Position: Rules Created by Symbolic Systems Cannot Constrain a Learning System

Shih-Wai Lin, Rongwu Xu, Xiaojian Li, Wei Xu

Abstract: This paper explores the inherent flaws of symbolic systems and their limitations in constraining artificial intelligence (AI). Symbols lack intrinsic meaning; their meaning depends on training, contextual confirmation, and social interpretation. *As a product of human cognitive limitations, natural language is a flawed system adapted to human-bounded intelligence.* However, in autonomous learning systems, it exposes deep issues within symbolic systems. For the first time, this paper proposes the *Triangle Problem* framework, revealing the complex relationship between symbols and conceptual spaces. It argues that symbolic systems cannot effectively constrain learning systems, leading to a new type of principal-agent problem. AI deviates from human expectations in areas such as context generation, dynamic adjustment of symbolic meaning, and symbolic jailbreak. By analyzing the ambiguity of natural language, its context dependence, and the differences in AI's perceptual capabilities, this paper calls for the establishment of *Symbolic Safety Science*, which aims to address symbol-related risks in AI development, providing a theoretical foundation for aligning AI with human.

Primary Area: System Risks, Safety, and Government Policy

Keywords: Concept Grounding Problem; Symbol Stickiness; Concept Localization; The Interpretive Authority of Symbols; Symbols; Cognitive Linguistics; Cognitive Science; Symbolic Systems; Symbolic Security; AI Safety; AI Alignment; Symbol Grounding Problem; Principal-Agent Problem; Agents; Symbolic Constraints; Symbolic Representation; Concept Formation in AI; AI Interpretability; Explainable AI (XAI); Philosophy of AI; Philosophy of Mind; Cognitive State Modeling; Deficiencies of Natural Language; Symbolic AI Alignment Problem; Triangle Problem; Value Knowledge; Innate Knowledge; Class-Based Symbolic System; Context; Context Accuracy

Submission Number: 500