Primary research areas
Facilitating data access while protecting sensitive data is a significant challenge for public interest research. Some privacy-enhancing techniques—such as injecting noise into data or creating ‘synthetic’ datasets—can fundamentally distort data in unknown and potentially harmful ways. For example, rare diseases may be suppressed in synthetic data, or vulnerable communities may be further marginalised.
We study the impact of modern secure data sharing with strong privacy guarantees on reproducibility and research integrity, and build open source tools for the public to understand how their data is being used.
Studying humans or algorithms in isolation reveals little about the real-world implications of data-driven technology. These systems cannot be studied by abstracting out interfaces, interactions, or affordances of real-world deployments. Societal harms often arise at these interaction layers, and thus cannot be identified without accounting for how humans use and interact with a system in practice.
Through user studies and simulations of user-AI interactions, we aim to propose better ways to conduct sociotechnical evaluations.
Automated algorithms set prices for consumers in online markets, match humans in dating apps, and recommend posts on social media feeds. These algorithms affect the prices we pay, the people we date, and the information feeds we consume. However, many of these algorithms are proprietary and opaque, not just to the public but to researchers and regulators as well.
We work to uncover the effects of such algorithms, focusing on how they serve or undermine the public interest. We also develop tools and methods to help users better control how algorithms affect them.
