📄️ Ethics and responsible AI
How to audit a machine learning system end to end: bias, fairness metrics, explainability, privacy, documentation, oversight and the EU AI Act.
📄️ 1. What responsible means
Module 1 of the Responsible AI premium course: translating fairness, transparency, accountability, safety and privacy into requirements you can measure.
📄️ 2. Sources of bias
Module 2 of the Responsible AI premium course: selection bias, measurement bias, label bias, feedback loops and proxy variables, illustrated on credit data.
📄️ 3. Fairness metrics
Module 3 of the Responsible AI premium course: demographic parity, equalized odds, calibration, the impossibility theorem, and Fairlearn on the credit scorer.
📄️ 4. Explainability
Module 4 of the Responsible AI premium course: local and global explanations, Shapley values, LIME instability, and why explanation is not justification.
📄️ 5. Privacy
Module 5 of the Responsible AI premium course: re-identification, k-anonymity, differential privacy with epsilon budgets, and membership inference attacks.
📄️ 6. Model cards and datasheets
Module 6 of the Responsible AI premium course: writing a model card and a dataset datasheet with intended use, exclusions and subgroup performance.
📄️ 7. Oversight and recourse
Module 7 of the Responsible AI premium course: human-in-the-loop versus human-on-the-loop, automation bias, right to explanation and a working contestation procedure.
📄️ 8. EU AI Act
Module 8 of the Responsible AI premium course: prohibited practices, high-risk systems, obligations, timeline and extraterritorial reach of the EU AI Act.
📄️ 9. Personal data protection
Module 9 of the Responsible AI premium course: GDPR lawful basis, minimization, purpose, data subject rights, DPIA and Article 22 on automated decisions.
📄️ 10. Case study audit
Module 10 of the Responsible AI premium course: the full audit grid on the credit scorer, findings, corrective actions and what should have been done pre-deployment.
📄️ Recap and exam
Complete recap of the Responsible AI premium course: bias, fairness, explainability, privacy, documentation, oversight, AI Act, GDPR, then the 40-question exam.