人工智能时代的信息物理学与智能信息演化

Information Physics and Intelligent Information Evolution in the Age of Artificial Intelligence

摘要: 信息物理学关注信息与物理状态、计算过程和智能行为之间的关系。信息这一概念经历了从经典物理中的状态描述、Shannon信息论中的统计表征，到人工智能时代语义表示的发展过程。本文从信息物理学视角出发，讨论信息产生、表示、传输和利用过程中物理规律与智能信息处理之间的关系，重点关注信息如何由物理状态经过观测和数据处理形成语义表示，并进一步服务于具体智能任务。以自动驾驶感知决策系统为贯穿案例，本文系统讨论信息从物理状态、统计表征到语义信息的演化过程——从传感器获取的物理信号，到深度学习识别和表示的行人、车辆等语义对象，再到驾驶决策。在此基础上，分析人工智能在信息表示、知识形成和生成式处理中的作用，以语义通信为例讨论信息如何从物理观测经智能表示服务于任务决策，并探讨信息价值描述以及人工智能与物理规律融合等方向。分析表明，人工智能时代的信息科学正在由“描述信息”逐步扩展到“理解信息”和“利用信息”，而信息物理学视角强调信息演化过程受到物理观测条件、计算实现和任务需求的共同影响。

Abstract: Information physics focuses on the relationship among information, physical states, computational processes, and intelligent behaviors. The concept of information has evolved from the description of physical states in classical physics, to the statistical representation in Shannon information theory, and further to semantic representations in the era of artificial intelligence. From the perspective of information physics, this paper discusses the relationships between physical laws and intelligent information processing during the generation, representation, transmission, and utilization of information. Particular attention is given to how information evolves from physical states into semantic representations through observation and data processing, and further supports specific intelligent tasks. Taking an autonomous driving perception-decision system as a representative case, this paper systematically analyzes the evolution of information from physical states and statistical representations to semantic information, covering the process from physical signals acquired by sensors, to semantic objects such as pedestrians and vehicles recognized and represented by deep learning models, and ultimately to driving decisions. Furthermore, the roles of artificial intelligence in information representation, knowledge formation, and generative processing are analyzed. Using semantic communication as an example, this paper discusses how information is transformed from physical observations into intelligent representations to support task-oriented decision-making, and explores future directions including information value characterization and the integration of artificial intelligence with physical laws. The analysis indicates that information science in the era of artificial intelligence is gradually evolving from “describing information” toward “understanding information” and “utilizing information”. From the perspective of information physics, the evolution of information is jointly influenced by physical observation conditions, computational implementations, and task requirements.