Artificial intelligence → Module 5 → Ethics and safety of AI

Ethics and safety of artificial intelligence

Artificial intelligence is a tool – and every powerful tool can be used well or badly, carefully or carelessly. In this lesson you will get to know the main ethical questions: bias, fake content, privacy, copyright, responsibility and safety – and the rules being made for them.

The part about laws describes the situation as of November 2026 and is updated once a year.

Bias and unfair decisions

From the lesson Data – food for AI we know: a model learns from data and inherits its flaws. When AI helps with important decisions, this can be serious:

• a program that helps choose job candidates learned from old decisions – so it also repeats old injustices;
• face recognition makes mistakes more often with people who were poorly represented in the data;
• a system for grading written work gives lower grades to a different writing style.

That is why models are tested on different groups of people, and in important decisions (jobs, school, health, courts) the final word should belong to a person who can explain the decision.

Fake images, videos and voices

Generative AI makes images, video and voices that look real – this is called a deepfake. It can be used for fun and film, but also for lies, scams and hurting people.

Five checking steps

1. Who published it? A well-known and reliable source, or an account without a name?
2. Do other reliable media report it too? Big news doesn't appear in only one place.
3. Details: hands and fingers, text in the background, shadows, reflections in the eyes, lip movements that don't follow the voice.
4. Reverse image search: was the same image published earlier in a different context?
5. Emotion: does the content want to make you angry or scared so that you share it quickly? Then stop and check twice.

Never make fake images, videos or voices of other people – it can seriously hurt them, and in many countries it is also a crime.

Privacy

• AI models learn from huge amounts of data, which can include data about people.
• What you write to a chatbot, the images you send and voice messages can be stored on servers.
• Cameras with face recognition can track where people go – that is why they are strictly limited in many countries.

A rule for you: don't share your own or other people's personal data; for school projects record objects, not people (as in the Teachable Machine exercise).

Copyright: whose work is it?

• Models for images and text learned from the work of millions of people – artists, writers, programmers. Was that all right without their permission? There are court cases and debates about this, and the answer isn't clear yet.
• If AI makes an image from your description, who is the author – you, the company or nobody? The rules differ from country to country.
• A simple rule applies at school: always say what was made by AI and don't hand in AI work as your own.

Who is responsible when AI makes a mistake?

A self-driving car hits a fence. Who is to blame – the owner, the company that made the car, the programmers, or the people who prepared the data? Such questions are new, and laws are only beginning to deal with them. The important principle is: responsibility always lies with people – those who make, sell and use the AI system. “The computer did it” is not an excuse.

Energy and the environment

Training large models and answering millions of questions a day uses a lot of electricity and water for cooling data centers. That is why more efficient models and chips are being developed – and we can be sensible too: we don't use a big model for something a calculator can do, and we choose a smaller model when it is enough (remember TinyML).

Rules and laws

• The European Union has adopted the Artificial Intelligence Act (AI Act), which came into force in 2024 and is applied gradually. Since February 2025 some practices have been banned (for example, systems that “score” people by their behavior or manipulate them), since August 2025 obligations for large general-purpose models apply, and obligations for AI in high-risk areas (employment, education, health care…) apply later – according to the 2026 amendments, from the end of 2027 and 2028. The law also requires content made with artificial intelligence to be labeled.
• Serbia has adopted the Strategy for the Development of Artificial Intelligence for 2025–2030, and on its way to the EU it is aligning its regulations with European rules.
• There are rules at school too: check what is allowed in tests, competitions and homework.

Activity: the ethics traffic light

For each example, decide: green (it's OK), yellow (it depends – under what conditions?) or red (it's not OK). Explain your choice.

1. A student uses AI to explain the material before a test.
2. A student makes a funny fake photo of a classmate and sends it to a group chat.
3. A shop uses cameras with face recognition to track customers.
4. A doctor uses AI to help spot a disease on a scan, and makes the decision herself.
5. A student hands in an essay written by a chatbot and claims it is his own.
6. A city uses AI to predict traffic jams and adjust the traffic lights.

Common misconceptions

• “A computer is objective, so AI is always fair” – AI inherits flaws from the data and from the decisions of the people who made it.
• “If something was made by AI, nobody is responsible” – responsibility always lies with the people who make and use the system.
• “A fake image is easy to recognize” – the newest fake content is very convincing; that is why you check the source, not just the look.

Practice tasks

Try it yourself first, and only then open the solution.

1. A model for choosing scholarship candidates learned from the decisions of the previous 10 years, when scholarships almost always went to students from one school. What can go wrong, and how can it be prevented?
2. Name five checking steps before you share a shocking video.
3. Why is a “secret word” recommended in a family?

Check your knowledge