Ai Ethics for Everyday People

• A. It thinks and understands the world the same way humans do
• B. It finds patterns in data and uses them to make predictions or generate outputs ✓
• C. It has its own goals and desires independent of its programming
• D. It stores and retrieves exact copies of information from the internet in real time

Why: Most AI systems, including modern machine learning models, work by learning statistical patterns from large amounts of data and using those patterns to make predictions or produce outputs. They do not necessarily understand the world in the human sense, and they do not have independent desires.

• A. AI is always objective because it is based on mathematics
• B. AI systems can reflect and amplify biases present in their training data ✓
• C. AI systems cannot make mistakes once they have been trained
• D. AI systems automatically know when their answers are wrong

Why: Because AI learns from data created by humans and society, it can absorb and even magnify existing biases. Mathematical foundations do not guarantee fairness, and trained systems can still produce errors.

• A. Narrow AI is designed for a specific task, while general AI would be able to handle a wide range of tasks across domains ✓
• B. Narrow AI runs on small computers, while general AI requires supercomputers
• C. Narrow AI is free to use, while general AI must be purchased
• D. Narrow AI cannot access the internet, while general AI can

Why: Narrow AI is built to perform specific tasks, such as recognizing faces or recommending videos. General AI refers to a hypothetical system with flexible, human-like intelligence across many domains. Today's AI tools are narrow, though discussions about future general capabilities are ongoing.

• A. AI systems require users to register before each conversation
• B. Knowing helps you judge how much to trust the output and understand how your data may be used ✓
• C. AI systems are legally required to introduce themselves with a name
• D. It determines which language the AI will respond in

Why: Recognizing that you are dealing with an AI helps you evaluate its answers critically, understand the possibility of errors or bias, and think about privacy and how your personal information might be handled.

• A. Summarize a long article quickly
• B. Recognize spoken language in a noisy room
• C. Verify that every fact they state is true without any external checking ✓
• D. Translate a sentence between two common languages

Why: AI language models can produce fluent but incorrect statements, sometimes called hallucinations. They do not inherently check facts against reality, so important claims should be verified with reliable sources. The other tasks are areas where AI performs well in many cases.

• A. AI systems that always refuse to answer questions
• B. Systematic errors in an AI system that produce unfair outcomes for certain groups of people
• C. A programmer's personal opinion being written into the code on purpose
• D. The tendency of computers to run more slowly over time

• A. It will recognize all faces equally well because it is a computer
• B. It will perform less accurately on darker-skinned faces than on lighter-skinned faces
• C. It will refuse to process any images at all
• D. It will only work in bright lighting conditions

• A. Random hardware failures inside the computer
• B. Only from deliberate instructions given by malicious programmers
• C. From the training data, which can reflect historical inequalities or underrepresent certain groups
• D. From the electricity used to power the servers

• A. Because AI bias only affects large companies, not individuals
• B. Because AI tools can influence decisions like hiring, lending, and healthcare, and users should question outputs rather than accept them automatically
• C. Because understanding bias allows users to fix the AI's code themselves
• D. Because AI systems are always biased in ways that cannot be reduced

• A. Training it only on the largest available dataset, regardless of who is represented in it
• B. Testing its performance across different demographic groups and using diverse, representative data
• C. Avoiding any evaluation so the system is not influenced by results
• D. Assuming that a system with high overall accuracy must be fair to everyone

• A. The average temperature of a city over a decade
• B. A user's browsing history and search queries ✓
• C. The publicly listed boiling point of water
• D. A general weather forecast for the whole country

Why: Browsing history and search queries are tied to an individual, so they count as personal data. The other options are general, public, or not linked to any person.

• A. The app has already collected your data and is asking for a thank-you
• B. You are told clearly what data will be collected, why, and how it will be used, and you can agree or decline ✓
• C. The app automatically assumes you agree if you keep using it
• D. The company must notify the government before collecting data

Why: Informed consent means you understand what is being collected and for what purpose before you agree. Simply continuing to use a service or collecting data first and asking later does not meet that standard.

