Skip to content
Idaho AI Classroom
All lessons
AI ImpactEthics and Privacy9-12.AII.EP.4

Used Without Asking

11–1250 minutes90 minutes (extend seminar and writing)

Standard quoted exactly

Assess how unauthorized data collection has influenced the practice of training AI models.

Example from the standards. An AI model is trained on art without the artist's permission which can lead to legal implications and can change the training process.

Student-friendly learning targets

  • I can assess how training on writers’, artists’, photographers’, and public social posts without permission has changed both lawsuits and training practice.
  • I can explain why fair use is a contested argument in this setting, not a rule I can recite as settled law.
  • I can describe at least one way a training process might change in response to unauthorized-data claims (licensing, filtering, attribution, opt-out).

Essential questions

  1. If a work was posted or published, was it offered as training fuel?
  2. What is the difference between a fair-use argument and a property argument, and why have courts not handed schools a slogan?
  3. How should training practice change when permission was never asked?

Objectives

  1. Assess unauthorized collection across four source types: writers, artists, photographers, and public social posts.
  2. Present fair use and licensing/property claims as competing arguments with evidence, not as a verdict.
  3. Trace at least one legal implication to a concrete change in training practice (data filtering, licensed corpora, opt-out tools, synthetic data).
  4. Evaluate a student-posted photo or caption as a training example without using any real classmate’s account.
  5. Keep student work and names out of live models while discussing scraping of public web data.

Key vocabulary

Unauthorized collection
Gathering works or data for training without a license or a clear, informed agreement from the person who made or is depicted in them.
Fair use
A U.S. copyright doctrine that sometimes allows limited use of protected works without permission. Whether mass training qualifies is contested in courts and legislatures.
License
A permission to use a work under stated conditions, often in exchange for pay, credit, or limits on reuse.
Training corpus
The assembled collection of texts, images, or other examples used to train a model.
Opt-out
A mechanism that asks rightsholders to request removal after collection, shifting the burden onto the person who did not consent.
Style imitation
A model producing work that resembles a living creator’s distinctive manner, raising questions beyond exact copying.
Provenance
The record of where a training example came from and under what permission, if any.
Transformative use
A fair-use factor arguing that the new use serves a different purpose than the original; companies invoke it for training, and many creators reject it at this scale.

Teacher background

The standards example is an AI model trained on art without the artist’s permission, with legal implications that can change the training process. Widen that example in class to writers, photographers, and people who posted in public. Students should assess influence on practice: filtering copyrighted domains, buying licensed libraries, offering opt-outs, watermark research, and the use of synthetic data. They should not memorize a holding as if the Supreme Court had settled classroom policy. Fair use is a real doctrine with four factors. Whether copying millions of whole works to build a commercial generator is transformative remains a live dispute. Teach both arguments with equal care. A property argument says the copies were made to create a competing product without pay. A fair-use argument says the model learns statistical patterns, does not replace any single book in the ordinary sense, and would be impossible if every token needed a handshake. Neither argument is a personality test. Use public docket summaries and creator statements, dated. Do not download pirated book heaps with students. Do not scrape Instagram. Do not feed student art or essays into a generator to prove a point. FERPA and ordinary consent still apply. Offline fallback: printed creator statements, a blank four-factor fair-use grid, and fictional but realistic posting scenarios.

Materials and prep

Materials

  • Four source packets: Writer, Artist (the standards example), Photographer, Public social post.
  • Fair-use factor grid (purpose, nature, amount, market effect) labeled Contested — not a verdict sheet.
  • Training-practice change cards: licensed corpus, domain filter, opt-out registry, attribution, paid residuals, synthetic data.
  • One public artist statement and one public company or researcher statement, both dated, both excerpted fairly.
  • Offline fallback: no live image-generator demos that imitate a living artist’s name; describe the practice in prose if needed.

Before class

  • Remove any living local student’s work from examples. Professional public cases only, plus fictional teen posting scenarios.
  • Read enough of a current case summary to avoid stating a reversed holding as law. If the law moved, say the date.
  • Print the standards example at the top of the Artist packet.
  • Block plan: full hearing with opening statements, cross on fair-use factors, and a written judgment that still admits uncertainty.

Instructional sequence

Posted is not the same as offered

5 min
  1. Display a fictional public photo: a student posts a concert picture from the Idaho Center. A vision model later trains on it. A generator produces a similar crowd scene for an ad.
  2. Students mark: permission, implied permission, no permission. No discussion yet.
  3. Add a second beat: a living illustrator’s portfolio, scraped, then a prompt in the style of…
  4. Read the standard and the art example. Tell students the law is disputed and that their job is assessment, not a fan vote.

Four sources, two arguments, one training pipeline

10 min
  1. Name the four sources on the board. Unauthorized collection can be illegal, legal-but-contested, or merely unethical; students must not collapse those words.
  2. Teach fair use as four factors. Write Contested in large letters. Schools do not get a mnemonic that replaces litigation.
  3. Teach the property/licensing argument: wholesale copying to build a substitute market, unpaid.
  4. Show how practice already shifts when claims land: companies negotiate licenses with some publishers, publish opt-out forms, or claim they filtered certain domains. Those are influences on training, which is what the standard asks.
  5. FERPA note: a student’s unpublished essay in a learning-management system is not a public post. Different legal and ethical floor.

