Ethical issue: repurposing and sharing of data by AI systems

Name:

Date:

Class:

Quick Questions

Shade in the A or B circle for each question.

1. A student uploads a photo to a social app to share it with friends. The company later uses the photo to train an AI model without telling users. What is this?

A Normal use: data collected for one purpose can always be used for anything

B Data repurposing: the data is reused for a different purpose without consent

2. Which situation is an example of sharing data with a third party by an AI system?

A A chatbot company sells user conversations to an advertising company

B A chatbot shows a user their own conversation history on their own account

3. True or false: Once a person's data has been used to train an AI model, it is easy to remove just that person's data from the model.

A False: removing data from a trained model is very difficult

B True: the data can be deleted from the model with one click

4. Which option best describes an ethical way to use user data for AI training?

A Hide the training use in the terms and conditions so users are unlikely to notice

B Clearly tell users and get their consent before using their data for training

5. Mia deletes her account. Her posts were already used to train an AI. What is the likely problem?

A Her data may still influence the trained model even though her account is gone

B The AI instantly forgets everything it learned from her posts

Match the Pairs

Draw a line connecting each item on the left to its match on the right.

Left column		Right column
1Policy: chats may be used to train models		AUser data is passed on to another organisation, which the user may not know about or agree to
2share_with(third_party, user_data)		Bcollected_for = 'support'; used_for = 'AI training'
3Once a model is trained, it is hard to remove one person's data from it		Cdelete(user_row) // model still keeps what it learned
4Data collected for one purpose is reused for a different purpose such as AI training		DDeleting the database record does not remove what a trained model has already learned from the data. Removing it from the model is hard.
5if consent == 'support': train_ai(chats)		EConsent was given for support only. Using the chats to train an AI is a new purpose, so the person must be told and asked for consent again.
6delete(my_row) // so the AI forgets me		FData collected from users for one purpose is reused as AI training data, often without them really knowing

Definitions

Data repurposing

Fill in the Blanks

Complete each line of code by filling in the missing parts.

IF consent = _____ THEN DISPLAY "Do not reuse data" END IF

IF newPurpose != oldPurpose THEN ___ consent END IF

_____ userData WITH partner LOG "Third party has data"

model = _____(userData) DELETE userData OUTPUT "Model kept the data"

IF _______ = TRUE THEN OUTPUT "Deleting will not undo it" END IF

Spot the Error

Shade in the letter of the line that contains an error.

Question 1

A consent ← user.trainingOK

B IF consent = FALSE THEN

C ADD chat TO trainingSet

D END IF

Question 2

A IF user.shareOK = TRUE THEN

B SEND data TO partner

C ELSE SEND data TO partner

D END IF

Question 3

A note ← "Chats are not stored"

B SHOW note TO user

C ASK user FOR consent

Question 4

A DELETE record FROM trainingSet

B // model already learned it

C SET record.inModel ← FALSE

D // slow and costly

Question 5

A purpose ← "support"

B data ← COLLECT(chatLogs)

C USE data FOR training

Answer Key

Ethical issue: repurposing and sharing of data by AI systems

Section A: Quick Questions

B — Data repurposing: the data is reused for a different purpose without consent

A — A chatbot company sells user conversations to an advertising company

A — False: removing data from a trained model is very difficult

B — Clearly tell users and get their consent before using their data for training

A — Her data may still influence the trained model even though her account is gone

Section B: Match the Pairs

Policy: chats may be used to train models → Data collected from users for one purpose is reused as AI training data, often without them really knowing

share_with(third_party, user_data) → User data is passed on to another organisation, which the user may not know about or agree to

Once a model is trained, it is hard to remove one person's data from it → delete(user_row) // model still keeps what it learned

Data collected for one purpose is reused for a different purpose such as AI training → collected_for = 'support'; used_for = 'AI training'

if consent == 'support': train_ai(chats) → Consent was given for support only. Using the chats to train an AI is a new purpose, so the person must be told and asked for consent again.

delete(my_row) // so the AI forgets me → Deleting the database record does not remove what a trained model has already learned from the data. Removing it from the model is hard.

Section C: Definitions

Data repurposing — When an AI system reuses or shares data collected for one purpose, such as training, without the knowledge or consent of the people it came from. Removing it from a trained model is hard.

Section D: Fill in the Blanks

IF consent = FALSE THEN
  DISPLAY "Do not reuse data"
END IF

IF newPurpose != oldPurpose THEN
  ASK consent
END IF

SHARE userData WITH partner
LOG "Third party has data"

model = TRAIN(userData)
DELETE userData
OUTPUT "Model kept the data"

IF trained = TRUE THEN
  OUTPUT "Deleting will not undo it"
END IF

Section E: Spot the Error

B — The chat should only be added to the training set when the user has consented. Testing for FALSE reuses data from users who said no. It should be IF consent = TRUE THEN.

C — When the user has not agreed to sharing, the ELSE branch must keep the data private. Sending it to a third party anyway shares data without consent. It should be ELSE KEEP data PRIVATE.

A — The notice is misleading because the chats are used to train the AI. Users cannot give informed consent if they are not told this. It should say the chats train the AI.

C — A trained model cannot forget data just by changing a flag, because it has already learned patterns from it. The model must be retrained without the record: RETRAIN model WITHOUT record.

C — The data was collected for support. Using it for training is repurposing without the users' knowledge or consent. It should be USE data FOR purpose.