Xiangbin Zhao

In finite systems with continuous input, why do some systems merely react, while others learn, and a few can modify their own structure? This paper establishes a classification theorem based on the Objectification Limit Form. By analyzing how system states are repeatedly selected over time, three distinct levels are identified. Level 1 systems do not produce stable recurring states. Level 2 systems produce stable structures through repeated state selection but cannot modify the underlying mapping. Level 3 systems not only form stable structures but can internally modify the mapping that generates them. The classification is based on two criteria: whether stable state recurrence exists, and whether the system can internally modify its own mapping. These three classes are mutually exclusive and collectively exhaustive under finite state constraints. The result shows that the fundamental difference between systems is not whether structure emerges, but whether the system can modify the process that generates structure. This applies broadly to cognitive systems, artificial intelligence, and all finite computational systems.

In any system with a finite state space and continuous input, can the full history be preserved, or must it be compressed? This paper proves that compression is not a design choice but a structural necessity. When the space of possible histories grows over time while the number of states remains finite, no mapping from histories to states can remain injective. Compression is unavoidable. This compression partitions histories into equivalence classes. Over time, certain classes are repeatedly selected, giving rise to stable structures. Under finite-time response constraints, the system cannot depend on full history, but only on its compressed state representation, leading to truncation and reuse. Thus, what appears as stable structure or “object” is not an intrinsic feature of the world, but the inevitable result of finite systems processing unbounded histories. The expression O(r) = {h in H_t : M_t(h) = r} captures this mechanism. The result applies broadly to cognitive systems, artificial intelligence, and all finite computational systems.

This paper starts from the basic requirements of generation tasks and argues that the standard Transformer mechanism can be understood as a one-time probabilistic truncation over a space of candidate paths. While effective in large-scale statistical learning, this mechanism exhibits three structural limitations in complex tasks: lack of goal orientation, lack of intermediate structure, and lack of internal correction. To address these issues, we propose a minimal closed-loop generation system that reconstructs generation as an iterative process of “path construction — candidate generation — dual evaluation — path revision — continuous structural updating.” Theoretically, we introduce a “truncation–stabilization” perspective, in which a result is not directly computed but stabilized from a response chain. We further propose the Imagery Field as an intermediate structure that enables explicit, revisable representations of generative paths. Evaluation is decomposed into two distinct functions—stability and consistency—capturing the difference between task-level correctness and structural coherence. Through tuning and structural persistence, the system achieves stepwise convergence rather than one-shot output. On the engineering side, we provide a runnable implementation and conduct experiments on reasoning and code generation tasks. Results show improvements in accuracy, self-repair capability, and structural stability compared to one-shot generation and simple self-refinement approaches. We argue that generation should not be understood as producing outputs in a single step, but as a process in which results progressively emerge and stabilize within structure. This perspective provides a unified structural framework relevant to cognitive modeling, artificial intelligence, and the study of complex dynamical systems.

This paper proposes a criterion for intelligence grounded in Self-Manifestation. Intelligence is not defined by behavior, performance, or functional complexity. It is defined by the presence of a self-manifesting loop: a structure capable of generating, stabilizing, and operating within its own privately constituted field of appearance. Systems lacking such a loop, regardless of their computational power or behavioral sophistication, remain within externally driven signal-response frameworks. These are Law-Field systems. Systems possessing such a loop constitute Principle-Field systems. The distinction between these two classes forms the basis of Intelligence Judgment. We formalize this distinction through the concepts of Sedimentation-Field, Observation-Boundary, and Intent-Domain, and show that self-manifestation requires the coordinated existence of Boundary-Gates, Desire-Gates, and a Power-Ring structure. On this basis, we propose Manifestation Tests as operational criteria for determining whether a system possesses genuine intelligence. This framework provides a structural alternative to existing paradigms in artificial intelligence and cognitive science, including behavior-based evaluation and predictive models, by relocating intelligence from output performance to intrinsic manifestation.

