“How do humans make sense?” multiscale dynamics and emergent meaning

Abstract

Abstract The challenges posed by the composite nature of sense-making encourage us to study how that composite is dynamically assembled. In this paper, we consider the computational underpinnings that drive the composite nature of interaction. We look to the dynamic properties of recurrent neural networks. What kind of dynamic system inherently integrates multiple signals across different levels and modalities? We argue below that three fundamental properties are needed: dynamic memory, timescale integration, and multimodal integration. We argue that a growing area of investigation in neural networks, reservoir computing, has all these properties (Jaeger, 2001). A simple version of this model is then created to demonstrate “emergent meaning,” using a simplified model communication system.