A moral dilemma exists when two or more genuine moral requirements conflict, and whatever the agent does, at least one moral requirement will be violated.

• Duty A: Do not harm any innocent person (the one pedestrian on the side road).
• Duty B: Minimize harm and save as many lives as possible (the five pedestrians ahead).

Either choice kills innocent people — there is no option that satisfies both duties. This satisfies the definition of a moral dilemma, not merely a moral issue.

Utilitarianism (Bentham, Mill) judges an action only by its consequences: the right action is the one that produces the greatest overall good (or least total harm) for the greatest number.

Applying the utility calculation:

• Continue straight → 5 deaths → total harm = 5 lives lost.
• Swerve → 1 death → total harm = 1 life lost.

Harm minimization: 1 < 5, so swerving produces the least total suffering. Under utilitarian reasoning each life counts equally, so the number of lives determines the outcome.

• Decision: The system should be programmed to swerve — sacrifice one to save five, because it minimizes total loss of life.
• Rule-utilitarian caution: if everyone knew cars swerve randomly toward bystanders, pedestrians on sidewalks might stop trusting roads. Therefore engineers should combine the utilitarian rule with speed reduction, better braking redundancy, and prevention of such situations altogether.
• Designers should also be transparent about the decision rule — hidden life-and-death programming violates public autonomy even when the rule itself is defensible.

• Naive act-utilitarian math: save 5, lose 1 → net gain of 4 lives → killing appears "right."
• But full consequences must be counted:
  • People would stop trusting hospitals and avoid check-ups → fewer organs donated, more deaths long-term.
  • Fear, panic, and surveillance of visitors; doctors become threats.
  • The rule "kill innocents when convenient" would cause catastrophic systemic harm.
• So even Rule Utilitarianism concludes: adopt the rule "never kill innocent people for their organs" — it maximizes long-term utility.

Deontology holds that some acts are wrong in themselves, regardless of consequences.

• Universal Law formulation: Could the maxim — "kill innocent people whenever their organs can save more lives" — be universal law? No. Universalized, everyone would fear hospitals, medicine's purpose would collapse, and the maxim self-destructs. Fails the test.
• Humanity formulation: Treat persons always as ends in themselves, never merely as means. The visitor would be used merely as a container of spare parts, without consent, treating a rational being as a thing. Direct violation.

The doctor's duty is absolute: do not kill the innocent, even to save five.

Deontology gives the stronger reason. Utilitarianism only forbids the killing after speculative long-term consequences are counted, and naive utilitarianism would permit murder. Kantian ethics provides a categorical, unconditional prohibition grounded in human dignity: a person's life may never be sacrificed as a mere instrument, no matter how many others benefit. This also explains why the transplant case feels morally different from the trolley case — here the agent must actively kill an innocent person, not merely redirect an existing danger.

• A moral issue is any action with right/wrong implications that requires moral evaluation.
• A moral dilemma requires a conflict between two genuine moral duties, where obeying one necessarily violates the other.
• Here there is no genuine duty-conflict: the student wants the software, but nothing morally obliges him to possess it. Legitimate alternatives exist (free/open-source tools like Python, VS Code, Eclipse; university academic licenses; student discounts). Since no duty forces him to pirate, this is a moral issue, NOT a moral dilemma.

Morality is guided by rules of conduct at three levels:

• Societal rules/laws: copyright law and licensing agreements — the student violates them (piracy is illegal even for personal, educational use).
• Professional rules: software developers' work is protected intellectual property; unauthorized copying breaks professional norms of respecting others' work.
• Self-imposed rules: personal integrity and honesty — using something one has no right to use erodes one's own standards.

• Consequentialist view: The student argues "no one is harmed — I don't sell it." But aggregated piracy causes industry-wide revenue loss, reduces developers' incentive to create, and raises prices for paying users. The utilitarian case is not clearly in the student's favor.
• Deontological view: Using software without permission is dishonest and unjust — taking a benefit without the owner's consent. Universalizability test: if everyone who "couldn't afford" software pirated it, the software industry would collapse — the maxim fails; hence the act is wrong.
• Justice view: The student receives a benefit while others pay their fair share — free-riding is unfair.
• Mitigating factors: poverty and educational purpose reduce blameworthiness, but do not make the act right.

