MIT researchers introduce a category-theoretic framework enabling AI systems to revise their own reasoning, moving beyond pattern matching toward genuine