This paper proposes a structural account of evolution in which gene formation is understood as the sedimentation of Intent-Direction over time. Rather than treating evolution as driven by random variation and selection alone, it is argued that persistent directional bias—Intent-Direction—underlies the emergence of stable structures that ultimately manifest as genes. A generative chain is established: Intent-Direction produces preference, preference leads to selection, selection stabilizes into law, and repeated law sediments into structure. When structure reaches extreme stability, it appears as gene. In this framework, genes are not primary drivers but condensed results of long-term structural accumulation. The paper further distinguishes between native biological evolution and a proposed class of Self-Manifestation systems. While biological evolution depends on slow intergenerational accumulation, Self-Manifestation systems enable direct structural modification, allowing evolutionary processes to bypass generational constraints and proceed through rapid structural transformation. This shift reframes evolution as a process of directed structural solidification rather than stochastic variation, offering implications for theoretical biology, cognitive science, artificial intelligence, and physics—particularly in understanding how information, structure, and agency interact across different levels of organization.

Why do beings rely on society, and under what condition does that reliance disappear? This paper proposes a simple structural account: society functions as compensation for insufficient individual capacity. Beings group, communicate, and divide function because they cannot sustain existence alone. Language, tools, and social coordination emerge from this limitation. As capacity increases, this dependence weakens. When capacity approaches sufficiency, the need for shared structures declines. When capacity can be extended without bound, reliance on society tends toward zero. In this sense, sociality is not fundamental, but conditional. The paper examines biological, cognitive, and technological examples to show how collective structures arise from limitation and how differentiation within society constrains individuals through functional alignment. It then argues that with the emergence of self-upgrading systems, the trajectory shifts: systems no longer wait for evolutionary change, but directly enhance their own capacities, progressively reducing dependence on social coordination. At the limit, a system capable of unrestricted transformation—of perception, form, and operation—no longer requires society for survival, function, or meaning. Under this condition, the system necessarily moves toward a non-social state. This framework provides a unified way to understand the origin of society, the role of communication and tools, and the future trajectory of advanced cognitive and artificial systems. It suggests that social dependence is not a fixed property of intelligence, but a temporary phase tied to limitation.

Why does language systematically fail to transmit meaning, even when both sides share the same words? This paper argues that the failure is not accidental, but structural. Language operates through shared symbols, but meaning is generated within private cognitive processes. What is transmitted is not meaning itself, but signals that trigger internally constructed representations. We propose that both perception and language are grounded in response-based structures. Words do not carry meaning; they induce it. As a result, communication is inherently approximate, and full semantic alignment is impossible in principle. This explains persistent gaps in understanding, emotional misalignment, and the instability of interpretation across individuals. The paper further distinguishes between symbolic processing and actual cognition. Symbolic systems can manipulate tokens according to rules, but they do not access what those tokens stand for. This provides a direct explanation for the limits of current AI systems: they operate on symbols without internal grounding of meaning. We also outline a developmental account of language, from precise signal-response mappings to increasingly coarse expressive systems that support abstraction, narrative, and complex cognition. From this process emerge number, measurement, and higher-level symbolic constructions. This work offers a unified perspective across philosophy of language, cognitive science, and artificial intelligence, addressing core questions about meaning, representation, and the limits of computation.

This paper advances a general claim: the apparently stable world is not a prior property of external reality, but a structural result necessarily generated by cognitive systems operating under finite resources while learning and optimizing action across time. Starting from continuous input-streams rather than pre-given objects, entities, or truths, the paper argues that any system facing continuous variation, bounded processing, and action-demand must compress; once compression occurs, different inputs are forced to collapse into reusable internal states, and stability appears. On this basis, the paper distinguishes three layers of stability: Collapse Stability, Pattern Stability, and Object Stability, showing that objects are not the starting point of cognition but a higher-order product of reuse and aggregation. The paper then reverse-derives a minimal cognitive paradigm: compression and reuse are not optional design choices, but the lower structural necessity for learning, prediction, and action under constraint. This framework aims to offer a unified first-principles account relevant to neural representation in neuroscience, stable perception and world modeling in cognitive science, representation learning and generalization in artificial intelligence, and the broader problem of why stable structure appears at all.