• A. AI systems can combine many small pieces of data to infer sensitive details, such as health conditions or habits ✓
• B. AI systems can only process data that is already labeled as sensitive
• C. Harmless data is always deleted automatically by AI systems
• D. Combining data is illegal, so AI systems never do it

Why: AI is good at finding patterns across large datasets, so separate harmless details can be pieced together to reveal private information. This is why aggregated data can still be sensitive.

• A. Sharing your full name, address, and bank details so the tool can personalize its help
• B. Reviewing privacy settings and avoiding sharing sensitive personal information unless necessary ✓
• C. Assuming that anything you type is automatically private and permanently deleted
• D. Using the same password across all AI tools for convenience

Why: Checking privacy settings and limiting what you share reduces your exposure. Assuming automatic privacy or reusing passwords leaves you vulnerable.

• A. A right to have your personal data deleted or removed from certain systems under specific conditions ✓
• B. A right to forget your own passwords without penalty
• C. A requirement that AI systems forget everything they learned from training
• D. A rule that companies must erase all public records about everyone

Why: The right to be forgotten lets individuals request deletion or removal of their personal data in certain circumstances, such as when it is no longer needed. It applies to personal data, not to general knowledge or every public record.

• A. The model deliberately lies to the user to achieve a hidden goal
• B. The model generates confident-sounding information that is false or unsupported by its training data or sources ✓
• C. The model refuses to answer questions because it detects harmful intent
• D. The model produces images that look like photographs of real people

Why: A hallucination refers to output that sounds plausible and confident but is factually wrong or fabricated. Language models predict likely text rather than checking facts, so they can invent names, dates, or citations without any intent to deceive.

• A. AI systems are always wrong, so their outputs are never useful
• B. AI outputs can contain errors or fabrications, and sharing them can spread misinformation even unintentionally ✓
• C. Only government agencies are permitted to use AI-generated information
• D. Verification is only necessary for images, not for text

Why: AI can be useful but is not infallible. Because its errors often look polished and credible, people who share unverified AI output can unknowingly spread false information to others.

• A. A deep analysis report generated by a data science team
• B. A synthetic image, video, or audio recording that realistically depicts someone saying or doing something they never did ✓
• C. A type of encryption used to protect private messages
• D. An AI model trained on very large datasets

Why: Deepfakes use AI to fabricate realistic-looking media of real people. They raise serious trust concerns because seeing or hearing someone is traditionally strong evidence that it happened.

• A. Trusting any AI answer as long as it sounds confident and detailed
• B. Avoiding AI entirely because any use of it is dangerous
• C. Understanding the tool's limitations and cross-checking important information against reliable sources ✓
• D. Assuming that a company's AI is always accurate because the company is large

Why: Calibrated trust means relying on AI where it performs well and verifying where it may fail. Knowing limitations and checking important claims against independent, reliable sources helps people avoid both blind trust and blanket rejection.

• A. They are designed to show users only verified scientific research
• B. They are often optimized for engagement, which can prioritize sensational or emotionally charged content over accurate content ✓
• C. They cannot process any text that contains numbers
• D. They automatically delete all content that users disagree with

Why: When systems are tuned to maximize clicks, watch time, or shares, emotionally provocative content often performs better than accurate but dull content. This optimization can amplify misleading material without anyone explicitly intending it.

• A. Workers voluntarily choosing to use new software tools
• B. Machines or software performing tasks that were previously done by humans ✓
• C. Companies hiring more people to handle customer requests
• D. Employees taking breaks scheduled by a computer system

Why: Automation refers to machines or software carrying out tasks that people used to do. The other options describe voluntary tool use, hiring, or scheduling, which are not the core idea of automation.

• A. Because displaced workers are always less skilled than others
• B. Because the benefits of automation often go to business owners while costs fall on workers ✓
• C. Because AI systems are legally required to employ a minimum number of humans
• D. Because automation always reduces the total amount of goods produced

Why: Ethicists point out that productivity gains from automation can concentrate wealth with owners and investors, while displaced workers bear the hardship of lost income. This imbalance is a central reason for calls for retraining, safety nets, and fair distribution of gains.