Argue the art example without a verdict

10 min
  1. Read the artist packet together. Fill the four-factor grid in two columns: strongest fair-use claim, strongest property claim.
  2. Ask which training-practice change would actually respond to the artist (a license, a filter, an opt-out after the fact, or nothing).
  3. Refuse a class vote on who won. The assessment is whether students can state both arguments with evidence.
  4. Point out style imitation: even if no single pixel is copied, a living artist’s market may still be affected. That is part of the dispute, not a solved lemma.

Assess a second source

12 min
  1. Pairs take Writer, Photographer, or Public social post and complete: unauthorized how, legal implication, likely change to training practice.
  2. Social-post pairs must address the teen-concert scenario without using real usernames.
  3. Writer pairs must address books and journalism, not only famous novelists: local reporters’ work can be in corpora too.
  4. Exit product: a twelve-line assessment that uses the word contested correctly.

Real-world examples

  • Visual artists discovering prompts that target their names after portfolio sites were crawled.
  • News organizations and book authors suing over wholesale copying of text into training sets.
  • Wedding and sports photographers whose images, once online for clients, reappear as generic stock-like generations.
  • Public social posts, including Idaho event photos, treated as free vision data because the privacy setting was public.
  • Companies announcing licensed deals with some publishers while still disputing fair use for other corpora — a change in practice under legal pressure.

Hands-on activity

Hearing: change the training process

8 min
  1. Each group draws a practice-change card and must argue how it would, or would not, answer unauthorized collection for their source type.
  2. Opt-out groups face a challenge: why is the burden on the artist after the scrape?
  3. Licensed-corpus groups face a challenge: who cannot afford to be in the licensed set, and whose work then disappears from culture the model can see?
  4. Record one unresolved question per group on the board. Uncertainty is a legitimate academic product.

Discussion questions

  1. Is a public Instagram photo more like a published newspaper (in the world) or more like a conversation that happened to be visible?
  2. If fair use protected training completely, what would happen to working illustrators’ commissions? If it banned training completely, what would happen to research?
  3. Should a student who posts a poem in a public hashtag expect it not to become a training sentence?
  4. Does paying some publishers and not others make the training process more ethical or only more expensive?
  5. How is a school essay stored in a district system different from a public post?

Differentiation

Support

  • Give a two-column notes sheet: Fair-use argument / Property argument, with factor prompts.
  • Allow students to assess the art example in greater depth instead of switching sources.

Challenge

  • Write a model card addendum describing provenance and known unauthorized sources, with a recommended mitigation.
  • Compare opt-out and opt-in as governance designs and argue which one matches the word permission.

Multilingual learners

  • Note that non-English writers are often in corpora without local legal support. Invite reflection without requiring personal publishing history.
  • Provide a glossary of copyright, license, and fair use with examples, not idioms.

IEP / 504

  • Provide large-print factor grids and a quiet copy of excerpts.
  • Accept a recorded oral assessment if writing speed is the barrier; content standards stay the same.

Assessment

Formative

  • Two-column fair-use grid from guided practice.
  • Listening for contested versus is illegal as a slogan.

Summative

  • Twelve-line assessment of a source type linking unauthorized collection to a training-practice change.
  • Block extension: a written judgment of 350 words that states both arguments and still ends in a recommended practice, with uncertainty named.

Success criteria

  • Treats fair use as contested, with at least one factor on each side.
  • Includes writers, artists, photographers, or public posts with a concrete example.
  • Names a change to training practice, not only a feeling about theft.

Responsible use, ethics, and privacy

Responsible use

Do not run live prompts that name a living artist in order to imitate them. Do not download unauthorized book datasets. Do not scrape social platforms. Fictional teen posts and public, already-reported cases are enough. Offline packets are preferred.

Ethics

Creators’ unpaid labor and researchers’ need for large corpora are both facts. The ethical failure is pretending only one fact exists. Students should leave able to describe how unauthorized collection pushed some firms toward licenses and filters — influence on practice — without declaring the legal war over.

Privacy

Student essays, art, and photos stored by the school are education records or school work, not public web scrapes. FERPA and district policy limit disclosure. A public social post is still a person’s data; class will not harvest classmates’ accounts to illustrate the point. Use invented scenarios.

Reflection

  1. Which source type felt most like unauthorized collection to you, and which factor made the difference?
  2. What would an honest model card say about your own public posts?
  3. Where did you feel tempted to treat contested law as settled, and what pulled you back?

Homework

On paper, write a twelve-line policy for a student literary magazine: may archived poems be used to train a model, by whom, with what permission, and what happens if a graduate later objects. Do not paste the poems into a tool.

Closing

Return to Posted is not offered. Collect assessments. Name the four sources once more. Next class turns to cybersecurity, where the same dual-use pattern — a tool that writes and a tool that detects — has benefits, risks, and ethics of its own.

Extensions and cross-curricular links

Go further

  • 90-minute block: structured hearing plus a written judgment that a school board could read.
  • Invite a local photographer, journalist, or art teacher to describe commission markets; record a backup.
  • Compare two companies’ public training-data statements dated a year apart; list what changed.
  • Draft an opt-in policy for student publications (yearbook, literary magazine, news site) covering future training uses.
Visual arts
Portfolio posting, style, and commission markets after generators.
English / journalism
Copyright, quotation, and whether a model that read a paper replaces the paper.
Government / law
How unsettled federal cases should be taught: competing arguments, dated holdings, no fake certainty.
Photography / CTE media
Client contracts, public galleries, and scraping of images posted for a different purpose.