AI bias is not about a machine secretly deciding to be unfair. Unequal outcomes can emerge from the data, labels, measurements, goals and design choices around a system. This challenge uses evidence and a small simulation to investigate how that happens and what people can do about it.
Bias can enter through data, labels, goals, measurement or deployment. A system can also make errors that are not evidence of unfairness by themselves.
You do not need a real AI model to see the problem. A small decision exercise can show how incomplete examples create misleading conclusions.
Real cases are more useful than invented horror stories. Focus on what happened, the suspected cause, who was affected and what changed afterwards.
I can summarise a documented bias case without overstating what the evidence proves.A system can treat everyone with the same rule and still produce unequal outcomes. In other situations, changing rules for different groups can create new questions. Fairness involves values as well as maths.
Your finished audit should avoid blaming “the AI” as if it has intentions. Focus on data, measurement, design choices, deployment and impact.
The goal is not to conclude that every AI system is biased. It is to know what evidence and safeguards to look for.
How did your definition of AI bias change during the challenge?
What evidence would you want before trusting a claim that an AI system is “fair”?
Who should be involved when organisations test whether an AI system works fairly for different people?
Prioritise original studies, regulators, government sources, court documents or transparent company documentation for real bias cases. An approved AI research tool may help locate sources but should not be treated as the source itself.
