The paper proposes autoformalization—automatically converting informal natural language mathematics into formal, machine-verifiable statements—as a promising approach to achieving AI with human-level general reasoning and significant impacts on mathematics, programming, and AI systems.