AI Ethics, Safety and Law
Practice questions
Which statement gives the most balanced view of AI and jobs?
- AI will certainly end all jobs within a few years
- AI never affects anyone's work
- Only manual jobs can ever be affected by AI
- AI changes tasks within jobs, routine tasks are most exposed, and reskilling helps people move to new work
Answer and explanation
Answer: D
AI tends to change tasks before whole jobs, with routine tasks most exposed, while new kinds of work also appear; reskilling and upskilling help people adapt. The other statements are extreme claims that are not supported.
How does 'human-on-the-loop' differ from 'human-in-the-loop'?
- In human-on-the-loop there is no human at any stage
- In human-on-the-loop a person must approve every single decision first
- In human-on-the-loop the system acts on its own while a person monitors and can step in
- There is no difference; the terms mean exactly the same
Answer and explanation
Answer: C
In human-in-the-loop a person approves before a decision takes effect. In human-on-the-loop the system acts by itself while a person watches and can intervene. A system with no human at all is human-out-of-the-loop.
Which of these best describes a risk-based approach to AI rules such as the European Union's AI Act?
- Uses with unacceptable risk are prohibited, high-risk uses face strict duties, and lower-risk uses face lighter duties
- Every AI system faces exactly the same rules whatever its use
- Only AI made outside the region is regulated
- AI is regulated only after it has caused harm
Answer and explanation
Answer: A
A risk-based approach matches the strictness of rules to the possible harm of each use, from prohibition down to light transparency duties. Identical rules for all uses, rules only for foreign AI, or rules only after harm do not describe it.
A family agrees on a secret code word to be used on calls in an emergency. This mainly protects against:
- Slow internet speeds
- Voice-clone emergency scams
- Bias in hiring tools
- Data minimisation failures
Answer and explanation
Answer: B
A cloned voice can sound exactly like a relative, but the fraudster will not know the agreed code word, so the family can tell a real emergency from a fake one. It has nothing to do with internet speed, hiring bias or data collection.
Which check is most useful to find out whether a viral clip of a public figure is a deepfake?
- Counting how many likes it has
- Checking whether it has background music
- Looking for the original on the person's official channels and checking reliable news sources or fact-checks
- Checking whether the video is in high definition
Answer and explanation
Answer: C
Authenticity is established by tracing the source: official channels and independent verification. Likes, background music and picture quality say nothing about whether the content is real; good fakes can be high definition and very popular.
Ravi realises he has just been cheated of money through a fake UPI request. What should he do first?
- Wait a week to see whether the money comes back
- Delete the chat so he can forget about it
- Pay the fraudster a 'refund fee' to get the money back
- Report at once on 1930 or cybercrime.gov.in, inform his bank and keep screenshots and transaction IDs
Answer and explanation
Answer: D
Reporting quickly on 1930 or cybercrime.gov.in improves the chance that the money can be traced and blocked, and evidence helps the investigation. Waiting loses time, deleting chats destroys evidence, and a 'refund fee' is a second fraud.
Which statement about the Puttaswamy judgment is correct?
- A nine-judge bench of the Supreme Court held that privacy is a fundamental right
- It held that the Constitution does not protect privacy
- It was delivered together with the DPDP Act in 2023
- It struck down the whole of the IT Act, 2000
Answer and explanation
Answer: A
In 2017 a nine-judge bench unanimously held privacy to be a fundamental right, including under Article 21. It affirmed, not denied, constitutional protection; it came years before the DPDP Act, 2023; and it did not strike down the IT Act.
Under the DPDP Act, 2023, a payroll company that processes employees' data only on behalf of an employer is a:
- Data principal
- Data processor
- Data fiduciary
- Consent manager
Answer and explanation
Answer: B
A data processor processes personal data on behalf of a data fiduciary. Here the employer decides why and how the data is used (fiduciary), the employees are the data principals, and the payroll company acts for the employer.
A coaching app collects students' phone numbers for login and later sells them to a loan company without asking. What is the main problem in terms of data protection principles?
- The app used an encrypted connection
- The students were data fiduciaries
- The data was used for a purpose the students never consented to
- Phone numbers are not personal data
Answer and explanation
Answer: C
Data collected for login was used for a different purpose without consent, breaking purpose limitation and consent. Phone numbers are personal data, the students are data principals rather than fiduciaries, and encryption is a protection, not a problem.
Which measure most directly addresses representation bias in a face recognition system?
- Collecting a more balanced, consented set of training photos and testing accuracy separately for each group
- Making the system run faster
- Reporting only the overall average accuracy
- Removing the option for people to appeal a wrong match
Answer and explanation
Answer: A
Representation bias comes from groups being missing or too few in the data, so the fix is more balanced data plus group-wise testing. Speed does not help, an overall average hides the problem, and removing appeals makes the harm worse.
