Modern AI is built on data, and much of that data is about people: what they search for, where they go, what they write and how they look. Privacy is a person’s ability to control how information about them is collected and used. Data protection is the set of legal and technical rules that put this into practice, such as the UK and EU GDPR, which set out principles like lawfulness, purpose limitation, data minimisation and security, and give individuals rights over their data.
The lessons begin with the basics: what counts as personal data, the main data protection principles and the rights of individuals. They then turn to privacy techniques designed for machine learning. Differential privacy adds carefully measured noise so that results reveal little about any single person, and federated learning trains a shared model without moving raw data to a central server.
The final lesson looks at a risk specific to AI: large models can memorise parts of their training data, and attackers can sometimes extract it. Technical protections such as differential privacy help, but they involve trade-offs in accuracy and do not replace good practice like collecting less data in the first place. Watch the lessons in order, then continue on to Accountability & Governance.
A short animated explainer from the UC Berkeley Center for Long-Term Cybersecurity on differential privacy as an approach to privacy-preserving data analysis.
Video 4minExplains how models are trained on private data in many places and only the model updates are combined on a central server, so the raw data is never shared.
Video 7minAsks whether differential privacy helps with memorisation in deep learning models, reviews research on memorisation attacks and gives practical tips. A short but more technical, optional watch.
Video 14min