Definition
A self-driving car's brakes fail. It can stay in its lane and hit three people, or swerve and hit one. Who should decide what it does, and when?
An ethical dilemma is a situation in which every available choice has a serious moral cost, so that no option is simply right. AI creates new dilemmas because people must decide in advance, in rules and goals, how a machine will act in situations that may arise later.
In simple words
A doctor with one bed and two critically ill patients faces a dilemma: whoever is not chosen may suffer. People face such moments rarely and decide in the moment. A machine cannot decide in the moment the way a person does. Someone has to write the decision down beforehand, for thousands of cases, and then live with the result.
Why it matters
AI is increasingly used where choices affect safety, freedom and opportunity. Recognising a dilemma, and seeing that it involves values and not only technology, is the first step towards deciding it fairly and openly.
Key terms
- Ethical dilemma
- A choice in which every option has a serious moral cost.
- Trade-off
- Giving up some of one good thing to gain more of another.
- Stakeholder
- Anyone affected by a decision.
- Trolley problem
- A famous thought experiment about whether to redirect harm from many people to fewer.
- Value judgement
- A decision based on what matters most, which facts alone cannot settle.
Explained in detail
The trolley problem goes digital
In 1967 the philosopher Philippa Foot described a runaway tram heading towards five workers. You can pull a lever to send it down a side track, where it will kill one worker instead. Should you? For decades this was a classroom puzzle. Self-driving cars made it practical, because a car's software must be told in advance what to prefer if a collision cannot be avoided. Between 2016 and 2018, researchers at MIT ran an online game called the Moral Machine, in which millions of people from 233 countries and territories chose whom a driverless car should spare. Their answers showed broad agreement on some points, such as sparing more lives and sparing children, and clear differences between cultures on others.
Engineers point out that real crashes rarely look like the puzzle, and that the best thing a car can do is brake hard and avoid such situations altogether. The deeper lesson stands: a decision that a person would make in panic becomes, in a machine, a policy chosen calmly in advance, and someone must be responsible for that policy.
Dilemmas that are already here
| Situation | The competing goods |
|---|---|
| Face recognition in public places | Catching criminals against everyone's privacy and freedom |
| Removing harmful posts automatically | Protecting people from abuse against free expression and wrongful removal |
| Using health data to train medical AI | Better care for future patients against today's patients' privacy |
| Children's learning apps that track progress | Personalised help against constant monitoring of children |
| Screening loan applications quickly | Speed and lower cost against the risk of unfair refusals |
How to think through a dilemma
There is rarely a formula, but there is a method. Name everyone affected, including people who are not in the room. Ask what each option would cost each of them, and how likely and how severe the harm would be. Look for a third option that reduces the dilemma: better design, a human check, a limit on use. Decide openly, record why, and arrange to review the decision when you learn more. Above all, recognise that choosing a goal or a threshold for an AI system is a moral choice, even when it looks like a technical setting.
Why ordinary people should have a say
Engineers can say what is possible, but they have no special authority to say what is right. Dilemmas involving public safety, privacy and fairness affect everyone, so they belong in public debate, in law and in the decisions of elected representatives. Listening to the people most affected, who are often the least consulted, is part of deciding well.
How it works
- Describe the situation and the decision to be made.
- List everyone affected.
- For each option, note the benefits and harms to each group.
- Look for a better option that reduces the conflict.
- Choose, explain why in plain words, and plan to review.
Try it yourself
A dilemma at the city council
The council must decide whether to install face recognition cameras at the railway station.
Moral dilemma vote
Play with a group. There are no right answers, only reasons.
- Read this case aloud: a school can use an AI tool that predicts which students may fail, so teachers can help them early, but the predictions are sometimes wrong and labels can stick.
- Everyone votes: use it, do not use it, or use it with conditions.
- Each person gives one reason for their vote.
- Together, list the conditions that would make it acceptable.
- Vote again. Did anyone change their mind?
Think about it
- Who is affected but was not represented in your group?
- Which argument from the other side was strongest?
Common mistakes
- Believing every dilemma has a technical solution.
- Treating a setting such as a threshold as neutral, when it decides who bears the harm.
- Leaving the decision only to engineers or only to the company.
Check yourself
Summary
- An ethical dilemma has no option without a serious moral cost.
- AI turns split-second human choices into policies decided in advance.
- Name everyone affected, seek better options, decide openly and review.
Finished reading?
Sources: MIT Moral Machine · UNESCO: Recommendation on the Ethics of Artificial Intelligence