COMP90087 The Ethics of Artificial Intelligence

How COMP90087 is assessed

What COMP90087 covers

COMP90087 builds a moral-reasoning toolkit — principlism plus utilitarianism, deontology and virtue ethics — in Weeks 1-5, then applies it to accountability, transparency, algorithmic bias, data governance and generative AI in Weeks 7-12. It is assessed by tutorial participation (10%), a Week-6 closed-book in-class essay (30%), a Week-12 research essay (30%) and a closed-book examination-period exam (30%) that is a hurdle you must pass.

This map follows the real UniMelb teaching order so you can revise each week toward those assessments.

Weeks 1-5 assemble a moral-reasoning toolkit — the history of AI, the ethics-versus-law distinction and principlism (beneficence, non-maleficence, justice, respect for autonomy), then the three classical theories: utilitarianism (consequences), deontology (duty) and virtue ethics (character).

Weeks 6-12 apply that toolkit to live AI problems — creative economies, trust and accountability, transparency and explainability, algorithmic bias and equity, data governance, contemporary issues (governance, sustainability, right to repair) and generative AI.

It is a reasoning-and-writing subject: marks come from applying a theory to an AI case, arguing a clear position, and recalling who-owns-which-concept, not from calculation.

Assessment runs through Canvas as four components: tutorial participation and online discussion (10%, 1% per tutorial across Weeks 2-12, best 10 of 12 count); a Week-6 closed-book handwritten in-class essay (30%, ~90 minutes, ~1000 words, applying Weeks 1-5 theory to a scenario); a 1500-word take-home research essay (30%, due Week 12); and a 120-minute closed-book digital exam (30%) sat on campus via the LockDown browser, with roughly 30-40 multiple-choice questions (including “select all that apply”) plus 3-4 short-answer questions.

The exam is a hurdle requirement: score 49% or below on it and you fail the whole subject (grade NH), no matter how strong your essays are. This 12.5-credit-point subject feeds the Weighted Average Mark (WAM), and its H1 / H2A / H2B / H3 / P grade bands sit behind every result.

Worked example · free

One AI case, three theories: how utilitarianism, deontology and virtue ethics diverge

Key terms

Principlism

Applying four classic bioethics principles — beneficence (do good), non-maleficence (avoid harm), justice (fairness) and respect for autonomy — to AI decisions. Named explicitly in the in-class essay prompt; the four principles are the scaffold for evaluating a deployment.

Principle of utility

The utilitarian criterion (Bentham/Mill): the right action maximises overall happiness — the greatest total balance of pleasure over suffering across everyone affected, counted impartially. In an AI case you sum benefits and harms and pick the option with the greatest net utility.

Categorical imperative (Formula of Humanity)

Kant's duty-based test: always treat rational beings as ends in themselves, never merely as means. Grounds AI limits on deception, non-consensual data use and uncontestable automated decisions, independent of good outcomes.

Phronesis (practical wisdom)

In Aristotle's virtue ethics, the master virtue needed to apply the other virtues to concrete situations and adjudicate when they conflict. Shannon Vallor recasts it as “technomoral wisdom” for technology design.

Accountability (Nissenbaum)

Being answerable for an outcome: blameworthy or praiseworthy, liable to sanction or restitution, and able to give an account of what was done. Nissenbaum names four barriers that erode it in computerised systems (many hands, bugs treated as inevitable, computer-as-scapegoat, ownership without liability).

Hurdle requirement

An assessment component you must pass on its own to pass the subject, regardless of your total. In COMP90087 the 120-minute exam is a hurdle: 49% or below on it means grade NH (fail), even with strong essays.

COMP90087 FAQ

Questions students ask in each COMP90087 chapter

Do I need to memorise every date and machine in the AI-history timeline?

Not every date, but you should know the sequence and what each milestone was or was not — especially “programmable” vs “general-purpose” (the Jacquard loom is programmable but not a general computer; Babbage’s Analytical Engine is the general-purpose design, with Ada Lovelace as first programmer). The highest-value single item is the AI-winter definition, followed by the two winters’ causes and the revival triggers (compute, big data, backpropagation). Chapter 01 · The History of Artificial Intelligence

What exactly is ‘principlism’ and why does it matter for assessment?

Principlism is the practice of applying four bioethics principles — beneficence, non-maleficence, justice and respect for autonomy — to an AI decision. It matters because the in-class essay prompt names it explicitly, so you may be asked to defend a position on an AI scenario by trading these four principles against each other. Practise applying all four to a fresh case and reaching a conditional verdict. Chapter 02 · Introduction to AI Ethics and Principles

How do I actually compute utility in the exam’s numeric question?

For each option compute net utility = (number pleased × per-person gain) − (number annoyed × per-person loss), then pick the option with the highest total. The classic trap is to choose the option that helps the most people rather than the one with the greatest net balance — always work out every option’s number before selecting. Chapter 03 · Utilitarianism

What is the exact wording examiners want for Kant’s Formula of Humanity?

“Always treat rational beings (humanity, in yourself and others) as ends in themselves, never merely as means.” This appears as a near-verbatim MCQ, so learn it word-for-word, and be ready to apply “merely as means” to an AI case (deception, non-consensual data use, uncontestable automated decisions). Chapter 04 · Deontology

How do I not mix up which side of the mean a vice is on?

