The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) AI and the Responsible Data Economy with Dawn Song - #403
Aug 24, 2020
Dawn Song, a Professor of Computer Science at UC Berkeley and CEO of Oasis Labs, dives into the future of data privacy and security. She discusses creating a responsible data economy that empowers users while leveraging techniques like blockchain and differential privacy. The conversation highlights the risks of adversarial attacks on AI, and how innovations in homomorphic encryption can secure sensitive data. Dawn also shares insights on balancing privacy with effective COVID-19 contact tracing, emphasizing the importance of technology in protecting personal information.
AI Snips
Chapters
Transcript
Episode notes
Responsible Data Economy Challenges
- Users lose data control and don't get sufficient direct benefits.
- Businesses face data breaches, privacy regulation compliance costs, and data silo challenges.
Controlled Data Use
- Responsible data usage requires blockchain for logging data rights and usage, and secure computing for controlled data access.
- This "controlled use" restricts data buyers to confined environments, preventing raw data access and misuse.
Language Model Data Leakage
- Language models trained on email datasets can leak sensitive info like social security and credit card numbers.
- Attackers can exploit this vulnerability by querying the model without needing model parameters.