Definition
Two people agree on every fact about an AI system and still disagree about whether to use it. How can that be?
An ethical framework is a structured way of judging what is right. The main frameworks look at consequences (which choice leads to the best outcomes), duties and rights (which rules must never be broken), virtues (what a good person would do) and justice (whether benefits and burdens are shared fairly).
In simple words
Imagine looking at a building through four different windows. From one you see the garden, from another the road, from a third the roof. Each window shows something true, and none shows everything. Ethical frameworks are windows on a decision. Looking through several gives a fuller picture than looking through one.
Why it matters
Arguments about AI are often arguments between frameworks without anyone saying so. Knowing them helps you understand why people disagree, and to make your own reasoning clearer and stronger.
Key terms
- Consequentialism
- Judging an action by its results: the right choice produces the best overall outcome.
- Duty-based ethics
- Judging an action by whether it follows moral rules and respects rights, whatever the results.
- Virtue ethics
- Judging an action by whether it is what an honest, fair and caring person would do.
- Justice
- The fair sharing of benefits, burdens and opportunities among people.
- Autonomy
- A person's right to make their own informed decisions.
- Beneficence and non-maleficence
- Doing good, and avoiding harm: two of the four principles of medical ethics.
Explained in detail
Consequences: what produces the most good?
This framework, associated with the philosophers Jeremy Bentham and John Stuart Mill, judges an action by its results. The right choice is the one that produces the greatest overall benefit and the least harm for everyone affected. It suits many AI decisions: an early-warning flood model is good because it saves lives. Its weakness is that adding up benefits can hide harm to a few. A system that helps ninety people and badly harms ten may score well on totals while being deeply unfair to those ten.
Duties and rights: what must never be done?
This framework, linked to the philosopher Immanuel Kant, holds that some things are wrong even if they produce good results: lying, using people merely as tools, violating their rights. From this view, secretly recording people's faces is wrong even if it catches criminals, because it breaks a duty to respect their privacy and consent. Human rights law follows this kind of thinking. Its weakness is that duties can clash, and following a rule rigidly can sometimes cause avoidable harm.
Virtue: what would a good person do?
This framework, going back to Aristotle in Greece and found in many Indian traditions, asks not "which rule?" or "which result?" but "what kind of person, or organisation, do we want to be?" Would an honest company hide that its product uses AI? Would a caring developer release a tool they knew failed for village speakers? It is useful when rules run out, and it reminds us that ethics is about character as well as calculation.
Justice: who gains and who pays?
The justice framework asks how benefits and burdens are shared. Does a new AI service help mostly those who are already well off, while its errors fall on the poor? The philosopher John Rawls suggested a test: design the rules as if you did not know which person in society you would be. Many fairness debates about AI, from loans to policing, are at heart debates about justice.
The four principles of medical ethics
In 1979 Tom Beauchamp and James Childress set out four principles that doctors around the world use, and that apply well to AI in health and beyond: respect for autonomy (people decide for themselves, with full information), beneficence (act for people's good), non-maleficence (first, do no harm) and justice (share benefits and burdens fairly). An AI tool for patients should honour all four: patients should know AI is involved and be able to refuse, the tool should genuinely help, it should be safe, and it should work for every group of patients.
How it works
| Framework | Key question | Applied to a face camera at a station |
|---|---|---|
| Consequences | Which choice gives the best overall result? | Lives saved by catching criminals against harm from errors and lost privacy |
| Duties and rights | Does it respect people's rights? | Recording everyone without consent may violate their rights |
| Virtue | What would a fair and honest authority do? | Be open about the cameras and use them sparingly |
| Justice | Are benefits and burdens shared fairly? | Wrong matches fall more on some groups |
Try it yourself
Four windows on one decision
A hospital considers an AI tool that decides which emergency patients are seen first.
Match the argument to its framework
Which framework is each person using?
- "It saves more lives than it costs, so we should use it."
- "Recording people without consent is wrong, whatever the benefit."
- "An honest company would tell its customers."
- "The poor bear the errors while the rich get the benefits."
Common mistakes
- Thinking one framework is always right and the others are useless.
- Adding up total benefit and ignoring who is harmed.
- Using a framework only to defend a choice already made.
Check yourself
Summary
- Ethical frameworks are structured ways of judging right and wrong.
- Consequences, duties and rights, virtue and justice each highlight different concerns.
- Medicine's four principles, autonomy, beneficence, non-maleficence and justice, apply well to AI.
Finished reading?
Sources: UNESCO: Recommendation on the Ethics of Artificial Intelligence · OECD AI Principles