• Privacy — users' right to control their personal information.
• Autonomy & informed consent — people must knowingly agree to how their data is used.
• Honesty / transparency — hiding data sales is a form of deception.
• Trust — the invisible foundation of the company–user relationship.
• Non-maleficence — data sales expose users to identity theft, profiling, manipulation, spam.

Fair exchange requires that both parties knowingly benefit. Here:

• The company takes 100% of the benefit (millions of dollars).
• Users carry all of the risk (privacy loss, targeting, fraud) — without consent and without compensation.
• This is a one-sided, exploitative exchange: the company profits from something that belongs to the users. Even if data sales are common practice, "everyone does it" is not a moral justification.

• Distributive justice: benefits and burdens must be distributed fairly — here all benefits flow to the company, all burdens to users.
• Procedural justice: a fair process (informed, opt-in consent) is absent entirely. The users are denied any voice.
• Rights-based justice: privacy is a right; rights are not to be traded away secretly for profit.

• Short-term: millions in revenue — a utilitarian might be tempted.
• Long-term: once exposed (such secrets usually leak), the company faces lawsuits, regulatory fines (GDPR-style penalties), customer exodus, and reputation collapse. Societal trust in the digital economy erodes — harming everyone.
• Total long-term utility is negative: hidden sales destroy far more value than they create.

Impartiality requires that decisions be based only on morally relevant criteria (ability, skill, qualification), and that identical cases be treated identically, free of arbitrary bias.

• University wealth is not a job-relevant criterion in itself. Two candidates with equal skill must receive equal evaluation; here they don't → partiality confirmed.
• Bias laundering: past "better performance" of rich-university graduates largely reflects opportunity, networking, and socioeconomic advantage, not superior talent. The algorithm converts historical inequality into seemingly "objective" scores — discrimination with a mathematical disguise.
• Feedback loop: the AI hires rich-university graduates → future "historical data" shows they succeed → bias deepens. Inequality becomes self-perpetuating.

1. Audit the training data for historical bias before training; document known distortions.
2. Use only job-relevant features: skills tests, work samples, structured interviews, verified competencies — not university prestige, name, address, or other class proxies.
3. Remove/neutralize proxy variables correlated with wealth (extracurriculars that require money, unpaid internships as sole criteria).
4. Disparate-impact testing: compare selection rates across socioeconomic/demographic groups; investigate and fix significant gaps.
5. Human oversight & appeals: applicants can request human review of automated rejections.
6. Transparency: tell applicants what criteria are used and allow correction of input errors.
7. Continuous monitoring: re-audit after deployment because models drift and bias can re-emerge.

Act Utilitarianism (Bentham): judge this particular act by its own consequences in this particular situation.

• Expected benefit = probability of survival × amount of life saved.
• Elderly patient: low probability × fewer remaining years = low expected utility.
• Young patient: high probability × many remaining life-years = high expected utility.

Conclusion: give the dose to the young patient — this single act maximizes total expected well-being.

Rule Utilitarianism (Mill): choose the rule whose general adoption maximizes utility; then follow that rule.

• Best rule: "In triage, allocate scarce medicine by medical criteria — urgency of need and probability of survival — never by wealth, status, or personal preference."
• Why this rule maximizes utility: it saves the most lives system-wide, keeps public trust in hospitals (people won't fear being denied care for arbitrary reasons), and gives doctors a clear, fair protocol in chaotic disasters.
• Applying the rule to this case → the young patient's higher survival probability is the decisive medical criterion → same answer.
• Note: the rule permits probability-of-survival but forbids judgments like "the young person's life is worth more" as such — that would undermine equality and trust.

Both theories converge on the same decision, but by different routes. The doctor should also document the decision and involve an ethics/triage team if time permits, following standard medical triage codes.

W. D. Ross's Act Deontology (pluralistic deontology) holds that we have several prima facie duties — duties that are binding on the surface unless overridden by a stronger duty in the actual situation. Ross lists:

• Fidelity — keep promises and tell the truth.
• Beneficence — help others improve their condition.
• Non-maleficence — avoid harming others.
• Reparation — make up for wrongs done.
• Gratitude, Justice, Self-improvement.

Unlike Kant, Ross does not rank duties by a fixed formula — the context determines which duty is weightiest.

