← Members
Zhonghao works on Martingale training as a principled method to address Bayesian irrationality — confirmation bias, sycophancy, and polarization — and on alignment algorithms that seek human reflective equilibrium. He develops formal modeling, training, evaluation, and computational social science methods to study and mitigate AI's influence on human epistemics.
He is currently a Cosmos Fellow and Research Associate at the HAI Lab, University of Oxford, working with Prof. Philipp Koralus (Philosophy) and Prof. Jakob Foerster (Engineering). He holds an MSt in AI Ethics and Society from the University of Cambridge.