). The result is fluent, but the model is producing plausible language, not checking facts. When it states false information confidently, people call this a hallucination (or confabulation). Some tools connect a model to search results, which can help, but they can still misread or mis-cite sources.
Image generators are commonly diffusion models, trained on many captioned images. They start from random noise and remove it step by step, guided by your prompt.
Voice and video tools can imitate a person's voice from samples, generate realistic video and edit footage (removing objects, extending a scene).
What this means for reading AI media
- Fluency is not reliability. Polished language and realistic images no longer signal effort, expertise or truth.
- Training data carries perspectives. Models learn from data that over-represents some languages, cultures and viewpoints. Tests by researchers and journalists have found image generators producing stereotyped defaults (for example, mostly men for prompts about leaders or engineers). That is a bias issue under expectation 1.5.
- The average aesthetic. Generated content tends toward the most common patterns, which can make it generic.
- Detection is unreliable. AI-text detectors can wrongly flag human writing; studies have found this happens more often with writing by English language learners. Visual glitches are disappearing.
Evaluating AI-generated or AI-assisted texts
Because surface features are unreliable, evaluate accountability and process:
- Who is responsible? Named authors, editors, an address, a corrections policy.
- Is AI use disclosed? Responsible outlets and creators say how they used AI. Several major platforms ask creators to label realistic AI-generated content, and some read content credentials to add labels automatically.
- Can claims be verified? Do quotes, numbers and events appear in independent, reliable sources?
- Provenance: do images carry credentials or a traceable original? Some AI companies also embed invisible watermarks in generated media, which special tools can detect.
- Purpose: why was this made? Sites that mass-produce AI articles often aim to earn ad revenue from search traffic.
Industry and law (1.6)
- Copyright and training data: authors, artists, music companies and news organizations in several countries have sued AI companies over training on their work without permission or payment. In Canada, a group of major news organizations sued OpenAI in Ontario in late 2024; in late 2025 the Ontario Superior Court ruled that the case can proceed there, and OpenAI appealed. These questions are still being decided.
- Labour: AI tools change jobs in writing, illustration, voice acting and translation, which raises questions of consent and payment.
- Likeness and consent: using a real person's face or voice without consent can be harmful and, depending on the use, illegal.
- Who profits: a few large technology companies make most of the leading models, which concentrates power in the media industry.
Worked exampleLet's evaluate the *Lakeshore Daily Wire* article
Evaluate the credibility of the invented library-closure article and explain what kind of text it is.
- Responsibility: no bylines, no staff list, no address, no corrections page. Nobody is accountable. A serious warning sign.
- Disclosure: no statement about AI use, even though the site publishes 40 articles a day with no visible staff.
- Verification: the library's own website and the city's news page have no closure notice; local news outlets haven't reported it. The library spokesperson has no name. The central claim fails verification.
- Provenance: a reverse image search finds no earlier original of the photo, and the picture has no content credentials. That alone proves nothing, but combined with everything else, the image is likely generated or misattributed.
- Purpose: the site is covered in ads and its headlines target common local searches. Likely purpose: ad revenue from search traffic, not public information.
- Classification and judgment: the article is probably AI-generated content presented as journalism; the false closure claim is a hallucination or fabrication. When parents share it in good faith, it spreads as misinformation. The text communicates its idea fluently but fails the purpose a news text claims, because its information is false and unaccountable.
The article is not credible: no accountability, no disclosure, a claim that fails verification, an image with no traceable original, and an ad-revenue purpose. Fluent writing and a realistic photo don't change the verdict, because accountability and verification, not polish, are the tests.
AI and honesty
Using AI to brainstorm or edit is a choice; hiding AI use where audiences expect human work, or presenting unchecked AI output as fact, is the ethical problem. Disclose, verify and take responsibility.
Just for funEarly image generators were famous for giving people the wrong number of fingers. Newer ones rarely do, which is why media literacy teachers had to retire their favourite trick: count the fingers.
Hands-on activityAI output fact-check
You need: access to a chatbot or AI search summary (with a grown-up's OK and following your school's rules), paper and a pencil.
- Ask an AI tool three factual questions about your own community (when a local school was built, what a local landmark is named after, a local sports team's history).
- Copy each answer. Underline every checkable claim (names, dates, numbers).
- Verify each underlined claim using reliable sources (official sites, local news, the library).
- Mark each claim correct, wrong or can't verify. Count them.
- Write a 5-sentence evaluation: how reliable was the tool on local topics, and why might local topics be harder for it?
What to look for: every claim checked, honest can't verify labels, and an explanation such as less local information in the training data.
Your turn
1How does a large language model generate text?
It repeatedly predicts a likely next piece of text (token) based on patterns learned from huge amounts of training text.
