Johns Hopkins University hosts AI safety-relevant research led by Prof. Anqi (Angie) Liu, whose group focuses on machine learning for trustworthy AI, including distributionally robust learning and uncertainty quantification under distribution shift.
Johns Hopkins University hosts AI safety-relevant research led by Prof. Anqi (Angie) Liu, whose group focuses on machine learning for trustworthy AI, including distributionally robust learning and uncertainty quantification under distribution shift.
People– no linked people
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Johns Hopkins University's connection to AI safety grantmaking primarily runs through Prof. Anqi (Angie) Liu, an Assistant Professor of Computer Science at the Whiting School of Engineering. Her research group works on machine learning for trustworthy AI, developing algorithms that address three central challenges: making models robust to changing data distributions, providing accurate and honest uncertainty estimates, and incorporating human preferences and values into AI decision-making. Liu completed her PhD at the University of Illinois Chicago and her postdoctoral research at Caltech before joining Johns Hopkins. Liu's key research projects include distributionally robust learning under covariate shift, uncertainty quantification for AI safety and fairness, conformal prediction methods for safety assessment, off-policy reinforcement learning, and fair machine learning. Her work on JAWS and JAWS-X introduces distribution-free wrapper methods for uncertainty quantification that balance sample efficiency and computational tractability. She also works on calibrated uncertainty for large language models, including cross-lingual prediction calibration. Liu is affiliated with multiple Johns Hopkins research centers: the Mathematical Institute for Data Science (MINDS), the Institute for Assured Autonomy (IAA), the Center for Language and Speech Processing (CLSP), and the Laboratory for Computational Sensing and Robotics (LCSR). She is listed as a faculty member in the AI Existential Safety Community at the Future of Life Institute. Open Philanthropy recommended a grant of $94,600 to Johns Hopkins University in 2022 to support course buyouts enabling Prof. Liu to explore and prepare for AI safety research. She has also received an Amazon Research Award, a Johns Hopkins University and Amazon Initiative for AI Faculty Research Award, a Johns Hopkins Discovery Award, and an Institute for Assured Autonomy Challenge Grant. As of early 2026, the group includes approximately eight PhD students, several co-advised with other Hopkins faculty in related areas including robotics, NLP, and biostatistics.
Theory of Change
Updated 05/18/26Liu's approach holds that near-term and long-term AI risks are substantially driven by AI systems that are brittle under distribution shift, poorly calibrated in their uncertainty, and misaligned with human values. By developing rigorous mathematical methods for robust learning, uncertainty quantification, and conformal prediction, the group aims to give practitioners and researchers the tools to build AI systems that fail safely, honestly represent their own limitations, and remain reliable in deployment environments that differ from training conditions. The work primarily targets the technical foundations of trustworthy AI rather than policy, with the belief that principled ML algorithms are a prerequisite for safe, high-stakes AI deployment.
Grants Received
Updated 05/18/26Projects– no linked projects
Updated 05/18/26Discussion
No comments yet. Be the first to share your thoughts.