AI Ethics & Safety
Bias, hallucination, privacy — and how to use AI responsibly.
- ✓Why models inherit bias from their training data
- ✓What 'hallucination' is and how to guard against it
- ✓Privacy basics when working with AI
Explanation
AI learns from human data, so it can absorb and amplify human bias — for example, favouring one group if the training data was skewed.
Language models also hallucinate: they can produce fluent, confident answers that are simply wrong, because they predict plausible text rather than look up facts.
The practical response is to verify important outputs, keep a human in the loop for high-stakes decisions, and protect privacy by never pasting sensitive personal data into tools you do not control. Responsible AI is less about fear and more about good habits.
A résumé-screening model once down-ranked candidates from certain backgrounds because it learned from biased historical hiring data — a textbook example of inherited bias.
- • Treating confident AI output as automatically true.
- • Pasting private or sensitive data into tools without checking their policy.
Find one AI answer online, fact-check it against a reliable source, and note whether it held up.
1. Why can an AI model be biased?
2. A 'hallucination' in an LLM is...