AIethics.
Bias, responsibility, transparency: from the start. Not as an appendix, not as PR. As the foundation for everything that follows.
Ethics is not a
brake.
In the tech world, ethics has a bad reputation. It is seen as a disruption to progress, the thing that slows teams down and complicates products. That is wrong. Ethical AI is better AI: more precise, more reliable, fairer.
When systems harm people, there is a cost. Trust, money, human dignity. A facial recognition system that recognises darker skin less well is not a "value-free" tool. It is a system that discriminates. The question is not whether AI has values. It always does. The question is: Whose values? And who gets to decide?
Every AI system encodes values. In the dataset, in the design, in the decisions of the teams that build it. Ethics in AI is not an add-on. It is the question of which values get built in, and who controls that.
Algorithmic fairness
An algorithm is not neutral just because it is a machine. It reflects the data it was trained on and the decisions of the people who selected that data. Fairness is measurable, but measurement is a choice too.
Algorithmic bias
Bias arises when systems systematically disadvantage certain groups, often unintentionally, but with real consequences. It can be hidden in the training data, in the model design, in the way success is defined.
Accountability
When an AI system causes harm: who is responsible? The developer? The company? The user? This question is often unclear, and that is no accident. Lack of clarity protects those who hold power.
Transparency & conscience
Transparency does not mean making everything public, but making it possible to explain how decisions come about. Conscience means asking these questions before you ship. Not afterwards.
This is what bias
looks like.
Bias is not an abstract term. It shows up in concrete systems with real effects on real people. Choose a scenario and see how systems treat different groups differently.
Distributing
responsibility.
When an AI system harms people, the same thing often happens: everyone points at someone else. Developers at the data. Companies at the users. Governments at the technology. This diffusion of responsibility is structural, and politically convenient.
Responsibility is not a binary question of "guilty" or "not guilty". It is a system of different actors carrying different kinds of responsibility.
| Actor | Type of responsibility | Typical excuse | Accountability |
|---|---|---|---|
| Developers | Technical decisions: architecture, data, metrics | "I just write code" | High |
| Companies | Deployment, product decisions, business model | "We have terms and conditions" | High |
| Users | Using the tool, passing on outputs | "I only use what is on offer" | Medium |
| Governments | Regulation, standards, enforcement | "Technology moves too fast" | Medium |
| Data suppliers | Quality and representativeness of the data | "We only collected it" | Medium |
| Affected groups | Formally: none, but 100% of the consequences | none | No power |
The group that is most affected typically has the least control and say. That is no accident. It is the structure of power in tech systems.
Algorithms do not discriminateRREBELL AI ethics module
because machines are evil,
but because people were.
What ethical AI
means.
There are many frameworks. Many of them are PR. These six principles are a practical foundation: not a cure, but an honest start.
Not equally wrong for everyone
Fairness does not mean everyone gets the same. It means nobody is systematically treated worse, especially not groups that are already marginalised.
Explainable, not infallible
Systems must be explainable: to those affected, to developers and to regulators. "The model decided" is not an explanation. It is a refusal of responsibility.
Someone has to answer for it
When a system causes harm, there must be a named, reachable person or body responsible. Not a structure that spreads responsibility until it disappears.
Before you deploy
Not afterwards. Not when the press reports on it. Conscience means asking the uncomfortable questions: Who does this harm? Whose data? Who profits, and at whose expense?
Involve those affected
The people most affected must have the biggest say. Not as a "user test" but as structural co-determination. This is the hardest principle.
The minimum standard
At the very least. No system that predictably causes harm should be deployed, however profitable. That sounds obvious. It is not.
No right
answers. Almost.
Ethics is not a multiple-choice exam, but some answers are still better than others. These exercises train the critical thinking you need to question AI systems, whether you build them, use them or regulate them.
Task: A company develops an AI system for HR decisions. It was trained on 10 years of historical hiring data. During that period 87% of the company's staff were men. The system now mostly recommends men for leadership roles.
Choose the statement that best describes the problem:
A hospital uses an AI system that assesses patient risk. The system systematically assigns Black women patients lower risk scores than white women patients with identical health conditions, because it was trained on historical cost data and Black patients in the US had on average received less healthcare (not because they were healthier, but because they had less access). Result: Black women patients are less often referred for intensive treatment.
In your view, who bears the most responsibility, and why? Write your answer in 3 to 5 sentences.
Your assessment:Task: Put these ethical principles in your personal order of priority: drag and drop them into the order you think is right. There is no objectively correct answer, but you will be asked to explain your decision.
A city is considering an AI system for crime prevention. It analyses public cameras and behaviour patterns to flag "suspicious persons". It has a 73% hit rate, but a 27% false-positive rate, and these false alarms disproportionately affect people of colour and queer people.
According to the manufacturer, the system would raise the clearance rate by 18%.
Several people argue differently. Choose the position most compatible with ethical use of AI:
Mariam, 34, is classified by the job centre's AI system as having "low employability". Her benefits are then cut by 40%. She receives no letter explaining why, only a standard form saying: "Your assessment is based on an automated decision."
Write an explanation for Mariam as it should be: understandable, honest, without legalese. What did the system decide? On what basis? What can she do?
Remember: transparency means the person can actually understand it , not that something has been explained in legal terms.
Last exercise. This one is personal. Imagine you are hired as an AI developer or AI decision-maker on one of the following projects. You are well paid. Which would you take on, and which not? Briefly explain why.
A system that pre-sorts asylum applications and creates risk profiles. Goal: cut processing time. Known: the training data reflects caseworkers' prejudices.
An AI for queer young people offering anonymous counselling and resources in countries where homosexuality is illegal. Risk: privacy and state surveillance.
A recommendation algorithm for news articles that deliberately prioritises emotionally charged content because it increases time on site. The company says: "We only deliver what people want."
An audit tool that examines existing AI systems for bias and produces reports for regulators. Low profit. Big impact.
of technology.
It is a question of power.
Who decides what gets optimised? Who bears the consequences? Who has a voice? These are the real questions behind every AI system.
← Back to the learning hub