Skip to content
Idaho AI Classroom
All lessons
AI ImpactEthics and Privacy9-12.AII.EP.2

Who Gets a Better Door

11–1250 minutes90 minutes (extend seminar and writing)

Standard quoted exactly

Analyze how AI tools shape user experiences for people with diverse backgrounds and characteristics.

Example from the standards. Real-time translation tools and AI-driven accessibility features (like speech-to-text for the hearing impaired or image-to-text descriptions for the visually impaired) remove barriers for non-native speakers and people with disabilities.

Student-friendly learning targets

  • I can analyze how the same AI feature creates different experiences depending on language, disability, accent, bandwidth, and cost.
  • I can give a precise example of an accessibility gain and a precise example of a remaining barrier or new harm.
  • I can recommend a design or policy change that would make a tool more usable without treating any group as a test afterthought.

Essential questions

  1. When an AI tool opens a door, who walks through first, and who still finds a step they cannot climb?
  2. How do translation, captions, and image descriptions both remove barriers and introduce new errors?
  3. What would it mean to design AI so that diverse users are expected users, not exceptions?

Objectives

  1. Analyze at least three user profiles (language, sensory disability, and one additional characteristic such as accent, literacy, rural bandwidth, or age) against the same AI feature.
  2. Distinguish access (can the person use the tool) from quality (does the tool work as well for them as for the default user).
  3. Evaluate a translation or accessibility feature for both barrier removal and error risk in high-stakes settings (health, legal, school).
  4. Propose one design or school-policy change that treats disabled and multilingual users as intended users.
  5. Follow FERPA and disability-privacy rules: no student is required to disclose a diagnosis to participate.

Key vocabulary

Accessibility
The degree to which a product can be used by people with a wide range of abilities, including vision, hearing, mobility, and cognitive differences.
Default user
The unspoken person a product was designed around, often English-speaking, sighted, hearing, urban, and well connected to broadband.
Real-time translation
Automatic conversion of speech or text from one language to another with little delay, useful and error-prone in the same moment.
Speech-to-text
A system that converts spoken language into written text; accuracy often varies by accent, microphone, and background noise.
Image-to-text / alt text
A description of an image generated for people who cannot see it, or who need text that a screen reader can speak.
Inclusive design
Building for the edges of human variation from the start, rather than adding an accommodation after a product ships.
Error cost
What happens when the tool is wrong: a missed joke is cheap; a mistranslated medical or legal phrase is not.

Teacher background

The standards example is clear: real-time translation, speech-to-text, and image descriptions can remove barriers for non-native speakers and for people with disabilities. That gain is real and should be taught as real. The analysis the standard asks for is not a celebration of features. It is a study of how the same tool shapes experience differently depending on who the user is. A captioning model that works on a quiet classroom recording may fail on a gym pep rally. A translation app that handles cafeteria Spanish may mangle a parent letter about special education rights. A voice assistant that hears a Treasure Valley newscaster may not hear a student whose first language is Somali or whose speech is dysarthric. Rural bandwidth is also a characteristic of users in Idaho: a feature that requires a live connection is not an open door in every district. Teach access and quality as separate questions. Teach that disabled and multilingual people are expected users, not afterthoughts. Do not require any student to disclose a disability, language spoken at home, or immigration status. Use public, teacher-created personas. FERPA and disability records stay closed. Offline fallback: printed screenshots of captions, alt text, and translation output, plus a paper user-experience matrix.

Materials and prep

Materials

  • Printed user-experience matrix: rows are features (translation, speech-to-text, image description, voice assistant); columns are user characteristics.
  • Five teacher-written personas on cards: a parent who prefers Spanish; a student who is Deaf and uses captions; a student who is blind and uses a screen reader; a student with a rural satellite connection; a student whose accent the model often misses. No real student data.
  • Printed screenshots: a good alt-text example, a bad or invented alt-text example, a translation that is fluent but wrong on a rights-related phrase.
  • Error-cost scale: Low (inconvenience) to High (health, legal, discipline, or rights).
  • Offline fallback: no live microphone tests on students; all audio examples are teacher-recorded public-domain or staff-read text.

Before class

  • Audit personas so none resemble a specific student in the room.
  • If you demonstrate captions, use a school device and a prepared clip, not a student speaking on the spot.
  • Print a short excerpt of the district accessibility or language-access policy if one exists.
  • Tell the special education case manager you will not be asking students to self-identify.
  • Block plan: extend the seminar to argue whether schools should require human review of AI translation for IEP and discipline documents.

Instructional sequence

Same door, different step

5 min
  1. Show a photo of a school entrance with an automatic door, a curb cut, and a narrow side gate. Ask: who is this entrance for?
  2. Translate the metaphor: software has doors too. Captions, translation, and alt text are doors. So are default English menus and required live video.
  3. Students jot one AI feature they use and one person for whom that feature might fail. No names of classmates.
  4. Read the standard and the state’s example aloud. Tell students analysis means gains and remaining barriers.

Access is not the same as quality

10 min
  1. Define accessibility, default user, and error cost with the state’s three examples: translation, speech-to-text, image-to-text.
  2. Walk a Magic Valley dairy-family example: a Spanish-language auto-translation of a school attendance letter that turns unexcused into a harsher legal term. The door opened; the quality failed; the error cost is high.
  3. Walk a captions example from a noisy gym livestream of a playoff game: access exists on paper, quality collapses with crowd noise.
  4. Name rural bandwidth as a characteristic, not a complaint: an Idaho student on a metered satellite link does not have the same experience as a Boise student on fiber.
  5. Set the privacy rule: we analyze personas, not classmates’ medical or language histories.

