What Can Complex Systems Theory Tell Us About Understanding in the Human–AI Communication System?

This paper explores communication dynamics between humans and AI, specifically how well and in what ways humans and AI can communicate. In order to evaluate these questions, this paper introduces a statistical measurement called understanding, which is based on Claude Shannon’s concept of information in a communication system. Understanding is calculated using cross-entropy and measures the uncertainty of the receiver about the sender’s message. In order to develop this measurement, human–AI communication is formulated as a complex system, which has certain properties such as hierarchical structure, nonlinearity, complexity, and open vs. closed system dynamics. This formulation helps support the usage of a statistical measurement for evaluating communication efficacy. The paper then explores the connection between understanding and internal language processing systems and determines that understanding is correlated with internal language models.