AI Ethics

Correct answer: C. जब प्रशिक्षण डेटा एकतरफा या असंतुलित हो

Explanation: AI systems learn from patterns in their training data. If that data overrepresents some groups or situations and underrepresents others, the model may produce systematically unfair results for the less-represented cases. Diverse data and regular evaluation can reduce this risk, while a clearly defined objective helps the project but cannot alone remove bias.

Correct answer: B. विविध और संतुलित डेटा का उपयोग करना

Explanation: Diverse and balanced data gives the system examples from different relevant groups and situations, making underrepresentation less likely. It does not guarantee perfect fairness by itself, so outputs should also be tested and monitored. Using only one group’s data, hiding errors, or stopping evaluation can conceal or increase unfair performance.

Correct answer: A. व्यक्तिगत जानकारी की सुरक्षा और उचित उपयोग

Explanation: Privacy concerns how personal information is collected, used, stored, shared, and protected, and whether people have appropriate control or permission regarding it. Preventing unauthorised access is an important security measure that supports privacy, but it is only one part of the broader idea. Processing speed and training time are performance issues, not privacy itself.

Correct answer: A. एआई के परिणामों की जाँच करना और व्यक्तिगत जानकारी की गोपनीयता बनाए रखना

Explanation: Responsible AI use requires checking outputs for errors, limitations, or bias and protecting personal information throughout the process. Human judgement remains important, especially for significant decisions. Sharing an unchecked result, publishing private data, or removing all oversight can cause harm and make accountability difficult, so those actions do not represent responsible use.

Correct answer: A. एक भर्ती एआई प्रणाली समान योग्यता वाले उम्मीदवारों में एक विशेष समूह के उम्मीदवारों को लगातार कम अंक देती है।

Explanation: If equally qualified candidates receive systematically lower scores because they belong to a particular group, the system is producing an unfair group-based disparity. That is a clear example of bias. Requesting documents for incomplete applications, sorting valid dates, or asking for a password change can be neutral procedures when applied consistently and for a relevant purpose.

Correct answer: A. A staff member reviewing an AI-recommended loan application before final approval

Explanation: A loan decision can significantly affect a person, so an employee should review the AI recommendation, check the data, and look for unfair treatment or special circumstances before approving it. Restarting a computer, typing, and alphabetising files are routine actions and do not represent oversight of a high-impact AI decision.

Correct answer: A. Transparent processes and verifiable outputs

Explanation: People are more likely to trust an AI system when they can understand its purpose, know how it is used, and check whether its outputs are reasonable. Transparency and verification also support accountability when errors occur. Hiding mistakes, concealing the purpose, or sharing private information weakens safety and trust.

Correct answer: A. Respect and privacy

Explanation: Permission expresses informed consent and respects a person’s control over personal data, images, or voice recordings. It helps protect privacy and dignity, especially when the information may identify or affect someone. Faster processing, model accuracy, and storage capacity are technical matters and do not replace the need for consent.

Correct answer: A. The AI system may reproduce past bias in its decisions.

Explanation: Historical selection data may reflect unequal opportunities or earlier unfair decisions. If the AI learns only from those patterns, it may reproduce or even strengthen the same bias when ranking new applicants. Training data alone does not guarantee fairness, so data review, fairness checks, and human oversight remain necessary.

Correct answer: A. Using diverse and balanced data

Explanation: Diverse and balanced data gives the system examples from different groups, conditions, and relevant situations. This can reduce the chance that the model will learn a narrow or unfair pattern. It does not guarantee fairness by itself; testing, monitoring, appropriate design, and human review are also important for finding remaining problems.

Correct answer: B. Protecting personal information from unauthorised access and misuse

Explanation: Privacy means handling personal information in a way that protects people from unauthorised access, disclosure, or misuse. Responsible practice includes collecting only what is needed, seeking appropriate consent, securing the data, and respecting its purpose and retention limits. Collecting without consent or sharing publicly can violate privacy.

Correct answer: B. Verify the date using a reliable textbook, museum, or official archival source

Explanation: A confident chatbot answer can still contain a factual error, so confidence is not evidence of correctness. The student should compare the date with an independent, reliable source such as a textbook, museum, or official archive. Asking the same chatbot again is not independent verification because it may repeat the same error.

Correct answer: A. Safety, fairness, and privacy

Explanation: Responsible AI considers how a system affects people throughout its design and use. It includes safety, fairness, privacy, transparency, accountability, appropriate human oversight, and checking for errors or harmful effects. Spreading misinformation, stealing personal data, and removing evaluation are irresponsible practices that can undermine trust and cause harm.

Correct answer: A. When an AI decision can affect people’s opportunities, safety, or access to services

Explanation: An AI decision affecting opportunities, safety, or access to services is high-impact. An incorrect or biased result in such a case can cause serious real-world harm, so human review is especially important. Low-risk suggestions such as colours, titles, or note organisation may still be checked, but their likely consequences are usually smaller.

Correct answer: B. Understanding, checking, and using AI output responsibly

Explanation: A responsible user treats AI output as assistance rather than unquestionable truth. The user should understand the purpose, check important facts and possible bias, protect privacy, and use the result ethically. Accepting every output, spreading misinformation, or sharing private information without consent can cause harm and violates responsible digital behaviour.

Correct answer: B. To identify and reduce potential harms caused by AI

Explanation: Risk awareness helps people identify problems such as biased decisions, privacy violations, inaccurate predictions, security failures, or unsafe recommendations. Testing, monitoring, human review, and safeguards can then reduce the chance or impact of harm. Attractive appearance is unrelated, while rushed deployment and blind acceptance increase risk rather than controlling it.

Correct answer: C. An AI output may contain errors or bias.

Explanation: AI output may contain factual errors, missing information, unsuitable recommendations, or bias inherited from data and system design. Checking important results helps users compare them with evidence, notice unfair effects, and correct mistakes before acting. Verification does not make AI less useful; it makes its use safer and more reliable. AI can produce many kinds of outputs, not only numbers.

Correct answer: A. A clear and safe purpose, reliable data, and human oversight

Explanation: Responsible AI begins with a clear, beneficial, and safe purpose. It also requires suitable and reliable data, attention to privacy and fairness, testing for errors, and human oversight of important outputs. Accepting results blindly, using inaccurate or biased data, or ignoring privacy can harm people and therefore contradict responsible use.

Correct answer: D. Digital ethics and academic integrity

Explanation: The governing concept is digital ethics, especially honesty and academic integrity when using technology. Looking at another candidate’s screen gives an unfair source of answers and violates the examination rules, regardless of whether the device or network works properly. File storage, network speed, and screen brightness describe technical conditions, not the student’s dishonest conduct. Thus option D is the only relevant answer.