AI Ethics & Limitations
You'll be able to
- Identify common sources of bias in AI
- Explain why models can be confidently wrong
- Reason about responsible AI use
AI is not neutral. A model learns from data created by people, so it can absorb and amplify human biases. If a hiring tool trains mostly on one demographic, it can encode unfair patterns — a serious risk that responsible builders design against.
Models also fail in ways that surprise. A large language model can produce fluent, confident-sounding answers that are simply wrong — a problem called hallucination. Learning to verify output is a core skill for every AI builder.
Key concerns
- Bias — unfair outcomes baked in through unrepresentative data.
- Hallucination — confident, fabricated answers.
- Privacy — models trained on sensitive personal data.
- Transparency — users deserve to know when AI is involved.
Responsible AI is a technical and human discipline: clean data, evaluation against biases, clear labels, and human oversight.
Challenge
Audit a system
Think of an automated decision that could go wrong if biased — e.g. credit scoring. Write down one way it could harm people and one mitigation.
Knowledge Check
Ethics & limitations
What is a hallucination in a language model?
AI bias can enter a system through the data it was trained on.
Answer all questions to submit.