Santa Fe Institute's video: Patterns and Parsimony

Tyler Millhouse, The University of Arizona The ability to develop good high-level representations is essential for modeling one's environment successfully. This ability is also central to modeling in science, where scientists seek to identify high-level regularities in physical systems (e.g., psychological regularities realized by the human brain). Philosophers, cognitive scientists, and data scientists have, in different ways, proposed a central role for compression in understanding what distinguishes good high-level representations from bad ones. This idea is facially plausible. Good high-level representations are often highly informative despite their simplicity—simplicity they achieve by abstracting away from the fine-grained details of their target systems. Despite its plausibility, I argue that this account of high-level representations is substantially incomplete and that a more complete account requires careful attention to the details of specific models and their relationships to possible representations. Ultimately, I conclude that good high-level representations are those under which the target system exhibits model-relevant patterns. Learn more at https://santafe.edu Follow us on social media: https://twitter.com/sfiscience https://instagram.com/sfiscience https://facebook.com/santafeinstitute https://facebook.com/groups/santafeinstitute https://linkedin.com/company/santafeinstitute