• A. Deciding whether a person should receive a medical treatment without human review
• B. Sorting large volumes of repetitive data entry records ✓
• C. Counseling a grieving family about a loss
• D. Making final hiring decisions about job candidates' character

Why: Repetitive, well-defined data tasks are the most common and least risky candidates for automation. Decisions involving medical care, emotional support, or character judgments carry high stakes and benefit from human oversight.

• A. A program that replaces workers with machines permanently
• B. A program that teaches workers new skills so they can move into different kinds of jobs ✓
• C. A program that reduces working hours for all employees equally
• D. A program that bans AI from being used in any workplace

Why: Reskilling means training people in new abilities so they can shift to roles that remain in demand or emerge with new technology. It is one common response to job disruption caused by automation.

• A. Hide the changes from employees until the system is fully running
• B. Involve affected workers in planning, communicate openly, and consider their wellbeing ✓
• C. Adopt AI only in departments where workers cannot object
• D. Replace the workforce as quickly as possible to cut costs before anyone notices

Why: Transparent communication, worker involvement, and attention to wellbeing are widely recommended principles for responsible workplace AI. Secrecy, avoiding objections, or rapid unannounced replacement undermine trust and fairness.

• A. The AI always gives every person exactly the same outcome.
• B. The AI treats people in comparable situations consistently and does not disadvantage people based on irrelevant traits like race or gender.
• C. The AI makes decisions faster than a human could.
• D. The AI only uses data that the developers personally approve of.

• A. The algorithm will be too slow to process applications.
• B. The algorithm may learn to favor male candidates because it reflects patterns from the historical data.
• C. The algorithm will automatically reject all female applicants.
• D. The algorithm cannot be used for any hiring decisions at all.

• A. It makes the AI run more efficiently.
• B. It allows people to check for errors or unfair treatment and seek correction.
• C. It guarantees the AI will never make a mistake again.
• D. It is only important for companies, not for the public.

• A. Zip codes are always inaccurate, so the AI will fail.
• B. Removing a sensitive attribute does not guarantee fairness, because other features can act as proxies for it and produce discriminatory outcomes.
• C. Zip codes are illegal to store in any database.
• D. Loan decisions should never depend on any data at all.

• A. Only testing the system on the developers' own data.
• B. Measuring error rates and outcomes separately for different demographic groups and comparing them.
• C. Assuming fairness is achieved if the system has high overall accuracy.
• D. Avoiding any measurement, since measuring groups is itself unfair.

• A. The AI itself, since it made the decision
• B. The bank and the people who chose to deploy and rely on the system ✓
• C. No one, because AI decisions are automatically objective
• D. The customer, for providing incomplete information

Why: AI systems cannot be held morally or legally responsible in the way people can. Organizations that choose to deploy an AI system, and who oversee its use, are generally held accountable for its outcomes.

• A. AI systems are always too small to cause real harm
• B. Many people are involved, including developers, data providers, deployers, and users, so responsibility can be spread out and unclear
• C. Laws against AI harm have never been written
• D. AI only makes mistakes that no human could have predicted

• A. The color of the car
• B. Whether the system was tested properly, used as intended, and whether the company warned about its limits ✓
• C. How much the car cost to buy
• D. Whether the car's name was popular with the public

Why: Responsibility usually depends on whether the developer and operator tested the system adequately, followed safety requirements, and used it within its intended conditions, along with whether the user acted reasonably.

• A. Having a person check or be able to override AI decisions, especially high-stakes ones
• B. Letting AI run without any human involvement to avoid bias
• C. Requiring a human to write every line of AI code by hand
• D. Allowing AI to supervise other humans only

• A. AI companies never pay any fines for their products
• B. When AI acts in ways no one specifically intended or predicted, it can be unclear who should be held accountable ✓
• C. Users are always blamed for AI errors
• D. AI systems are legally required to apologize when they fail

Why: The responsibility gap arises when an AI produces harmful outcomes that no single person directly intended or foresaw, making it hard to trace accountability to any one party.