• Duty of Fidelity: keep your promise → help your friend study. Important, but the stakes are an exam — significant, yet not life-threatening, and the friend can study alone or with someone else later.
• Duty of Beneficence + Non-maleficence: your grandmother faces a serious health emergency — life or permanent harm may be at stake, and she needs you now.

The stakes, urgency, and irreplaceability of your presence make the grandmother's case overwhelmingly weightier.

• The duty that "wins" the weighing is called the actual duty — here: take your grandmother to the hospital.
• Ross adds that when we fail a prima facie duty, a duty of reparation arises: afterward, sincerely apologize to your friend, explain, and help him prepare at the earliest opportunity (e.g., late-night session, share notes, arrange alternative help).
• This shows Ross's theory is context-sensitive and humane: promises matter, but they do not bind us when life and health are on the line.

Every major engineering code (NSPE, IEEE) begins with: engineers shall hold paramount the safety, health, and welfare of the public (which includes workers). An engineer's approval is a professional certification of safety — it may never be sold to the highest bidder.

• Safety regulations (occupational-safety law, machine-guarding standards) set minimum mandatory protection levels.
• If the cheaper machine's injury risk exceeds legal limits, approving it is illegal — the engineer must refuse outright and may be personally liable.
• Even where legal, the law is a floor, never a license: legality does not equal safety or morality.

• The standard of care: what a reasonable, competent engineer would do in the same situation.
• No competent engineer, knowing test results show elevated serious-injury risk, would approve the machine. Doing so is professional negligence and misconduct, risking license revocation and criminal liability if workers are harmed.
• "The company told me to" is never a defense — engineers are personally and professionally responsible for their approvals.

The engineer's duty is not only to say "no" but to constructively solve the problem:

• Propose the safer machine, or retrofit the cheaper one with guards, light curtains, emergency stops, and interlocks.
• Present a full cost-benefit case to management: injury costs (medical, compensation, lawsuits, OSHA fines, downtime, lost morale and reputation) usually exceed the price difference — safety is often cheaper long-term.
• Document the test results and all communications in writing.
• If management insists on approval against the engineer's judgment: refuse to sign, escalate to higher management/board, and if serious danger remains, notify the appropriate authorities (organized whistle-blowing) as the codes direct.

• Modern engineering ethics and products-liability law require designers to anticipate reasonably foreseeable misuse — not just intended use.
• The scenario states that misuse is easily possible and reasonably expected. Therefore misuse is foreseeable by definition, and designing against it becomes a design obligation, not an optional courtesy.
• Modern product-liability doctrine: if foreseeable misuse causes injury and the design ignored it, the manufacturer is liable — a warning label alone will not protect the company.

The hierarchy of hazard controls ranks protections by effectiveness:

1. 1. Design out the hazard — the strongest control.
2. 2. Physical safeguards — guards, barriers, interlocks.
3. 3. Warnings & training — the LAST and weakest resort.

Warnings fail because they can be unread, forgotten, mistranslated, ignored under time pressure, or lost with the manual. When injury is serious and misuse is likely, relying on paper is negligence.

• Install an interlock: the machine physically cannot start or run unless protective equipment (e.g., a two-hand control, a guard door, a sensor detecting PPE) is engaged — making misuse impossible, not merely forbidden.
• Add physical guards, emergency stops, and fail-safe defaults.
• Keep the warning label and training — but only as a supplementary layer, never the only defense.

• Local residents: chronic health effects (respiratory illness, cancers), risks to children's development, reduced property values, contaminated air/water, constant anxiety.
• They are involuntary risk-bearers and a vulnerable population — poorer communities have less political power, less ability to move away, and often less information.

• Company/shareholders: profits.
• Government & wider region: taxes, industrial growth, cheaper products for distant consumers.
• Residents: some jobs — but also the entire localized burden.

• No — this is a classic case of environmental injustice: concentrated risks on the least powerful, diffuse benefits to the most powerful.
• Procedural injustice: residents were not consulted or informed (no informed consent — they cannot knowingly accept the risk).
• Distributive injustice: those who bear the risk receive only a fraction of the benefit; those who receive the most benefit bear almost none of the risk.
• Selecting a poor area may simply reflect that its residents cannot resist, not that the site is objectively best.

