# Quiz — Session 6 🟢 “AI and ethics”

**Program:** Applied AI — Beginner Level · **Instructor:** Yann Isola
**10 questions · Only one correct answer per question · Answer key commented at the end**

---

### Question 1
Why can AI be biased (unfair)?

- A. Because she is programmed to be mean
- B. Because it learns from data that already contains imbalances or stereotypes
- C. Because it breaks down from time to time
- D. Because she doesn't have enough electricity

### Question 2
A company used AI to sort CVs. AI disadvantaged women. What was the most likely cause?

- A. The AI hated women
- B. A hacker had modified the AI
- C. The AI had learned about the company's past recruitments, which mainly hired men
- D. Women write worse CVs

### Question 3
Where does the data used to train large AI mainly come from?

- A. They are invented by computers
- B. From our photos, texts, searches, clicks… published or generated by us online
- C. Only very old books
- D. Aliens

### Question 4
What does GDPR mean?

- A. General Network of Data Hackers
- B. General Data Protection Regulation
- C. Giant Robot Programmed to Detect
- D. General Rule of Digital Advertising

### Question 5
Which of these rights is guaranteed by the GDPR in Europe?

- A. The right to use all apps for free
- B. The right to ask a company to delete your personal data
- C. The right to have faster internet
- D. The right to copy films

### Question 6
What is the “black box” problem in AI?

- A. AI records everything like in an airplane
- B. The AI only works at night
- C. We cannot always explain why the AI made a specific decision
- D. The AI is hidden in a box to protect it

### Question 7
Where is AI's lack of transparency MOST serious?

- A. Music playlist recommendations
- B. Funny photo filters
- C. Health, justice and bank credit
- D. Video games

### Question 8
A self-driving car causes an accident. Which of these statements is correct?

- A. The car's AI will be tried in court
- B. No one is responsible, it's bad luck's fault
- C. Responsibility is discussed between humans and companies (driver, manufacturer, etc.), because an AI is not a legal person
- D. It is always the pedestrian who is responsible

### Question 9
What do we say about the environmental impact of training very large AIs? ⚠

- A. It is useless: a computer consumes nothing
- B. It is important: lots of electricity and CO2 emissions comparable to several transatlantic flights
- C. AIs produce electricity
- D. Only small AIs consume energy

### Question 10
The European AI Act (2024 ⚠) classifies AI by level of risk. What use is INTERDIT (unacceptable risk)?

- A. A spam filter for emails
- B. A chatbot that clearly indicates that it is an AI
- C. The generalized social rating of citizens by a State
- D. An AI that recommends films

---

## Fixed commented

**1. B** — The AI ​​has no intention or malice: it faithfully reproduces what its training data contains, including imbalances.This is the golden rule: *an AI is never more accurate than the data that fed it.*

**2. C** — Nobody had planned to “reject women”. The AI ​​learned from 10 years of history where the company recruited mainly men, and deduced this (bad) model of the “good candidate”. Real case, tool finally abandoned.

**3. B** — Large AIs learn largely from content produced by humans: photos, texts, reviews, clicks, searches. Often ours — hence the question of consent.

**4. B** — GDPR = **General **Regulation on the **Protection of **Data. In force in Europe since 2018. In France, the police force which enforces it is the CNIL (National Commission for Information Technology and Liberties).

**5. B** — The “right to erasure” (right to be forgotten) is one of the key rights of the GDPR, with the right of access (“show me my data”) and the right of rectification (“correct what is wrong”).

**6. C** — Large AIs make their calculations via millions of automatically adjusted internal settings: the result works, but the precise explanation of a given decision is very difficult to provide. This is the “black box”.

**7. C** — The more a decision impacts a human life (getting treatment, being judged, obtaining credit), the more explanation and human control are essential. A failed playlist, we get over it; a credit refused without reason, no.

**8. C** — An AI has no conscience, no assets, no legal personality: it can neither be judged nor punished. Responsibility is shared between humans and companies (driver, manufacturer, sometimes authorities), depending on the cases and the manufacturer's promises. The law is currently being developed on this subject. ⚠

**9. B** — Training a very large model consumes a lot of electricity (and water for cooling data centers), with a CO2 footprint compared to several transatlantic flights in several estimates. ⚠ The exact figures are evolving — but the order of magnitude justifies the question: “is this use worth the energy?” »

**10. C** — Generalized social rating of citizens is one of the uses **prohibited** by the AI ​​Act (unacceptable risk 🔴). The spam filter is minimal risk 🟢, the transparent chatbot is limited risk 🟡, the movie recommendation is minimal risk 🟢. Memory tip: the more AI affects freedom, money, health or education, the higher it rises in the risk pyramid.

---

### Indicative scale

- **9–10 ✅**: Honorary Judge of the AI Tribunal ⚖️
- **7–8 ✅**: Brilliant lawyer — convincing argument!
- **5–6 ✅**: Attentive jury member — reread the risk pyramid
- **< 5**: Return to the file instruction: review the slides and redo exercise 1 (Bias Detective) 🕵️