Lijun Chen

Hume’s problem of causality contains two nested core dilemmas: the legitimacy dilemma of inductive logic, and the existential dilemma of causal validity. For two hundred years, mainstream solutions have failed to respond to both dilemmas simultaneously. This paper abandons the metaphysical presupposition that causality must possess inherent ontological necessity, and constructs an alternative cognitive‑genetic framework based on the structural isomorphism between the classical Chinese meta‑cognitive structure of “image abstraction and analogical categorization” (quxiang bilei) and modern Bayesian inference. It is demonstrated that Bayes’ theorem is the mathematical formalization of the cognitive loop of “image abstraction → analogical categorization → verification → feedback”. Causality is redefined as the limit cognitive state in which, after massive collective Bayesian updating over hundreds of millions of years of embodied survival practice, the posterior probability converges infinitely to 1. Hume’s double dilemma is thereby resolved: the logical gap of induction is dissolved, and the source of causal validity is explained by natural selection driven by survival costs. The paper further provides an epistemological foundation for Pearl’s ladder of causation, interprets scientific revolutions as paradigm shifts from low‑dimensional to high‑dimensional cognitive convergence, diagnoses the lack of an embodied verification loop as the fundamental bottleneck of AI causal reasoning, and establishes the stance of cognitive humility: all causal judgments are tentative optimal estimates under the current evidence set, always open to new evidence, rejecting both metaphysical dogmatism and relativistic nihilism.

This paper demonstrates, from an epistemological standpoint, the structural isomorphism between Quxiang Bilei—the classical Chinese cognitive method of image abstraction and analogical reasoning, literally "taking phenomena to form analogies and categorical associations"—and Bayesian inference. By systematically mapping the three components of Bayes's theorem—prior probability, likelihood function, and posterior probability—onto the three constitutive stages of Quxiang Bilei: image-taking, analogy-making, and verification, this paper reveals that Bayesian inference is not merely a specialized technical toolset, but rather the mathematical formalization of the metacognitive structure inherent in Quxiang Bilei. Building upon this isomorphic framework, the paper advances five core philosophical propositions: (1) objectivity is the statistical convergence outcome of collective Bayesian updating across epistemic communities; (2) causal laws represent the limiting case in which posterior probability infinitely approaches unity; (3) the interpretability of probabilistic calculations resides not within mathematics itself, but in its correspondence with underlying metacognitive structures; (4) the absolute boundary of artificial intelligence lies not in insufficient computational power, but in its lack of embodied physical anchoring and authentic survival feedback; and (5) there exists a fundamental, non-accidental isomorphism between Quxiang Bilei and Bayesian inference. These five propositions are not mere mutual projections between two disparate conceptual frameworks, but genuinely novel insights emerging from their productive dialogue and critical engagement. This work provides a robust epistemological foundation for Bayesianism, offers rigorous mathematical support for the modern transformation of classical Chinese epistemology, and establishes a unified analytical framework for understanding the cognitive essence and fundamental limitations of artificial intelligence.