Data Ethics: the Big Questions

• A. The technical process of encrypting data for secure storage
• B. A field concerned with the moral obligations related to collecting, using, and sharing data ✓
• C. The legal rules that govern which companies can own databases
• D. The statistical methods used to clean and analyze large datasets

Why: Data ethics examines the moral questions raised by how data is collected, used, and shared, including issues of fairness, privacy, and harm. It goes beyond technical security or legal compliance alone.

• A. Collecting data only from people who are legally required to provide it
• B. Ensuring people understand what data is collected, why, and how it will be used before agreeing ✓
• C. Obtaining consent from a company's board of directors before collecting data
• D. Notifying individuals only after their data has already been analyzed

Why: Informed consent means individuals are given clear, understandable information about data practices and can voluntarily agree or decline. Burying terms in lengthy legal text or notifying people after the fact does not meet this standard.

• A. A researcher publishes aggregated average income figures for a city
• B. A retailer uses purchase history to recommend products to a customer who opted in
• C. A company sells customers' location histories to third parties without their knowledge ✓
• D. A hospital uses anonymized patient records to track flu trends

Why: Selling detailed location histories without people's knowledge exposes sensitive personal information to parties they never agreed to share it with. The other options involve aggregation, consent, or anonymization that reduce privacy risks.

• A. The dataset contains only numerical values rather than text
• B. The dataset is systematically unrepresentative of the population it is meant to describe, leading to skewed conclusions ✓
• C. The dataset was collected by a government agency rather than a private company
• D. The dataset has been deleted or corrupted over time

Why: Bias in data means certain groups or perspectives are over- or under-represented, which can produce unfair or inaccurate outcomes when the data drives decisions. This is a central concern in data ethics because biased data can perpetuate discrimination.

• A. Anonymized data is always legally required to be stored in paper form
• B. Individuals can sometimes be re-identified by combining anonymized data with other available information ✓
• C. Anonymization permanently deletes all useful information from a dataset
• D. Anonymized data cannot be used for any statistical analysis

Why: Re-identification attacks show that seemingly anonymous records can be linked to specific people when combined with outside data such as zip codes, birth dates, or public records. This is why anonymization must be paired with other safeguards and careful ethical judgment.

• A. A user agreeing to terms of service that they never read
• B. A person freely agreeing to data collection after understanding what data will be collected, how it will be used, and who will have access to it ✓
• C. A company assuming users are willing to share data unless they opt out
• D. Data being collected automatically once a device is switched on

Why: Informed consent requires that the person understands the relevant details of the data practice and agrees voluntarily. Buried or unread terms of service and default-on collection do not meet this standard.

• A. It makes databases run more slowly
• B. It increases the amount of personal information exposed to risk of misuse or breach without a clear purpose ✓
• C. It reduces the accuracy of statistical analysis
• D. It violates copyright law

Why: Data minimization means collecting only the data necessary for a stated purpose. Collecting excess personal data creates more opportunities for misuse, leaks, and unauthorized secondary uses.

• A. A company stores encrypted backups securely
• B. A company uses customer purchase histories to infer and act on sensitive traits, such as pregnancy, without the customer knowing ✓
• C. A researcher anonymizes a dataset before sharing it
• D. A government publishes aggregate census statistics

Why: Harm can arise from inference and profiling even without a breach. Using data to reveal and act on sensitive characteristics without awareness undermines individual autonomy and privacy.

• A. People never want to share any information at all
• B. Consent is often given under pressure, with complex terms, or without realistic alternatives, so it may not be truly free or informed ✓
• C. Consent is illegal in most countries
• D. Consent can only be given in writing by a lawyer

Why: When services are essential and terms are lengthy or opaque, 'agreement' may be coerced or uninformed. This is why many ethicists argue for additional safeguards beyond consent alone.

• A. Information should always be made as public as possible
• B. Information flows should match the norms of the context in which they were originally shared ✓
• C. Information should never be shared between any two parties
• D. Information is only private if it is stored on paper

Why: Contextual integrity holds that privacy is preserved when information moves in ways consistent with the expectations and norms of the setting where it was shared, such as medical details flowing between patient and doctor but not to an employer.

• A. Algorithmic bias ✓
• B. Data compression
• C. Overfitting
• D. Latency

Why: Algorithmic bias refers to systematic errors that produce unfair outcomes, often disadvantaging particular groups. Data compression, overfitting, and latency are technical concepts that do not inherently describe unfairness toward groups.