2What is an AI hallucination?
When an AI tool states false information confidently, such as an invented quote or date.
3How does a diffusion image model make an image?
It starts from random noise and removes it step by step, guided by the prompt, using patterns learned from captioned images.
4Why is fluency not a sign of reliability?
AI produces plausible language whether or not it is true, so polish no longer signals expertise or fact-checking.
5How can training data create bias in AI media?
If the data over-represents some groups, languages or views, the model reproduces those patterns, such as stereotyped defaults.
6What are content credentials?
A signed record (an open standard from C2PA) of how a file was made and edited, which can travel with the file.
7Name three accountability checks for a possibly AI-generated news site.
Any three: named authors/editors, address/contact, corrections policy, AI-use disclosure, verifiable claims.
More practice: warm-up
2True or false: missing content credentials prove an image is fake.
False. Most images have none; and credentials can be stripped by screenshots.
3Why are AI-text detectors unreliable?
They can wrongly flag human writing, including writing by English language learners.
4Give one industry issue around generative AI.
Any one: copyright of training data, labour/jobs, likeness and consent, concentration of power in a few companies.
5What is a likely purpose of a site that publishes dozens of AI articles a day?
Ad revenue from search traffic.
More practice: on your own
1Explain why tracing the source matters more now than looking for glitches.
Generation tools keep improving, so visual and writing glitches disappear; but fabricated content still lacks an accountable source and a verifiable original.
2A creator posts a realistic AI video of a famous Canadian landmark covered in snow in July, labelled AI-generated. Is this misinformation? Explain.
Not as posted: the label discloses it is synthetic and the purpose seems creative. It could become misinformation if someone re-shares it without the label as a real event.
3Apply expectation 1.5: what perspective might an AI tool carry when asked to describe a typical Canadian family?
It may reflect the most common patterns in its training data, possibly favouring English-language, urban or majority-culture images, and under-representing Indigenous, Francophone, newcomer or rural families.
4Why do news organizations care about AI companies training on their articles?
They argue their copyrighted work is used without permission or payment to build products that may compete with them for readers and revenue.
5Write a short AI-use disclosure for a school media project that used AI to brainstorm titles and generate one clearly labelled example image.
Sample: I used an AI chatbot to brainstorm possible titles and chose and edited the final one myself. The image labelled "AI-generated example" was made with an image generator. All facts were checked by me against the sources listed.
Challenge
1Evaluate this claim: AI-generated media is always less trustworthy than human-made media.
Sample: Too simple. Trust depends on accountability, verification and disclosure, not on the tool. Human-made media can be false; AI-assisted media from an accountable outlet that verifies facts can be reliable. But AI makes fabrication at scale cheap, so unaccountable AI content needs extra caution.
2The liar's dividend means real evidence can be dismissed as fake because fakes exist. Explain how generative AI increases this risk and one way journalists respond.
Because realistic fakes are easy to make, someone caught on a real recording can claim it's AI. Journalists respond by documenting provenance: original files, credentials, multiple witnesses and metadata, and explaining their verification publicly.
3Analyze how the business model of a free AI image app might shape what it produces and how it is used. Refer to industry factors (1.6).
Sample: the app may earn money from subscriptions, ads or data, so it is designed to encourage frequent, shareable creations; it may favour eye-catching styles and minimize friction such as labels. Training on large, possibly unlicensed datasets keeps costs low but raises copyright issues. Its terms may let the company use user uploads to improve models, which raises privacy concerns.
Quiz
1. An AI chatbot confidently invents a quote. This is called:
2. A diffusion image model begins with:
3. Which is the most reliable test of a possibly AI-generated news article?
4. A missing content-credentials record means the image is:
5. Stereotyped defaults in AI image outputs are mainly a result of:
6. Which is an industry issue around generative AI (1.6)?
What this lesson teaches: how generative AI produces text and images (LLMs, diffusion models), why fluency isn't reliability, how training data carries bias, how to evaluate AI-generated texts through accountability, disclosure, verification and provenance, and the industry and legal questions around AI (expectations 1.3, 1.5, 1.6 and 2.1). The Lakeshore Daily Wire site is invented.
How to help at home:
- Do the activity together: ask an AI tool about your community and check the answers.
- Talk about your family's rules for AI use and your school's academic-honesty policy.
- Look for AI labels on the platforms you use.
Common difficulties:
- Believing AI is either magic or useless. It is powerful at plausible patterns and unreliable on facts it hasn't verified.
- Relying on detectors or glitches. Focus on accountability and verification.
- Treating disclosure as optional. Honest use means saying how AI was used.
Note: legal questions about AI and copyright were still unresolved in 2026; facts in this lesson are general and may change.