• Modern emission-control technology (scrubbers, filters, closed-loop systems) to cut toxicity.
• Buffer zones / build in designated industrial zones away from homes.
• Continuous emissions monitoring with public real-time data and independent regulators.
• Emergency-response plans, health screening for residents, insurance/compensation funds.
• Community benefit agreements: local hiring priorities, revenue sharing, residents' representatives on oversight boards.
• Genuine informed community consent before approval — the decision process itself must be just.

• Voluntary risk: knowingly and freely chosen, usually with compensation or benefit attached (smoking, mountain climbing — people accept surprisingly high voluntary risks).
• Involuntary risk: imposed on a person without their knowledge or consent (contaminated water, hidden hazards) — people tolerate far lower levels and object strongly.
• Key factors: choice, information, control, benefit/compensation, and consent.

Person A perceives the risk as far more acceptable — he made an informed trade-off (risk ↔ cheaper house), giving him control and benefit. Person B bears the identical physical risk but experiences it as imposed deception, which research shows people judge as far less acceptable and morally outrageous.

• Identical physical risk → different moral status, because acceptability is not just about magnitude but about how the risk was imposed.
• The factory/company has a duty to inform all residents (informed consent), a duty to reduce emissions, and possibly a duty to compensate Person B for the risk imposed without his knowledge.
• Hiding risk information treats residents merely as means to corporate profit — a deontological violation.

• Risk-perception research shows people fear immediate, visible, personally experienced risks more. Product 1's stomach discomfort would be perceived as more serious — consumers notice it instantly, complain, and abandon the product.
• Long-term risks suffer from temporal discounting: "years away" feels psychologically light, causes are hard to trace, and statistical/abstract harms feel unreal. So Product 2's grave danger is typically underestimated.

• Rational evaluation weighs severity × probability × irreversibility × duration.
• Mild, reversible discomfort (Product 1) — low total harm.
• Possibly irreversible serious disease affecting users for the rest of their lives (Product 2) — potentially enormous total harm, even if delayed.
• Conclusion: the delayed risk is likely the objectively greater one. Delay reduces perceived seriousness, not actual seriousness.

• It would be exploitative to exploit consumers' perception bias — marketing Product 2 as "gentle, no side effects" while knowing the long-term danger.
• Duty: conduct long-term testing, disclose delayed risks clearly and prominently (informed consent), and weigh delayed severe harm heavily in design decisions — precisely because consumers will not.

Expected annual harm = probability × severity:

• Risk A (flood): probability ≈ 1/100 per year = 0.01; severity = catastrophic (many deaths, massive destruction). Expected annual harm = 0.01 × catastrophic.
• Risk B (accidents): probability = very high (~52 events/year); severity = minor per event. Expected annual harm = 52 × minor.

Depending on actual magnitudes, the two expected values can even be similar — yet they feel completely different.

• Familiar, frequent, small risks (B) become normalized: "accidents are part of life." People adapt, take precautions, and tolerate them.
• Rare, catastrophic, dread risks (A) trigger strong anxiety: huge imagination, media amplification after any disaster, and the feeling that one event is uncontrollable and communal. People may demand total prevention of A while casually accepting B — even when B's cumulative toll is larger.
• So paradoxically: the frequent minor risk (B) is often considered more acceptable in everyday life, while the rare catastrophe (A) is psychologically less acceptable but may be objectively manageable.

Yes — but only under strict conditions (ALARP logic — "As Low As Reasonably Practicable"):

1. The probability is genuinely negligible — verified by data, not merely unmeasured.
2. No reasonably available safer alternative exists.
3. The benefits of the activity are substantial to society (coastal living, trade).
4. Active mitigation reduces both probability and consequences: flood defenses/levees (cut probability), zoning away floodplains, early-warning systems, evacuation plans, insurance (limit consequences).
5. Ongoing monitoring — because "once in 100 years" is an average, not a guarantee; two floods can occur in consecutive years.

A catastrophic risk accepted without these conditions (ignored, unmitigated, secret) is never acceptable — it is mere negligence.

This is the discussion stopper from Harris, Pritchard & Rabins: "People disagree about morality / everyone has their own morality, so moral questions cannot be rationally resolved and ethical discussion is pointless." It combines a claim of moral relativism with a claim of futility of ethical reasoning.