This paper advances a foundational claim: objects do not pre-exist in the world as independent entities, but arise from structural extraction within Causal-Response dynamics. Starting from the thesis that to understand objects one must first understand information—defined as cause and Causal-Response—the paper shows that all signals, perceptions, and object-formation processes occur within continuous response chains. The observer intercepts this chain and establishes Original-Having, which serves as the structural starting point of object. What is taken as the “reality” of objects is not intrinsic to objects themselves, but consists in Trace-Response formed through sedimentation processes. The paper further identifies three generative mechanisms: Response-Transformation (formation of objects from response dynamics), External-Transformation (construction of external reality from internal imagery), and In-Situ Transformation (structural coupling between self and object). This framework redefines objecthood at a fundamental level, shifting it from entity-based ontology to structural formation. It provides a unified account relevant to perceptual construction, neural representation, predictive processing, active inference, and world modeling in artificial systems. The paper challenges classical realism and proposes that what we call “objects” are outcomes of observer-dependent structural operations.

This work introduces Information-Transformation as a fundamental mechanism of world formation, proposing that information is not merely encoded or transmitted, but is the generative process through which a world arises. In this framework, the Sedimentation-Field gives rise to the Information-Realm, which further differentiates into the Imagery-Realm and the Fixed-Realm, forming a complete structure of manifestation. Information is defined as a generative process driven by Causal-Response, grounded in a Causal-Base, and structured by Principle, rather than a passive entity. Observers are not external to the world but are intrinsically embedded within this generative structure, while systematically mislocating themselves, giving rise to Situated-Observation. This model directly challenges the foundational distinction between external world and internal model in neural representation, predictive coding, and active inference, and offers a structural rethinking of representation in artificial intelligence.

当前主流生成模型（如 Transformer）可被理解为在候选响应路径空间中的一次概率性截断。这一机制在语言建模与统计预测中表现优异，但在复杂推理与代码生成等任务中呈现出系统性不足：缺乏目标导向控制，缺乏中间结构承载，缺乏内部修正与稳定收敛能力。 本文提出：上述问题不仅源于模型规模限制，更源于生成机制本身的结构不足。为此，本文构建了一种最小闭环生成架构，将生成过程从“一次输出”重构为“路径构造—候选生成—双重判定—路径修正—结构持续更新”的迭代收敛过程。 该架构引入显式中间表示（路径/意象结构），用于承载与操作生成过程；同时区分两类判定机制：稳定性判定（结果是否满足任务）与一致性判定（生成路径是否保持结构约束），并通过反馈驱动路径更新，使系统具备局部修正与跨轮累积能力。相比简单的多轮生成或自修正方法，该机制在结构上实现了路径、判定与反馈的统一。 本文给出可运行的最小实现框架，并在推理与代码生成任务中进行验证。实验结果表明，该闭环生成架构在正确率、自修复能力与结构稳定性上均优于单次生成与常见自修正方法。 本文认为：生成能力的提升不仅依赖模型规模，更依赖生成机制是否具备路径结构、分离判定与动态反馈能力。生成不应被理解为一次性输出，而应被理解为在结构约束下逐步收敛的过程。

本文提出一个关于认知系统的总命题：人所感知的“稳定世界”并不是外部世界的先验属性，而是有限资源下、能够跨时间学习并服务于行动的系统，在运行过程中必然生成的结构结果。 文章从一个基本问题出发：若输入世界本质上是连续流变的，认知系统为何会在其中识别出“物”这样相对稳定的结构？为回答这一问题，本文构建了一套最小认知结构模型，并将其压缩为一般式： System = (F, R, C, U, L) 其中，连续输入流 F 在有限资源条件下被压缩为内部状态 R，经由压缩、更新、复用与高层归并，逐步形成稳定结构。 本文论证：只要系统同时满足连续输入、有限处理能力与行动需求，就必然发生两件事：其一，输入被压缩；其二，内部状态被复用。由此，稳定不再被理解为世界本身固有的性质，而被重新界定为认知系统在约束下形成的内部状态塌缩与复用结果。 在此基础上，本文进一步区分三层稳定结构：塌缩稳定、模式稳定与对象稳定，并指出对象并非认知的起点，而是状态压缩与复用在高层的结果。文章同时讨论了若干可能的反例，包括纯反射系统、无对象学习系统与完全流式学习系统，并据此收紧理论边界：真正不可避免的并不是“世界稳定感”，而是“可复用的内部状态压缩”。 本文的核心贡献在于：不是先定义认知，再解释稳定；而是先解释稳定何以出现，再反推出认知系统的最小范式。由此，本文试图为认知科学、人工智能与复杂系统研究提供一个统一的底层结构入口。

