Tag
This paper proposes a lattice-theoretic approach to define unbiased canonical set-valued oracles that remain self-consistent even when their outputs are learned and acted upon, addressing the performativity problem in AI prediction.