Introduction: From Historical Narrative to Predictive Optimization
Traditional research on civilizational history has long been trapped in a methodological deadlock: heterogeneous, often contradictory sources make the pursuit of "objective truth" an endless cognitive labor. Cases such as Homer's oral transmission of events spanning centuries around the Trojan War, or cross-regional archaeological anomalies that clash with mainstream historical narratives, illustrate a deeper problem. Historical records are not mirror-like reflections of the past but probabilistic constructions filtered through multiple mediations.
The essay argues for a fundamental paradigm shift: relocate historical records to the status of Bayesian priors, and evaluate civilizational world models by their predictive validity rather than by truth-versus-falsity verdicts.
1. Core Thesis: From the Truth Controversy to Bayesian Updating
1.1 Limits of Traditional Truth-Or-False Debates
Oral traditions undergo systematic drift across generations: core plots may survive, but timelines, geography, and interpersonal relations are reshaped. The Homeric epics, for instance, span roughly four centuries from the Trojan War (~12th century BCE) to their final written form (~8th–7th century BCE). When material remains conflict with textual records, conventional source criticism imposes a hierarchy, but that hierarchy itself embeds specific epistemological presuppositions.1.2 Introducing Bayesian Epistemology
The Bayesian framework reframes historical inquiry from "judging truth" to "optimizing the model." Its philosophical basis is pragmatic epistemology: we can never access "history itself" directly, but only assess the quality of belief systems indirectly through their predictive performance.Bayesian mapping for historical cognition:
| Bayesian concept | Historical counterpart | Role | |---|---|---| | Prior | Historical accounts, civilizational narratives, traditional knowledge | Provides initial belief distribution, carries cultural memory | | Likelihood | Present-day understanding, predictive models | Transforms priors into testable predictive propositions | | Evidence | Contemporary observations, archaeological finds, new documents | Reference standard for predictive testing | | Posterior | Updated world model | Optimized belief state integrating prior and evidence |
The maxim that "historical records, being partly true and partly false, can only serve as priors" captures the core of Bayesian epistemology: the probabilistic nature of priors is naturally compatible with the uncertainty of historical records.
2. Theoretical Foundations: Multi-Disciplinary Corroboration
2.1 The Bayesian Nature of Human Knowledge
The entire development of human knowledge can be read as a Bayesian process. Researchers always start from some prior, update beliefs when facing new evidence, and the resulting posterior becomes the next round's prior.Case study: the trajectory of the Copernican revolution. Copernicus retained many traditional elements — circular orbits, uniform motion, deferent-epicycle structures — as strong priors that limited the radicalism of the new model while preserving predictive continuity. Kepler's elliptical-orbit corrections and Galileo's telescopic observations incrementally raised the heliocentric model's likelihood score, and Newtonian mechanics ultimately supplied the dynamical foundation that pushed its posterior probability past geocentrism.
Unlike Popperian falsificationism, Bayesian updating allows "erroneous" theories to retain partial informational value during correction — as long as their predictions outperform random guessing, they still contribute to the posterior distribution.
2.2 The "Bayesian Brain" Hypothesis from Neuroscience
Contemporary neuroscience supplies a biological foundation. The Bayesian brain hypothesis reconceives the brain as a perpetual prediction machine: higher-level cortex generates predictions about the world based on prior knowledge, lower-level sensory regions compute the error between prediction and actual input, and error signals are fed back upward for model revision.> "The brain is not a camera passively recording external signals, but constantly interrogating those signals with its rich prior knowledge to produce a 'best guess' of sensory data." — frontier neuroscience theory
Perception is therefore a "constrained hallucination" — the brain's optimal explanation balancing priors and sensory evidence. Classical illusions such as the spontaneous reversal of the Necker cube are necessary products of Bayesian inference.