1. Self-defeating statement: "Nobody can say cheating is wrong" is itself a universal moral claim about what others may assert. If no one can make moral judgments, then that very judgment is also forbidden. The position destroys itself.
2. Disagreement ≠ absence of truth: people disagree about science, medicine, and history, yet we still seek rational answers. Disagreement shows the question is difficult, not that it is unanswerable.
3. Is/ought fallacy (descriptive → normative): from the fact that people do have different moral views (descriptive relativism), it does not follow that no view is correct (normative relativism). That inference is a non-sequitur.
4. Preference vs moral judgment: "I prefer tea" is a personal preference needing no defense. "Cheating is wrong" is a moral judgment — it is public, action-guiding, applies to others, and must be backed by reasons (fairness, honesty, harm to others). The student conflates the two categories.
5. Self-contradictory behavior: even strict relativists protest when they are cheated, demand fairness, and apologize — behavior that presupposes shared moral standards.

• Morality is not a private taste — it is a public system: a set of standards that (1) applies to everyone, (2) is publicly known, (3) is taught and mutually enforced, and (4) exists so people can cooperate and live together.
• Examinations are part of such a public system: they work only on trust and fairness. If everyone cheated, grades, certificates, and merit-based opportunity would collapse — the practice would self-destruct (the universalizability test fails: "everyone may cheat" makes examinations meaningless).
• Cheating wrongs other students (unfair competition), institutions (corroded trust), and society (unqualified graduates). These are objective harms, arguable with reasons — hence ethics discussion is both meaningful and necessary.

• Virtue ethics asks not "what rule?" or "what outcome?" but "what would a person of good character do?"
• The relevant virtues: honesty (no concealment), responsibility (own your professional work), integrity (technical judgment not corrupted by commercial pressure), courage (delivering bad news to management despite cost), care (concern for users' welfare).
• A person of practical wisdom would report the bug immediately. Hiding it would be an act of cowardice and self-interest that corrodes the developer's character — and repeated small dishonesties shape who we become.
• Professional identity: engineers are trusted to tell the truth about their work. Concealment betrays the very virtue that makes the profession possible.

Compare total consequences, short- and long-term, for everyone affected:

• Hiding the bug:
  • Users harmed by the bug (data loss, wrong outputs, safety/privacy failures — "most may never notice" ≠ none).
  • A time bomb: if discovered later — lawsuits, regulatory fines, refunds, massive reputation collapse, job losses, and lasting erosion of public trust in the company and in AI generally.
  • Company culture poisoned: staff learn bugs get buried.
• Reporting the bug:
  • 3-month delay + millions in cost — real but bounded, known, survivable.
  • Safe product, preserved trust, avoided catastrophic tail risk, and a reputation for reliability (itself a business asset).

Expected-utility analysis: the small certain cost of delay is far outweighed by the probable and potentially catastrophic harms of concealment. Utilitarianism also says: report.

• Duty of honesty / non-deception: shipping while silently hiding a known defect is deceiving every customer who pays for a working product.
• Kant — universalizability: can "engineers may conceal serious bugs to protect launch dates" be universal law? No — universalized, no one could trust any software, the software industry collapses, and the maxim destroys itself. Fails.
• Kant — humanity formulation: customers would be treated merely as means to revenue, denied the information they need as rational purchasers. Violation.
• Professional duties: duty of care to the public, fiduciary duty to the employer (which includes honest reporting — hiding a bug actually betrays the employer's long-term interest), and contractual duty to customers. → Duty: report.

• All three theories converge — this makes the decision unusually certain: report the bug.
• Strongest reason: deontology. The duty of honesty is categorical and immediate — it does not depend on estimating how harmful the bug turns out to be or on uncertain outcome predictions, which utilitarian reasoning requires (and which, if the bug were truly trivial, could theoretically license hiding it). The developer knows deception is wrong the moment he contemplates it; customers' right not to be deceived settles the matter before any calculation begins.
• Virtue ethics supplies the motivation and character behind acting on that duty, and utilitarianism the reassuring proof that honesty is also the smartest policy — but the binding moral force comes from the deontological duty of honesty and non-deception.
• Practical steps: report immediately in writing, quantify the risk, propose a fix plan and timeline, and if overruled on a serious safety/privacy issue, escalate to higher management, the ethics board, or regulators.