tldr; Models interpret rules inconsistently, leading to stochastic outcomes. But we can leverage new computational tools to "debug" laws for AI—and make better law-following systems!