• A. Random noise in the data
• B. Historical bias reflected in the training data ✓
• C. The algorithm's processing speed
• D. Too many features in the model

Why: Historical bias occurs when training data reflects past discriminatory decisions. A model trained on such data learns to replicate those patterns, even if gender is not explicitly used as an input.

• A. Calibration
• B. Demographic parity ✓
• C. Individual accuracy
• D. Model interpretability

Why: Demographic parity requires that the proportion of positive outcomes is equal across groups. Calibration concerns whether predicted probabilities match observed frequencies, and the other options are not formal fairness criteria.

• A. Darker skin reflects less light, and training datasets may underrepresent these faces ✓
• B. The algorithm cannot process color images
• C. Facial recognition only works on children
• D. Skin tone changes the speed of the camera

Why: Facial recognition systems trained on datasets with few examples of darker skin tones tend to perform less accurately on those faces. This is a case of representation bias in the training data.

• A. Remove all data about people from the dataset
• B. Audit the training data for representation gaps and test performance across demographic groups ✓
• C. Increase the model's complexity as much as possible
• D. Only evaluate the model on its overall average accuracy

Why: Auditing data for representation gaps and disaggregating performance metrics by group are foundational steps for identifying bias. Relying only on overall accuracy can hide poor performance for minority groups.

• A. Making the source code of every system publicly available
• B. Openly disclosing how data is collected, used, and how decisions are made ✓
• C. Encrypting all data so that only authorized staff can read it
• D. Deleting personal data after a fixed period of time

Why: Transparency concerns openly communicating what data is collected, how it is used, and how decisions affecting people are reached. It does not require publishing all source code, nor is it the same as encryption or data deletion.

• A. Scalability
• B. Redundancy
• C. Explainability ✓
• D. Latency

Why: Explainability refers to the ability to understand and articulate the reasons behind a model's predictions or decisions. Scalability, redundancy, and latency relate to system performance and infrastructure rather than interpretability.

• A. It allows the bank to charge a higher interest rate to the applicant
• B. It lets the applicant understand and potentially contest a decision that affects them ✓
• C. It guarantees the model will always approve future applications
• D. It removes the need for the bank to keep any records of the decision

Why: When automated decisions significantly affect people, explainability gives them the information needed to understand the outcome and challenge it if it was unfair or based on errors. The other options do not reflect the ethical purpose of explainability.

• A. A model that has been certified as free of all bias
• B. A model whose internal workings are opaque, so its reasoning is hard to inspect ✓
• C. A model stored offline to prevent hacking
• D. A model that only uses data from a single source

Why: A black box model produces outputs without an easily understood internal process, making it difficult for developers, auditors, or affected individuals to see why a decision was made. This opacity is a central concern in transparency debates.

• A. Keeping the system's purpose undisclosed to avoid misuse by competitors
• B. Documenting the data sources, intended use, and known limitations, and sharing this with stakeholders ✓
• C. Updating the model frequently without recording the changes
• D. Relying solely on the developers' personal judgment to verify fairness

Why: Documentation such as data provenance records, model cards, and disclosed limitations helps stakeholders understand and scrutinize a system. Hiding purpose, undocumented changes, and unverified personal judgment all undermine transparency.

• A. The legal right to create new datasets from public information
• B. The question of who has rights to control, access, and benefit from data about people and their activities ✓
• C. The technical ability to store large volumes of data on servers
• D. The requirement that all data be published openly for research purposes

Why: Data ownership concerns who holds rights to control, access, and profit from data, including data generated about individuals. It is a central ethical and legal question, not just a technical one.

• A. Because they are required by law to collect all user data
• B. Because they control vast amounts of user data and can use it to shape behavior, markets, and information flows ✓
• C. Because they always operate at a financial loss
• D. Because they are owned entirely by governments

Why: Platforms that aggregate large volumes of user data can influence what people see, how they behave, and which competitors can succeed, concentrating power in the hands of a few firms.

• A. The company that collects and stores personal data
• B. The government agency that regulates data use
• C. The individual whose personal data is being processed ✓
• D. The software engineer who designs the database

Why: A data subject is the identified or identifiable person to whom personal data relates. Data protection laws give data subjects rights such as access, correction, and erasure.