2.3 Bayesian Implementation in Artificial Intelligence
Large language models (LLMs) — whose core training objective is next-token prediction — implement approximate Bayesian updating at scale. Given prior context, the model predicts a probability distribution over the next token; comparison with the actual token (evidence) updates the internal representation (posterior).The Bayesian framework therefore reveals a deep isomorphism between human and machine cognition. Carbon-based neural dynamics and silicon-based gradient descent both functionally approximate ideal Bayesian inference — a convergence rooted in shared physics and shared mathematics.
3. Bayesian Reconstruction of the Civilizational World Model
3.1 Layered Structure: Nested Individual and Collective Models
The proposition that "every person carries a world model in their brain" acquires precise technical content under the Bayesian-brain framework. Individual world models are probability distributions encoded in neural systems that describe belief states about external structure.| Individual world model | Civilizational world model | |---|---| | Predictive function: anticipations of future sensory input | Historical narrative: shared framework of collective memory | | Learning function: parameter updates from prediction error | Cultural paradigm: default assumptions about how the world works | | Biological endowment: evolutionary priors | Institutional framework: normative structure of action | | Cultural inheritance: injection of collective wisdom | Dual temporality: both past belief and future framework |
The claim that "every civilization has its own historical narrative, which is the civilizational world model" reveals the socially constructed dimension of collective cognition.
3.2 Treating Historical Records as Priors
The core operation converts historical records from "factual statements" into probabilistic beliefs. This shift acknowledges the complex causal processes that generate historical sources, redirects historiographical critique from truth-judgment to confidence assessment, and provides logical space for the coexistence of contradictory records.Prior-strength gradation:
| Layer | Typical features | Prior distribution | Updating elasticity | |---|---|---|---| | Core consensus | Widely embedded, multiply supported | Highly concentrated (P≈1) | Very low | | Marginal controversy | Significant disagreement, active debate | Dispersed, multimodal | High | | Anomalous records | Severe conflict, isolated existence | Very low but non-zero (P≈0) | Conditional |
3.3 Predictive Testing as Model-Selection Mechanism
Predictive accuracy can be measured along multiple dimensions: short-term/long-term, quantitative/qualitative, domain-specific.| Predictive domain | Comparative performance | |---|---| | Natural phenomena | Physics-based frameworks clearly superior | | Human behavior | Highly context-dependent | | Social evolution | Complex adaptive dynamics |
4. Operational Framework: Continuous Optimization of World Models
4.1 Implementation Steps for Bayesian Updating
Operationalizing a civilizational world model in probabilistic form requires work on three layers: conceptual, mathematical, and computational.- Prior specification — aggregation of expert judgment (Delphi method), frequency-based conversion of historical data, elicitation of implicit priors in models, selection of probability distribution families.
- Evidence acquisition — systematic integration of archaeological discoveries, structured analysis of textual records, real-time incorporation of contemporary observations, evidence-quality evaluation systems.
- Likelihood evaluation — predictive scoring of candidate models against assembled evidence, calibration against domain-specific benchmarks.
- Posterior updating — Bayesian combination of priors and likelihoods, uncertainty propagation, scenario generation.
- Historical records should be treated as Bayesian priors, not as truth-claims to be verified.
- Predictive accuracy, not veridical correspondence, becomes the primary criterion for evaluating civilizational world models.
- Three independent pillars support the framework: scientific revolutions as Bayesian updating, the Bayesian-brain hypothesis from neuroscience, and next-token prediction in LLMs as approximate Bayesian inference.
- Individual world models (encoded as neural probability distributions) are nested within civilizational world models (shared narratives, paradigms, and institutions).
- A graduated prior-strength scheme accommodates core consensus, marginal controversy, and anomalous records within a single formal structure.
- Operational pipelines require prior specification, evidence acquisition, likelihood evaluation, and posterior updating, with robustness safeguards and cross-civilizational triangulation.
- The framework is complementary to, not a replacement for, humanistic interpretation; boundary conditions include path dependence, singular events, and narrative irreducibility.