I10 · Responsible AI in Practice

Ethical Frameworks for AI

Ethical frameworks are systematic ways of deciding what is right. Consequentialism, including utilitarianism, judges actions by their outcomes, aiming for the greatest overall benefit. Deontology judges actions by duties and rules, such as never deceiving people, whatever the outcome. Virtue ethics asks what a person of good character, honest, fair and caring, would do. Justice-based approaches ask whether benefits and burdens are fairly shared, especially for the least advantaged. Rights-based approaches protect individuals' rights, such as privacy and dignity. Indian traditions add ideas such as dharma (duty) and ahimsa (non-harm), which shape debates in India too.

12 min Intermediate · Sublevel Advanced Jump to the activity

What you need: The earlier topics of this module.

Definition

A hospital's AI can save more lives overall by sending scarce ICU beds to the patients most likely to recover, but that may mean older patients are rarely chosen. Is that right? Different ethical traditions give different answers, and knowing them helps you reason clearly instead of guessing.

Ethical frameworks are systematic ways of deciding what is right. Consequentialism, including utilitarianism, judges actions by their outcomes, aiming for the greatest overall benefit. Deontology judges actions by duties and rules, such as never deceiving people, whatever the outcome. Virtue ethics asks what a person of good character, honest, fair and caring, would do. Justice-based approaches ask whether benefits and burdens are fairly shared, especially for the least advantaged. Rights-based approaches protect individuals' rights, such as privacy and dignity. Indian traditions add ideas such as dharma (duty) and ahimsa (non-harm), which shape debates in India too.

In simple words

Imagine four advisers. One asks, "What gives the best result for the most people?" Another asks, "What rules must never be broken?" A third asks, "What would a good person do?" The fourth asks, "Is this fair to everyone, especially the weakest?" Good AI decisions listen to all four.

Overview

  • An ethical framework is a structured way of deciding what is right or wrong.
  • Different frameworks can give different answers to the same AI decision.
  • The classic families are consequences (utilitarianism), duties (deontology), character (virtue ethics) and fairness (justice).
  • Applied AI uses principle lists such as fairness, transparency, accountability, privacy and safety.
  • Organisations such as the OECD (2019) and UNESCO (2021) have published AI ethics principles.
  • Using several frameworks together helps teams spot problems one framework would miss.

Why it matters

AI systems make or shape decisions about jobs, loans, health, education and policing at enormous scale. Ethical frameworks give a shared vocabulary for spotting trade-offs, explaining decisions and arguing about them reasonably, rather than relying on gut feeling or on whatever is easiest to build.

Key terms

Consequentialism
Judging actions by their results; utilitarianism seeks the greatest good for the greatest number.
Deontology
Judging actions by duties and rules, such as honesty and respect, regardless of outcome.
Virtue ethics
Asking what a person of good character would do.
Distributive justice
Fair sharing of benefits and burdens across people and groups.
Stakeholder
Anyone affected by a system: users, people it decides about, workers, society.
Trade-off
A choice where gaining one good means giving up some of another.

Types

There are several types of ethical framework. Each asks a different question about an AI system.

1. Consequentialism (utilitarianism)

Judging actions by their outcomes: the right choice produces the most overall benefit and least harm.

  • How it works: List who is affected, estimate benefits and harms for each group and choose the option with the best total.
  • Example: Deploying a disease-screening AI that catches many more cases, even if it adds some false alarms.
  • When to use it: Use it for cost-benefit decisions, while watching that minority groups are not sacrificed for the average.

2. Deontology (duty-based ethics)

Judging actions by whether they follow moral rules and respect rights, regardless of outcomes.

  • How it works: Identify duties such as honesty, consent and non-discrimination and reject options that break them.
  • Example: Refusing to train on personal photos without consent even if it would improve accuracy.
  • When to use it: Use it to set non-negotiable limits such as privacy, consent and human dignity.

3. Virtue ethics

Judging actions by what a person of good character, honest, fair and caring, would do.

  • How it works: Teams ask what their choices say about them and whether they would be proud to explain them publicly.
  • Example: An engineer speaking up about a misleading AI demo instead of staying silent.
  • When to use it: Use it to build a professional culture of honesty, humility and care.

4. Justice and fairness frameworks

Judging actions by how benefits and burdens are shared and whether people are treated equally.

