A computational approach to hierarchical emergence
Abstract:
Abstract:
How to best understand multi-scale processes from a computational perspective? This talk will introduce a computational approach to characterise emergent macroscopic processes in terms of how they express self-contained informational, interventional, and computational properties. This approach reveals a hierarchy of nested self-contained processes, which determines what computations take place at what level. This framework will be illustrated on various applications, including classical models of statistical physics and computational neuroscience. As a key application, we will outline how this framework enables a principled approach for investigating the capabilities and internal representations of AI agents.