Centre for Investing Innovation Guest Lecture: Sahra Ghalebikesabi
31st March 2:00 PM – 3:15 PM BST
Advanced AI assistants combine frontier LLMs and tool access to autonomously perform complex tasks on behalf of users. While the helpfulness of such assistants can increase dramatically with access to user information including emails and documents, this raises privacy concerns about assistants sharing inappropriate information with third parties without user supervision. To steer information-sharing assistants to behave in accordance with privacy expectations, we propose to operationalize contextual integrity (CI), a framework that equates privacy with the appropriate flow of information in a given context. In particular, we design and evaluate a number of strategies to steer assistants’ information-sharing actions to be CI compliant. Our evaluation is based on a novel form filling benchmark composed of human annotations of common webform applications, and it reveals that prompting frontier LLMs to perform CI-based reasoning yields strong results. More details here: https://arxiv.org/pdf/2408.02373
Speaker Biography
Dr Sahra Ghalebikesabi is a Research Scientist at Google DeepMind where she works on privacy-aware LLM-based agents. Her doctoral thesis on tackling privacy and model misspecification within generative modeling was funded by the Microsoft Research PhD fellowship, the ESPRC and Novartis. Previously, she worked with the Health Intelligence team at Microsoft Research Cambridge on robust representation learning, and in the Robust and Verified AI group at Google DeepMind.
Sahra was a Communications Chair at NeurIPS 2022 and 2023, and was previously AISTATS 2022 Publications Chair. She co-organized the MINT workshop at NeurIPS 2024, and the I can’t believe it’s not better – workshop at NeurIPS 2022 and 2023.