A tool sends more inspections to areas where past records show more violations. More inspections there find more violations, which the tool reads as proof that it should send even more. This is best described as:
- Data minimisation
- Purpose limitation
- Human-in-the-loop control
- A feedback loop that can amplify existing bias
Answer and explanation
Answer: D
The tool's own outputs shape the next round of data, so an initial imbalance grows over time. That is a feedback loop. Data minimisation and purpose limitation are privacy principles, and human-in-the-loop is a form of oversight.
A loan tool approves 80 of 400 applicants from Group X and 30 of 300 applicants from Group Y. Group Y's approval rate is what fraction of Group X's?
- 0.25
- 0.5
- 2
- 0.75
Answer and explanation
Answer: B
Group X: 80 ÷ 400 = 20%. Group Y: 30 ÷ 300 = 10%. Ratio = 10 ÷ 20 = 0.5, so Group Y is approved at half the rate of Group X. This gap is a warning sign that the tool needs investigation.
A company says its AI system is "95% accurate", so it must be responsible. Why is this not enough?
- Because 95% accuracy is illegal everywhere
- Because accuracy can never be measured
- Because AI may be used only when it is exactly 100% accurate
- Because the errors may fall mostly on one group, and accuracy says nothing about accountability or explanation
Answer and explanation
Answer: D
An average figure can hide much higher error rates for some groups, and it says nothing about who is responsible or whether decisions can be explained. Accuracy is measurable, 95% is not illegal, and no system is perfect, so the others are wrong.
Consider two statements. (1) Accountability means a named person or organisation answers for what an AI system does. (2) Transparency requires every AI system to publish its full source code free of charge. Which is correct?
- Only statement 1
- Only statement 2
- Both statements
- Neither statement
Answer and explanation
Answer: A
Statement 1 correctly defines accountability. Statement 2 is wrong: transparency means being open that AI is used, broadly how and with what data; it does not require publishing all source code free. So only statement 1 is correct.
A hospital uses an AI tool to rank patients for treatment. Nobody can explain its scores, and doctors are told to follow them without question. Which two principles are most at risk?
- Data compression and storage speed
- Personalisation and entertainment value
- Explainability and human oversight
- Encryption and data minimisation
Answer and explanation
Answer: C
No one can give reasons for the scores, so explainability fails; doctors cannot question them, so human oversight fails. For decisions about treatment, both are essential. The other pairs are unrelated to the problem described.
Which of these is an example of deliberate AI misuse rather than an accidental error?
- A weather app that misses a light shower
- Using a voice clone to pretend to be a bank officer and ask for an OTP
- A translation tool that chooses a slightly awkward word
- A map app that suggests a longer route by mistake
Answer and explanation
Answer: B
Misuse means someone intends harm; cloning a voice to cheat people is deliberate fraud. The other cases are honest errors by systems trying to do their job - accidents, not misuse.
A government officer approves every AI recommendation to reject applications without reading them. A wrong rejection harms a family. Who is responsible?
- The AI software alone
- Nobody, because a machine made the decision
- The applicant, for applying
- The officer and the office that used and approved the AI's output
Answer and explanation
Answer: D
AI has no legal personality and cannot be held to account. The people and organisation who chose to use the tool and approved its output remain responsible. Rubber-stamping recommendations is a failure of human oversight, not a transfer of blame.
AI tools take over much of the routine data entry in an office. Which response is most in line with responsible AI?
- Hide the change from employees
- Ban computers in the office
- Help employees reskill and upskill for new or changed tasks
- Let the AI decide who should be dismissed with no review
Answer and explanation
Answer: C
Automation affects routine tasks first, and the ethical response is to help people adapt through reskilling and upskilling. Hiding the change breaks transparency, banning computers ignores reality, and letting AI dismiss people without review removes human oversight.
Which statement best describes the European Union's AI Act?
- It sorts AI uses by level of risk, with stricter duties for higher-risk uses and some uses prohibited
- It bans all use of AI in Europe
- It applies only to video games
- It requires every AI system to be open to the public free of charge
Answer and explanation
Answer: A
The EU AI Act takes a risk-based approach: uses with unacceptable risk are prohibited, high-risk uses face strict obligations, and lower-risk uses have lighter duties. It does not ban all AI, is not limited to games and does not make all AI free.
What does 'human-in-the-loop' mean?
- The AI acts entirely on its own with no person involved
- A person reviews and approves the AI's suggestion before a decision takes effect
- A human builds the computer hardware
- The AI is switched off at night
Answer and explanation
Answer: B
In a human-in-the-loop system the AI only suggests, and a person must review and approve before anything happens. A system acting with no person involved is human-out-of-the-loop. Building hardware and switching off at night are unrelated.