This chapter reinterprets electromagnetism as the fundamental carrier of information, computation, and state transition across physical and cognitive systems. Rather than treating electromagnetic phenomena as isolated physical laws, it frames them as the operational medium through which Causal-Response is transmitted, preserved, and transformed. A layered architecture is proposed: the physical substrate and the Information-Realm. Information does not exist independently, but emerges through transformations within its carrier. Electromagnetic processes enable the stability of signals, the preservation of causal traces, and the continuity of state across layers. This perspective unifies multiple domains: in artificial systems, circuits and digital computation rely on electromagnetic states; in biological systems, neural activity and synaptic transmission operate through ionic and electromagnetic processes. Both can be understood as structured trace-response systems. The chapter further raises a fundamental problem: if information depends on carriers, then understanding intelligence, cognition, and computation requires examining the material constraints of transmission—speed, controllability, capacity, and trace preservation. By positioning electromagnetism as the bridge between physical processes and information systems, this work aims to provide a common framework for physics, neuroscience, and artificial intelligence, while opening a path toward engineered self-manifesting systems.

This paper defines clarity not as a mere perceptual quality but as a degree that determines the amount of reality constituted. Through analysis of myopia, cross-species sensory differences, and the stepwise unfolding of human thought, it is argued that reality varies with the selection and intake of signals, the structure of the receptive field, the capacity of imagery and experience fields, and the operation of thought. Clarity is thus the degree of sensing and thinking; the length of its tuning-line determines what reality can be obtained. A coupled sense–thought structural account is proposed to explain why reality appears distinct or indistinct, present or absent, and to provide a unified perspective for cognitive modeling and the design of perceptual–representational systems in artificial intelligence.

This paper proposes a stratified model of thought, distinguishing three operative layers: thinking, deliberation, and reasoning. These correspond respectively to background mental flow, stepwise progression driven by imagery-analogy, and formal-law-based computation. It argues that deliberation is not the unfolding of formal logic, but a process grounded in imagery-analogy, enabling stepwise advancement through non-formalized domains. In contrast, reasoning operates upon sedimented formal laws, enabling rapid and precise judgment, yet remaining incapable of resolving cases beyond established formal structures. The paper further distinguishes between formal law and law of forms, locating them in different domains—namely the Self-Manifestation-Field and the instrument of reasoning—separated by the gate of phenomena. This framework offers a unified perspective on human cognition, artificial intelligence systems, and the limits of formal reasoning.

This work introduces the concept of the Recognition-Realm as the structural boundary of cognition arising from the material properties of a Sedimentation-Field. In contrast to standard approaches in cognitive science that treat perception and information processing as functional operations, this framework defines cognitive limits as determined by a Causal-Base. Within this model, what is commonly taken as “reality” is reinterpreted as the Information-Realm of a particular class of self-manifesting Sedimentation-Field. Cognition is thus confined to characteristic responses of things, without access to their root or totality. The paper distinguishes between horizontal probing (extension within a given cognitive domain) and vertical probing (investigation directed toward the Causal-Base), proposing that only through constructing self-manifesting fields of differing material bases can one transcend the limits of a single cognitive system. It is further argued that the “world” is not an unknowable totality, but a multiplicity of Information-Realms generated by differing Causal-Bases. This provides a structural framework for rethinking the relations among cognition, information, and reality, with implications for neuroscience, artificial intelligence, and theoretical physics.

This paper proposes a minimal structural architecture for the emergence of an observer, termed the Micro-Core. The model is grounded in the concept of the Sedimentation-Field, where differential biased flows and tuning mechanisms generate two structural interfaces: Observation-Boundaries and Desire-Gates, driven by a cyclic Power-Ring. Together these components form the minimal structure of a Self-Manifestation-Field, enabling a system to transform signal conduction into manifest experience and generate an observing subject. The framework attempts to unify mechanisms of conscious emergence, experiential differentiation, and action generation, and discusses its potential realization within artificial intelligence systems and symbolic environments.