← Back to course
AI and You 9-12 / Week 07 / Friday
5/6
Week 07 · Privacy in the Machine

Friday

Impact Friday: Privacy versus help
// What a question log reveals
⏱ about 20 min

Friday: Impact Friday: Privacy Versus Help

Librarian Joss opens the panel meeting with a new question. "Some parents asked us this. Should SH-1 alert a teacher when a student studies past midnight?"

Comet answers fast. "Yes! If someone is up that late, a teacher could help."

Then she stops. "But that means keeping names and times, the exact things we want to drop."

Wren taps his wristwatch. "Both sides care about the student. What does the evidence say about each choice?"

Teacher Lin adds, "Some students study late because that is when the house is quiet."

Nova projects two columns titled Privacy and Wellbeing. "Would you like a hint?" she asks. "List what each choice protects and what it risks."

Wren looks at you. "This one is a real dilemma. What would you advise?"

When privacy and wellbeing pull apart

The CSTA standards describe a kind of ethical dilemma.

It happens when one person's right to privacy is at odds with a community's safety, security or wellbeing.

The midnight alert is that kind of dilemma. It could help a struggling student, but it would watch every student at night.

This is a made-up example from the SH-1 story. Real panels weigh questions like it.

ChoiceWhat it protectsWhat it risks
Alert a teacher after midnightA student who is struggling could get helpEvery student is watched at night; names and times must be kept
No alert; keep topic totals onlyEvery student's late-night habits stay privateA struggling student might not be noticed by SH-1
WEIGH THE DILEMMA
  • Read the question.
  • Tap your answer.
Which of the crew's fixes would the midnight alert break?
Teacher Lin says some students study late because the house is quiet. What does that show?

The crew's advice: no midnight alert. A teacher who is worried about a student talks with them in person.

You may disagree. A good answer weighs both sides and explains its reasons.

Rate R3: data privacy

Risk combines how likely a harm is with how big it would be. The crew rates R3 using this week's evidence.

Likelihood: High. P5 says SH-1 logs every question with a name, class and time.

Size of harm: High. The lab showed a log can point to one person, and personal information can leak or be misused.

RowRiskSH-1 exampleLikelihoodHarm
R3Data privacyLog stores names, class and timeHighHigh
MATCH EACH ACTION TO ITS FUNCTION
  • Read the question.
  • Tap your answer.
The panel writes a rule for the school: no names in any AI study log.
The crew lists what the log keeps and who could see it.
Each month, someone checks a sample of the log for names.
The crew plans what to do and who to tell if the log ever leaks.
AI actor spotlight: privacy governance
NIST lists privacy experts among the people who do AI design work: writing down goals, gathering and cleaning data, and documenting it.
NIST also lists legal and privacy governance experts among the people who do AI development work: building, training and testing models.
In the SH-1 review, the crew is doing privacy governance work: asking what is kept, why, and who can see it.

The model card: privacy section

The crew adds its six fixes to the SH-1 model card as the privacy section.

  • No names in the log.
  • IDs that change each week.
  • Keep topic totals, not times.
  • Delete the raw log after 14 days.
  • Never train SH-1 on student questions.
  • Tell students exactly what is kept.
On paper, write the privacy section of your SH-1 model card. List the six fixes and mark the one you think matters most.

Week review

  1. Privacy protects autonomy, identity and dignity. Anonymity, confidentiality and control should guide AI design.
  2. Data can be collected without people noticing, and training on personal data strains consent, transparency and purpose.
  3. Models can reveal training data and guess private facts. A wrong guess can still hurt.
  4. De-identification and aggregation help, but every privacy fix gives something up.
  5. Sometimes privacy and wellbeing pull in different directions.
  6. R3 data privacy: likelihood High, harm High.
Week checkTrue or false?
The crew rated R3 High for likelihood and High for harm.?
One person's privacy can be at odds with a community's wellbeing.?
The midnight alert would need no names or times.?
WHY THIS EXERCISEThe rating and the dilemma are what the crew brings to the panel.

What a week, reviewer. Tomorrow is the Explorer Quest: a family game about what a helper really needs to know.

← Thursday