April 3, 2023 | 9:00 AM – 9:30 AM ET | In-person | Washington, D.C.
Privacy Engineering Section Forum
April 4, 2023 | 2:30 PM – 3:45 PM ET | In-person | Washington, D.C.
A Practitioner's Guide to Managing and Mitigating the Privacy Risks of AI
Katharina Koerner, CIPP/US, Principal Researcher, Technology, IAPP
Oliver Patel, CIPP/E, Enterprise AI Governance Lead, AstraZeneca
Organizations use AI and machine learning to inform decisions and automate processes, like medical diagnosis. There are significant privacy risks, because vast amounts of sensitive data are often processed and more data usually means better performing AI. This session explores state-of-the-art privacy engineering techniques which practitioners can employ to mitigate the risks of AI systems, promote data minimization and ensure AI is privacy-preserving by design. For example, differential privacy can prevent an algorithm’s output from revealing information about its training data, whereas federated learning reduces data breach risk by distributing the training of an AI model across different servers. Machine learning models can also be hosted on user devices, and the number of data-points which algorithms evaluate can be reduced, minimizing data sharing and processing. Additional anonymization and pseudonymization techniques will be discussed. However, there are tradeoffs between safeguarding privacy and implementing wider responsible AI principles, like explainability and transparency.
Attendees will learn: