Professor of Electrical Engineering and Computer Science, Computer Science and Artificial Intelligence Laboratory, MIT

Abstract: Data privacy presents a multi-pronged challenge in the age of AI. We are seeing an exponential growth in the volume of personal data being generated and collected, ranging from social media interactions to financial transactions, and, unfortunately, a corresponding growth in the number of cyberattacks and data breaches, as hackers proliferate and become increasingly sophisticated. We need to balance the benefits of data-driven innovations with the need to safeguard individual privacy rights, as many services, ranging from e-commerce to healthcare to energy to government, rely on extensive data analysis for their functionality. We present a new privacy definition, termed Probably Approximately Correct (PAC) Privacy, which holds promise in tackling difficult privacy challenges in statistical data processing.