Ethics is easier to understand when you apply it. These six projects turn the course into practice: a bias audit of a model, an explainability analysis, a model card that documents a system honestly, a NIST AI RMF risk assessment, a red-team exercise against a chatbot, and a privacy impact assessment for an AI feature.
Each project is a code-along or walk-through, so you can pause, work through it yourself and adapt it to a system you care about. Pick the project that matches the module you have just finished and come back for the rest. The videos use different tools and datasets, so treat them as examples of the practice and apply the same thinking to your own context.
Train a simple classifier, then measure how its outcomes differ across groups using fairness metrics such as demographic parity and equalised odds, and try a mitigation to see the trade-offs. Pairs with the Bias & Fairness module.
Take a trained model and explain individual predictions and overall behaviour with feature-attribution tools, then judge whether the explanations would satisfy the person affected by the decision. Pairs with Transparency & Explainability.
Pick a realistic AI use case and work through the govern, map, measure and manage functions to identify risks, rate them and propose controls and owners. Pairs with Responsible AI in Practice.
Stress-test a language model with adversarial prompts, such as jailbreaks, prompt injection and harmful-content probes, then log what you find and propose safeguards. Pairs with AI Safety & Alignment and Generative AI Risks.
Assess an AI feature that uses personal data: map the data flows, identify the legal basis and risks, and recommend minimisation and safeguards. Pairs with Privacy & Data Protection and Regulation & Policy.