Fix the pattern deficiency → virtue → excess and test every pair against it: cowardice–courage–rashness, stinginess–generosity–extravagance, self-deprecation–truthfulness–boastfulness. The matching MCQ specifically checks whether you can say a vice is the deficiency or the excess, so practise a handful until the pattern is automatic. Remember the mean is context-relative, not the arithmetic middle. Chapter 05 · Virtue Ethics

Is Week 6 examinable even though it’s the essay week?

Yes. There is no new theory lecture in Week 6 (the essay-skills session itself is the one non-examinable lecture), but the AI-and-creative-economies module content is examinable in the closed-book exam. Know the four orienting questions and the four tensions — authorship/IP, labour, authenticity, platform power — well enough to apply them to a scenario. Chapter 06 · AI Ethics and the Creative Economies

What are the five mechanisms of harm and why do they matter?

Datafication, automation, scalability, transformative effects and inscrutability/opacity. They answer the week’s framing question — whether AI’s harms are distinctive or just more of the same — by showing how AI amplifies and reshapes harm (e.g. scaling a biased scoring model across many districts before anyone can intervene). Know them as a named set and tie each to one of the anchor books (O’Neil, Eubanks, Scott). Chapter 07 · Trust, Power, and Accountability

Why isn’t a highly accurate model good enough on its own?

Because accuracy and explainability are different things: an accurate model can still give an affected person no reason they can understand or act on. Without intelligible, actionable reasons, decisions can’t be contested, no one can be held accountable, and a discriminatory proxy can hide behind the accuracy figure. Procedural fairness can outweigh raw accuracy — that is the core of the loan-approval item. Chapter 08 · Transparency and Explainability

How should I structure the training-data-bias short answer?

As three tight sentences worth one mark each: (1) definition — data over/under-represents groups, skewing outputs; (2) why it’s wrong — unjust, often invisible discrimination at scale; (3) a concrete example — facial recognition performing worse on darker skin, or an AI recruitment tool down-ranking an under-represented group. Keep it to about 40 words and do not pad. Chapter 09 · Algorithmic Bias, Accessibility, and Equity

What’s the exam trick with the GDPR principles?

The item lists several plausible-sounding options and asks which is NOT a GDPR principle. The seven real ones are lawfulness/fairness/transparency, purpose limitation, data minimisation, accuracy, storage limitation, integrity & confidentiality, and accountability — so a term like “epistemic justice” (real ethics, wrong list) is the impostor. Learn the seven and the distractor jumps out. Chapter 10 · Data Governance

What does the exam want on the ‘AI Governance Cascade’?

It tests the direction of the layered model: broad principles sit at the higher levels (e.g. the Commonwealth sets broad principles, not a system’s parameter count), and narrow, local practice — like a hospital ward — sits lower. Distractor options flip these relationships, so be able to say which level sets broad principles and which handles local implementation. Chapter 11 · Contemporary Issues in AI Ethics

What’s the two-category framework for analysing chat-LLM use?

The lecture splits the implications into knowledge/epistemic (how tools affect information habits, learning, belief formation and the shared web) and wellbeing/flourishing (how they affect psycho-social needs, relationships, dependence and authenticity). Pairing these with the affordances lens — what a tool invites, enables and constrains — gives you a reusable structure for any generative-AI case, including the research essay. Chapter 12 · Generative AI and Ethics

What makes the exam a ‘hurdle’ and why does it matter so much?

The 120-minute closed-book exam (30%) must be passed at 50% or above on its own; score 49% or below and you receive grade NH — a fail for the whole subject — no matter how strong your two essays are. That is why exam revision cannot be left to the end: it gates the entire subject, so being able to start every topic matters more than polishing a few. Chapter 13 · Essay and Exam Strategy

How to study for the exam

Treat COMP90087 as two linked jobs — build the toolkit (Weeks 1-5), then apply it (Weeks 6-12) — and rehearse both weekly rather than cramming through SWOTVAC, because the recall load (theories, thinkers, definitions, cases) is the real difficulty.

First, lock the three-theory map cold: utilitarianism = consequences, deontology = duty, virtue ethics = character, with the signature machinery under each (principle of utility; Kant's Formula of Humanity and Ross's prima facie duties; Aristotle's golden mean and phronesis, plus Vallor's technomoral virtues and care ethics).

Make a who-said-what sheet, because the exam repeatedly tests attribution (phronesis = Aristotle, technomoral virtues = Vallor, Gilligan and Tronto = care ethics, which item is NOT a GDPR principle). Second, drill application: take one AI scenario a week and argue it once under each theory so you feel where they agree and diverge — that is the exact skill both essays are marked on.

Third, practise the two written formats: the argumentative essay (clear position, define your theory, apply it to case detail, rebut one objection, reasoned conclusion — evaluate, don't list) and the ~40-word short answer (one mark per component: state the concept, say why it matters ethically, give an example).

Because the exam is a closed-book hurdle you must pass at 50%, prioritise being able to start every topic over perfecting a few. When a distinction won't stick, ask Sia to explain it a different way and set you a fresh practice item; it teaches the method and checks your reasoning, and it never substitutes for your own graded work.

Confirm the examination-period date, room and rules on Canvas and the University of Melbourne exam timetable.

Your AI Computer Science tutor for COMP90087

Stuck on a hard COMP90087 question? Sia is AskSia’s AI Computer Science tutor — ask any COMP90087 The Ethics of Artificial Intelligence question and get a clear, step-by-step explanation grounded in how the course is actually taught and assessed. Read this whole study guide free, then take your hardest questions to Sia.