• A. Two friends sharing photos with each other on a private message
• B. A user who is unaware their browsing history is sold to advertisers, while the company profits and the user cannot easily negotiate terms ✓
• C. A researcher publishing anonymized results with consent from participants
• D. A library lending books to members who have signed up voluntarily

Why: The power asymmetry lies in the company's knowledge, resources, and control over data compared with the user's limited awareness and bargaining position, often reinforced by take-it-or-leave-it consent terms.

• A. Data should only ever belong to the first person who collects it
• B. Data about groups or communities can affect those groups even when no single individual is identified, so ownership may need to be considered collectively ✓
• C. Data has no value unless it is combined into a single global database
• D. Only governments can hold collective rights over any form of data

Why: Data about patterns in a group, such as neighborhoods or demographics, can lead to discrimination or harm against that group even when individuals are not named, raising questions about collective rather than purely individual ownership.

• A. Data is always stored in encrypted form
• B. People change their behavior when they believe they are being watched ✓
• C. Governments always misuse collected data
• D. Surveillance increases the cost of storage hardware

Why: The chilling effect describes how awareness of monitoring leads people to self-censor, avoiding speech or associations they fear could be judged or punished, even if no punishment occurs.

• A. Only criminals need privacy protections
• B. Individuals with no wrongdoing have no reason to fear surveillance ✓
• C. Surveillance data should be made public
• D. Governments must disclose all surveillance programs

Why: The 'nothing to hide' argument holds that if you are not doing anything wrong, being watched should not matter. Critics respond that privacy protects more than secrets, including autonomy, dignity, and protection against future misuse of data.

• A. To make investigations faster for police
• B. To require an independent judge to approve intrusion based on probable cause ✓
• C. To notify the public of every investigation
• D. To ensure surveillance equipment is standardized

Why: Warrant requirements place a neutral magistrate between the state and the citizen, requiring evidence of probable cause before searches or surveillance. This is a key civil liberties safeguard against arbitrary government intrusion.

• A. Targeted surveillance
• B. Bulk or mass surveillance ✓
• C. Forensic analysis
• D. Undercover policing

Why: Bulk or mass surveillance involves indiscriminate collection of communications or data across large populations, as opposed to targeted surveillance focused on specific individuals suspected of wrongdoing.

• A. Stored data always becomes corrupted over time
• B. It creates a permanent record exposed to future misuse, breaches, and changing government priorities ✓
• C. Storage costs make retention financially impossible
• D. Courts have ruled that all stored data must be deleted within a week

Why: Retained data creates a standing risk: it can be stolen in breaches, repurposed for new goals, or accessed by future governments with different intentions, so retention itself can threaten liberty even without current abuse.

• A. Transparency
• B. Accountability ✓
• C. Proportionality
• D. Interoperability

Why: Accountability means there is a clear responsibility for how data systems behave and for the consequences of their use. Transparency is related but concerns openness about how systems work, not who answers for their outcomes.

• A. To maximize the speed of data processing
• B. To identify and mitigate risks to individuals before a data processing activity begins ✓
• C. To sell anonymized datasets to third parties
• D. To replace the need for user consent

Why: A DPIA is a structured process for evaluating the risks a planned data processing activity poses to people's rights and freedoms, so that mitigations can be built in before harm occurs.

• A. A data ethics or data governance committee ✓
• B. The marketing department alone
• C. External social media users
• D. The hardware procurement team

Why: Data ethics or governance committees bring together legal, technical, and business stakeholders to set policy, review high-risk uses, and hold the organization to its commitments.

• A. It makes datasets load faster
• B. It records who accessed or changed data and when, enabling review and investigation of misuse ✓
• C. It automatically deletes outdated records
• D. It removes the need for documentation

Why: Audit trails create a verifiable record of data activity, which supports accountability by making it possible to detect, investigate, and attribute unauthorized or inappropriate actions.

• A. Can the system be made more profitable?
• B. Who is responsible for identifying, redressing, and preventing this harm, and what processes allow affected people to contest decisions? ✓
• C. Has the system been given a human-like name?
• D. Is the harm too small to matter?

Why: Good governance asks who is answerable, how harm is remedied, and whether affected people have a route to challenge decisions. Focusing on responsibility and redress is central to accountability.