I.32 Ethical Issues — Computer Science | CSS Notes

⬆️ Back to: P-I.I. Introduction to Computing

🧭

Page Outline — click a heading to jump.

🎯

Exam pattern (2016–2026): Ethics half of 2021 Q2(b) (full answer on I.30) · 2026 MCQ Q2 (plagiarism as the ethical concern, on I.34).

CSS Computer Science (Optional) — Paper-I, Section-A, Topic I, p.73. Covers: Ethical Issues.

Definition

Ethical Issues (in computing): questions of right and wrong conduct raised by computing technology, which may diverge from what is merely legal or illegal.

Beyond Legality — What Is Ethical vs What Is Illegal

Something can be legal but still ethically questionable, or illegal yet ethically defensible — legality and ethics don't always align in computing.

Key Ethical Issues in Computing

Digital divide: the gap between people/regions with reliable access to computers and the Internet and those without, whether due to cost, infrastructure, literacy, or geography — raising questions of fairness and inclusion.
- How it manifests: rural areas lacking broadband infrastructure, students unable to afford a computer or data package, or a country's cities having high-speed internet while remote regions have none.
- Why it's ethical, not just economic: as essential services (education, banking, government services, job applications) move online, the divide directly translates into unequal access to opportunity and rights, not merely inconvenience.
- Example: students in remote areas of Pakistan struggling to attend online classes during COVID-19 school closures due to lack of internet access, while urban students continued largely uninterrupted.
Surveillance: the tension between using monitoring technologies for legitimate security/management purposes and an individual's right to privacy and freedom from being watched.
- How it manifests: government interception of communications for national security, employers monitoring employee emails/keystrokes, CCTV and facial recognition in public spaces.
- Key ethical question: whether the monitoring is proportionate, transparent, and subject to oversight, or excessive, secret, and open to abuse (e.g., targeting dissidents or journalists rather than genuine threats).
- Example: a government justifying bulk communications interception as a counter-terrorism measure, while critics argue it lacks judicial oversight and chills free expression.
Automation and job displacement: the ethical questions raised when automating tasks or roles previously done by people results in job losses or deskilling.
- Key considerations: whether employers have a responsibility to retrain or transition displaced workers, whether the economic gains of automation are shared fairly, and how quickly a workforce can be expected to adapt.
- Example: a bank replacing customer-service staff with chatbots, or a factory replacing assembly-line workers with robots, raising questions about retraining programs and severance obligations.
Data ownership and consent: the question of who truly "owns" personal data generated by a user's activity, and whether their agreement to share it was informed and freely given (e.g., via long, unread terms-of-service agreements).
- How it manifests: apps that collect far more data than needed for their stated function, consent obtained via lengthy legalese that few users read, or data resold to third parties beyond the user's expectation.
- Example: a free mobile app that harvests contact lists and location history and sells them to advertisers, with users unaware because consent was buried in a terms-of-service document.
AI bias and accountability: the risk that an algorithm's decisions systematically disadvantage certain groups (by gender, ethnicity, age, etc.), typically due to biased or unrepresentative training data — with accountability unresolved between the developer, the deploying organization, and the data provider.
- How it manifests: a hiring algorithm trained on historical data that reflects past discriminatory hiring patterns and reproduces them; a facial-recognition system that performs less accurately on certain skin tones.
- Example: an automated recruitment tool that downgrades resumes from a particular gender because it was trained on a company's historically skewed hiring data.
Professional ethics: the standard of honest, competent, and socially responsible conduct expected of computing professionals, going beyond what the law strictly requires — formalized in codes of conduct like the IEEE (Institute of Electrical and Electronics Engineers) / ACM (Association for Computing Machinery) Code of Ethics (see dedicated page).
- Key features: includes honesty about a system's limitations, refusing to build systems intended to cause harm, protecting user data even when not legally mandated, and disclosing conflicts of interest.
- Example: a software engineer refusing to ship a product they know contains an unresolved security flaw simply because a deadline is approaching, and instead flagging the risk to management.

MCQ Traps

Legal ≠ ethical: something can be technically legal (e.g., a vague data-collection practice buried in terms of service) while still being ethically questionable — a distinction examiners like to probe.
AI bias is an ethical issue about fairness in outcomes, not simply a technical bug — it concerns whether a system's decisions systematically disadvantage certain groups.
Surveillance and data ownership/consent are related but distinct issues — surveillance is about monitoring by an external party (often the state or employer); data ownership/consent is about whether the individual whose data is collected agreed to how it is used.

