Across information theory, machine learning and AI, understanding is often identified with compression – echoing Gregory Chaitin’s claim that “comprehension is compression.” Philosophers, however, describe understanding as grasping connections, offering explanations and handling novelty. This paper unifies the two pictures through three interlocking theses. First, understanding acts as an efficient proxy for a robust competence, allowing us to decide whom to trust and from whom to learn. Second, to understand a domain is to hold a mental model of its relational structure that supports prediction; prediction itself yields compression because predictable information need not be stored separately. Thus compression is not identical to comprehension but its representational shadow. Third, the uniquely human form of understanding is shaped by fiduciary and transmission demands, which pressure human cognition toward principled simplicity, ensuring demonstrability and transmissibility. The resulting framework accounts for both the appeal and the limits of compressionist accounts of understanding and sheds light on the inscrutability of current AI systems.
Review