AI, data and ethics
Computer Science · Key Stage 3 · The School
Learning from examples
A machine learning model is not programmed with rules; it is shown many examples and finds patterns that predict the answer. Show it thousands of labelled photos and it learns features that distinguish cats from dogs. It does not understand cats — it has found statistical regularities, which is why it can be confidently wrong in ways no human would be.
Bias comes from the data
If a hiring model is trained on a company's past hires, and that company mostly hired one kind of person, the model learns to prefer that kind of person — while looking objective because it is a computer. The bias was in the data, not the code. Fixing it means examining the data and testing outcomes across groups, not just checking the maths.
Impacts, argued with evidence
Automation removes some jobs and creates others; the effects fall unevenly, and "the technology is neutral" is too easy an answer because someone chooses where to deploy it. A good argument names WHO benefits, WHO bears the cost, and what evidence supports the claim — that is what an exam question on impacts is really asking for.
A company trains a CV-screening model on ten years of its own hiring decisions. Those decisions favoured one group. The model learns the pattern — including proxies like which sports a candidate played — and reproduces it, while the developers never wrote a rule about anyone. The maths did nothing wrong. It learned what it was shown, which is why the question to ask about any model is what data it was trained on.