✅ Self-Test Questions (with Answers)

Q1. Give an example of something that could be legal but still ethically questionable in computing.

Ans: The clearest example is consent buried in terms-of-service agreements. A free mobile application presents a lengthy legalese document that almost no user reads, obtains a click on "I agree", and on that basis harvests the user's contact list, location history and device identifiers and sells them to advertising brokers. This is very likely legal — consent was formally obtained and disclosure was technically made — yet it is ethically questionable, because the consent was neither informed nor meaningfully voluntary: the user did not understand what was being collected, the disclosure was deliberately unreadable, and refusing meant losing access to the service entirely. Ethics asks whether the user genuinely agreed; law often asks only whether a box was ticked.

Other standard examples of the legal-but-unethical gap in computing:

Dark patterns: interfaces designed to nudge users into subscriptions or data sharing — pre-ticked boxes, a bright "Accept all" beside a greyed-out "Manage settings", or deliberately obstructed cancellation.
Planned obsolescence: ending security updates for working hardware to drive replacement sales.
Data collected for one purpose reused for another (function creep), permitted by broad drafting but outside what users expected.
Selling a lawful surveillance product to a client likely to misuse it, where no export rule is breached.
Employee monitoring disclosed in a contract but extending to keystrokes and private messages.

The underlying principle, and the point examiners test, is that legality and ethics are distinct standards. Law sets a minimum floor of enforceable conduct and lags behind technology; ethics asks what a responsible professional ought to do. Conduct can be legal yet unethical (the examples above), and conceivably ethical yet illegal (whistleblowing that discloses genuine wrongdoing in breach of a confidentiality clause). Codes such as the ACM/IEEE Code of Ethics exist precisely to fill the space the law leaves open.

Q2. What is the "digital divide," and why is it considered an ethical issue?

Ans: The digital divide is the gap between those who have reliable, affordable access to computers, the internet and the skills to use them, and those who do not. It is not a single gap but several overlapping ones: a geographic divide between urban and rural areas, an economic divide based on the cost of devices and data, a gender divide in device ownership and permission to use technology, a generational divide, and a skills or literacy divide in which a person has a connection but cannot use it meaningfully. Scholars often distinguish a first-level divide (access to the infrastructure) from a second-level divide (the capability to use it productively rather than only for consumption).

It is an ethical issue, not merely an economic or technical one, because essential services have migrated online. Education, banking, job applications, welfare registration, tax filing and government services increasingly presuppose connectivity, so lacking access is no longer an inconvenience but a barrier to opportunity, entitlement and civic participation. This engages the core ethical principles of justice, fairness and equality of opportunity: a person excluded from the digital sphere is excluded from rights that the state assumes everyone can exercise. The divide also tends to be self-reinforcing, since those already disadvantaged fall further behind as services digitise, widening existing inequality rather than correcting it.

Education: online-only learning excludes the unconnected.
Economic participation: digital-only job portals and freelancing markets are closed to them.
Governance: e-government services disenfranchise those who cannot reach them.
Health and information: unequal access to reliable information in a crisis.

CSS angle: Pakistan offers a direct illustration. During the COVID-19 school closures of 2020, students in remote districts could not attend online classes for want of connectivity or devices while urban students continued largely uninterrupted. The Universal Service Fund (USF) exists precisely to extend telecom infrastructure to unserved and underserved areas, and closing the divide is a recurring commitment in national Digital Pakistan policy — a useful concrete reference in an answer.

Q3. Explain what is meant by "AI bias and accountability."

Ans: AI bias refers to the tendency of an algorithmic system to produce outcomes that systematically disadvantage particular groups — by gender, ethnicity, religion, age, disability or region — rather than erring randomly. The usual cause is not malice in the code but the training data: a model learns the statistical patterns of the historical data it is given, so if that data encodes past discrimination, the model reproduces and then automates it at scale. Bias can also enter through unrepresentative sampling, through proxy variables that stand in for a protected attribute (a postcode standing in for ethnicity), or through an objective function that optimises for the wrong target. The critical point for an ethics answer is that this is a question of fairness in outcomes, not a technical bug: the system may be functioning exactly as designed and still be unjust.