Score one persona as a class

10 min
  1. Project the screen-reader persona using image-to-text on a science worksheet that is actually a photographed diagram with no true alt text.
  2. Fill the matrix: access (the tool runs), quality (the description misses the labeled parts of the cell), error cost (the student cannot study the diagram).
  3. Ask: is the ethical failure the student’s, the teacher’s, or the tool’s training and design?
  4. Model a recommendation: require human-written alt text on instructional images; treat AI description as a draft, not a substitute.

Matrix for three personas

12 min
  1. Pairs receive three persona cards and complete the matrix for one feature per persona.
  2. Each pair must record one gain that matches the standards example and one remaining barrier or new harm.
  3. Write a ten-line analysis: how the tool shapes experience, not whether the student likes the brand.
  4. A third of the class should take the rural-bandwidth persona so Treasure Valley defaults do not dominate.

Real-world examples

  • District auto-translation of a parent newsletter that is serviceable for sports schedules and unreliable for special-education rights.
  • Live captions at a school board meeting that help Deaf attendees and also leak errors into the public minute-taking stream.
  • Image descriptions that call a tribal regalia photo a costume, a harm that is cultural as well as technical.
  • Speech-to-text that underperforms for students with regional or additional-language accents during oral quizzes.
  • A premium accessibility tier that works well only for families who can pay, turning a door into a ticketed entrance.

Hands-on activity

Redesign the door

8 min
  1. Groups pick one high error-cost cell on the matrix and write a three-part redesign: product change, school practice, and human backup.
  2. Post redesigns. Peers star any proposal that still treats disabled users as extra.
  3. Teacher reads one strong proposal that names a human backup for IEP or medical language.
  4. Collect matrices. If time remains, vote on which redesign a school could implement this year without a new vendor.

Discussion questions

  1. If a translation is free and instant, why might a school still need a human interpreter for a discipline meeting?
  2. Who is the default user of the tools this school actually issues?
  3. When alt text is wrong, is that a technical glitch or a civil-rights problem?
  4. Should families have to disclose a disability to receive a tool that everyone might benefit from (captions, transcripts)?
  5. How should rural bandwidth change what a district calls an equitable digital experience?

Differentiation

Support

  • Provide a completed sample matrix row. Students fill two empty rows by analogy.
  • Allow bullet analysis instead of a ten-line paragraph.

Challenge

  • Research WCAG-style alt-text guidance and rewrite the science-diagram description to a standard a screen-reader user could study from.
  • Write a one-page memo to a principal on when AI translation is acceptable and when a qualified interpreter is required.

Multilingual learners

  • Invite students to judge a printed translation into a language they know, if they choose to, without being appointed class translator.
  • Provide the matrix headings in English with student-generated glosses.

IEP / 504

  • Do not call on students with disabilities to testify. Personas carry that load.
  • Offer extended time on the written analysis and a quiet copy of all screenshots.

Assessment

Formative

  • Matrix cells checked for both a gain and a barrier.
  • Listening for whether students separate access from quality.

Summative

  • Ten-line analysis plus one redesign proposal scored on specificity, error cost, and a human backup.
  • Block extension: seminar argument on AI translation of IEP documents, with a written position of 250–400 words.

Success criteria

  • Analyzes at least two characteristics, not a single disabled-user stereotype.
  • Uses the standards examples (translation, speech-to-text, or image description) with a concrete error-cost judgment.
  • Recommends a change that includes a human backup for high-stakes language.

Responsible use, ethics, and privacy

Responsible use

Do not record classmates to test speech-to-text. Do not upload photos of students to an image-description tool. Teacher-prepared public or staff-created media only. Offline packets meet the standard. If a district-approved captioning tool is demonstrated, the teacher operates it.

Ethics

Accessibility gains in the standards example are genuine and should be stated as genuine. Analysis still requires naming who remains outside the door, including people whose accents, languages, bandwidth, or bodies were scarce in training data. Charity framing (helping them) is weaker than rights framing (the product was incomplete).

Privacy

Disability status, IEP content, and home language are sensitive. FERPA covers education records; additional disability-privacy norms apply. Students must not be asked to disclose diagnoses to complete the matrix. Personas are fictional. Do not put a real student’s accommodation plan into any AI tool.

Reflection

  1. Which characteristic was easiest to forget when you pictured the default user?
  2. Where does this school already provide a human backup for machine translation or captions?
  3. What is one feature you use that probably works better for you than for someone else in this building?

Homework

On paper, pick one tool you used this week. Complete a mini-matrix for two fictional users who are not you. Note access, quality, and error cost. Do not test the tool on a person without their consent, and do not upload anyone’s voice or image.

Closing

Return to the automatic-door image. A door that opens for some and sticks for others is still a design choice. Collect matrices. The next lesson asks what it costs, in energy, water, consent, and labor, to collect the data those doors were trained on.

Extensions and cross-curricular links

Go further

  • 90-minute block: seminar on whether schools should forbid AI-only translation of special-education and discipline documents, then a policy paragraph.
  • Audit five pages of the school website for alt text using a teacher-controlled checker; report patterns, not student work.
  • Interview (with permission) a district interpreter or captioner about what tools help and what they still must redo.
  • Compare mobile captions on a pep-rally video versus a classroom lecture; quantify missed proper nouns.
World languages
What translation can and cannot do; when a fluent wrong phrase is worse than no translation.
Health science
Error cost of medical speech-to-text and why clinics still need human review.
Civics
Language access at public meetings and school-board hearings as a civic, not merely technical, duty.
Special education / advisory
Captions, transcripts, and alt text as ordinary instruction, not a favor.