  • How it works: Rawls's "veil of ignorance" asks which rules you would choose without knowing which group you belong to.
  • Example: Checking that a loan model does not give worse outcomes to applicants from rural areas.
  • When to use it: Use it whenever an AI system allocates opportunities such as jobs, loans or school places.

5. Principle-based AI ethics guidelines

Practical lists of principles published by governments and organisations.

  • How it works: Teams turn principles such as fairness, transparency, accountability, privacy and safety into checklists, reviews and metrics.
  • Example: India's NITI Aayog "Responsible AI for All" papers (2021) setting principles for AI in India.
  • When to use it: Use them as a shared vocabulary for project reviews and policies.

Explained in detail

Applying the frameworks to one case

Consider an AI that screens loan applications in a small-finance bank. A consequentialist asks whether the system increases total welfare: more good loans approved, fewer defaults, lower costs passed to borrowers. A deontologist asks whether applicants are treated with respect: are they told an AI is involved, given reasons for a refusal and allowed to appeal, and is their data used only for the purpose they agreed to? A virtue ethicist asks whether the bank is acting honestly and with care, for example not pushing loans on people who cannot repay. A justice perspective asks who bears the errors: if the model wrongly rejects first-time borrowers from rural areas more often, it deepens existing inequality even if overall accuracy is high. Each lens catches problems the others might miss.

FrameworkKey questionRisk if used alone
ConsequentialismWhich choice gives the best overall outcome?Can sacrifice minorities for the majority
DeontologyWhich duties and rights must be respected?Rules can clash or ignore serious consequences
Virtue ethicsWhat would an honest, caring professional do?Gives less precise guidance
JusticeAre benefits and harms fairly shared?Fairness can be defined in conflicting ways

From philosophy to principles

Governments and organisations have turned these ideas into AI principles. The OECD AI Principles, adopted in 2019 and updated in 2024, call for inclusive growth, human rights, transparency, robustness and accountability. UNESCO's Recommendation on the Ethics of Artificial Intelligence, adopted by its 193 member states in November 2021, is the first global standard on AI ethics. In India, NITI Aayog's "Responsible AI for All" papers of 2021 set out principles of safety and reliability, equality, inclusivity and non-discrimination, privacy and security, transparency, accountability, and the protection and reinforcement of positive human values, drawing on the Constitution's fundamental rights. Principles are only a start; the next topics show how to put them into practice.

Frameworks are tools for thinking, not machines that output answers. Two careful people using the same framework can still disagree, because they weigh consequences differently or see different duties as more important. That is why ethical decisions about AI should be made by diverse groups rather than individuals, and why the reasons behind a decision should be written down. A written rationale lets others check it, challenge it and learn from it, and it shows regulators and the public that the organisation took the questions seriously.

How it works

  1. List everyone affected by the AI system, including people who never use it directly.
  2. Ask the consequence question: who benefits, who is harmed and how much?
  3. Ask the duty and rights questions: consent, honesty, dignity, appeal.
  4. Ask the justice question: are errors and benefits fairly shared?
  5. Record the trade-offs you chose and why, and revisit them as evidence comes in.

Try it yourself

Allocating ICU beds

During a disease outbreak, a district hospital considers an AI tool to rank patients for its 20 ICU beds.

Match the question to the framework

Which framework asks each question?

  • Will this save the most lives?
  • Are we being honest with users?
  • What would a caring doctor do?
  • Do poorer districts bear more of the errors?

Common mistakes

  • Using only overall accuracy or profit as the measure of "good".
  • Ignoring people who are affected but never use the system.
  • Treating published principles as enough without practical checks.

Check yourself

Question 1

1. Which framework judges an action mainly by its outcomes?

Question 2

2. "Never deceive users, even if it would increase sales" reflects…

Question 3

3. Which international body adopted a global Recommendation on AI ethics in 2021?

Summary

  • Consequentialism, deontology, virtue ethics and justice each highlight different moral questions.
  • Using several lenses reveals trade-offs a single measure hides.
  • OECD, UNESCO and NITI Aayog principles turn ethics into shared commitments for AI.

Finished reading?

Sources: UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) · NITI Aayog: Responsible AI for All (2021) · OECD AI Principles