Accountability is the accompanying question of who answers for the harm. It is unresolved because responsibility is distributed across the developer who built and trained the model, the organisation that chose to deploy it, and the source of the data. This difficulty is sharpened by two features of modern systems:

Opacity (the "black box" problem): complex models cannot readily explain why a particular decision was reached, so an affected person cannot contest it — hence the push for explainable AI and a right to explanation.
The "many hands" problem: with many contributors, each can plausibly disclaim responsibility, leaving the victim without redress.

Real examples: Amazon abandoned an experimental recruitment tool in 2018 after finding it downgraded CVs containing the word "women's"; the Gender Shades study (Buolamwini and Gebru, 2018) showed commercial facial-analysis systems had far higher error rates for darker-skinned women than for lighter-skinned men; and the COMPAS recidivism-scoring tool used in US courts was found to misclassify defendants unevenly by race.

The practical remedies are representative data, bias auditing of outcomes across groups, documented model limitations, and keeping a human in the loop for consequential decisions such as hiring, lending, bail or admissions.

Q4. Write a 5-mark answer explaining the ethical tension between government surveillance and individual privacy, with an example.

Ans: The tension is a conflict between two goods that a state is obliged to protect: collective security and individual privacy. A government argues that intercepting communications, retaining metadata and deploying CCTV or facial recognition is necessary to detect terrorism, organised crime and child exploitation before harm occurs. Citizens and civil-liberties advocates reply that privacy is a precondition of other freedoms — that people who believe they are watched self-censor, associate less freely and criticise power less readily, so surveillance corrodes free expression and political participation even when no one is prosecuted. This is the chilling effect. The ethical difficulty is that neither claim is frivolous, so the question is not whether surveillance is permissible but under what constraints.

Ethical evaluation usually turns on four tests:

Legality: is the power grounded in clear, published law rather than administrative discretion?
Necessity and proportionality: is the intrusion the least intrusive means available, and is its scale matched to the threat? Targeted surveillance of a suspect is far easier to justify than bulk collection on an entire population.
Oversight: is authorisation independent — judicial warrant rather than self-authorisation by the agency conducting it?
Transparency and redress: can the public know the rules, and can a wronged individual obtain a remedy?

Example: a state justifies bulk interception of internet traffic as counter-terrorism, but the capability is later used to monitor journalists and opposition activists — the classic function creep in which a power granted for an exceptional purpose migrates to routine political use. The Edward Snowden disclosures of 2013 made this pattern globally salient.

CSS angle: In Pakistan, Article 14 of the Constitution guarantees the privacy of the home and the dignity of man, while PECA 2016 grants investigative powers over data and traffic records; the resulting debate over lawful-interception powers and content blocking by the PTA is a live domestic illustration of precisely this tension.

Q5. Who should bear responsibility when an automated hiring system is found to discriminate — the developer, the deploying company, or both? Justify your answer.

Ans: The defensible position is that responsibility is shared but differentiated, because the harm arises from decisions taken at both ends of the chain, and neither party can fully discharge its duty by pointing at the other. This is the problem of distributed responsibility or the "many hands" problem in computing ethics: when an outcome results from the combined choices of many actors, each is tempted to disclaim accountability, and the victim is left with no remedy. The applicant, meanwhile, suffers a concrete harm regardless of who caused it.

The developer is responsible for the design choices that made bias possible: selecting and curating training data, failing to test outcomes across protected groups, and shipping a system whose reasoning cannot be explained. The deploying organisation is responsible for procurement diligence, for choosing to automate a consequential decision at all, for the weight it gives the system's output, and for the absence of human review and an appeals route. A reasonable allocation is therefore:

Developer: duty of care in data selection, bias testing, documentation of limitations, and honest disclosure of what the system cannot do.
Deploying employer: duty to audit before and during use, to keep a human in the loop for adverse decisions, and to provide an appeal mechanism — the employer, not the vendor, is the party that actually rejects the candidate.
Regulator: duty to require transparency and to provide a remedy, since neither commercial party has a strong incentive to self-report.

The strongest answer notes that legal liability usually attaches to the employer, since it is the employer who makes the employment decision and is bound by anti-discrimination obligations, while ethical responsibility extends further back to the developer. "The algorithm decided" is not an ethical defence: automation transfers the labour of a decision, not the accountability for it.