Prediction as Truth: When Bayesian Inference Meets History and World Models

> *"True understanding lies not in remembering the past, but in foreseeing the future."*

*An essay from the zhichai.net forum, translated and edited.*

Introduction: The Fog of History

Two historians study the same ancient period. One insists a millennium-old chronicle is reliable history; the other, citing archaeological evidence, points out contradictions. Years of debate—no one is persuaded. When evidence is scarce, belief is more stubborn than logic.

What if we changed the question? Instead of asking "Is this history true?", ask "What does believing this history do for our ability to understand the present and predict the future?" That is the essay's central thesis: re-examining history and world models through a Bayesian lens.

Key points

• World models operate at three levels: personal mental models (everyday decisions), civilizational historical narratives (identity and collective action), and scientific theories (explaining and predicting nature). The lineage runs from Kenneth Craik's *The Nature of Explanation* (1943), through Richard Sutton's Dyna architecture, to Ha & Schmidhuber's *World Models* (arXiv:1803.10122, 2018), summarized as: observe the world (V) + predict it (M) + learn actions inside the internal model (C).
• The brain is a Bayesian prediction engine. Building on Helmholtz's "unconscious inference," Bayes' rule P(H|E) = P(E|H) × P(H) / P(E) describes how priors are updated by evidence into posteriors. Karl Friston's predictive coding / free energy principle holds that the brain minimizes prediction error. Perception is "controlled hallucination": the Kanizsa triangle illusion and blind-spot filling show we perceive the brain's best guess, not reality itself.
• History as prior. Following Maurice Halbwachs, collective memory is socially constructed. Historical evidence is scarce (<1% may be knowable), records are biased, memories are distorted, and interpretations diverge—so debates about historical "truth" often devolve into clashes of position rather than truth-seeking.
• Prediction as the touchstone. Reviving pragmatist epistemology (James, Peirce), the essay argues a world model should be judged functionally: does it explain current phenomena, predict future events, and guide successful action—even if its historical narrative is partly "invented"? A cyclical-history model (warns against complacency in booms; encourages patience in busts) vs. a linear-progressive model (invests in the future, embraces change) is evaluated not by which is truer but which predicts and acts better in the current environment.
• The bootstrapping problem. If a prior is artificially inflated (as conspiracy theories do), Bayesian updating can self-reinforce. Remedies: multi-timescale predictions (years, decades, centuries), counterfactual predictions (what if we acted differently?), and a metacognitive layer—does the model allow itself to be falsified and updated?
• Prediction has limits. Chaotic systems (butterfly effect), emergent phenomena, and free choice make some things unpredictable. Honest Bayesian thinking quantifies uncertainty rather than pretending omniscience. Models also specialize across levels: weather (micro), economics (meso), civilizational dynamics (macro), and meaning-of-life (transcendent)—a mature cognitive system divides labor among them and integrates them.
• Beyond true/false: functional value of history. Even contested narratives serve to bind community identity, provide meaning frameworks, and shape moral intuitions—functions not fully captured by predictive accuracy. Model evaluation should weigh three possibly conflicting criteria: predictive accuracy, functional effectiveness, and richness of meaning. Wisdom lies in balancing them.
• From prediction to action. Friston's active inference: agents act to gather information and test beliefs—hypothesis, experiment, evidence, updated posterior. Civilizations do this via policy pilots, diplomatic probes, R&D investment, and cultural exchange. The loop is: predict → act → observe → update → predict again.

Conclusion: Humility Under Uncertainty

Traditional thinking says "I know the truth"; Bayesian thinking says "I hold beliefs with acknowledged uncertainty—let prediction test them." The framework values openness to challenging evidence, empirical evaluation over positional debate, continuous updating, and pluralism across levels. Even if ultimate truth exists, our knowledge is always approximate. The elegance of the Bayesian view is that it needs no ultimate truth to guide better decisions under uncertainty. Truth is not an endpoint but a process of predicting, verifying, and updating—perhaps the essence of wisdom.

References

1. Craik, K. (1943). *The Nature of Explanation*. Cambridge University Press. 2. Friston, K. (2010). The free-energy principle: a unified brain theory? *Nature Reviews Neuroscience*. 3. Ha, D., & Schmidhuber, J. (2018). World Models. arXiv:1803.10122. 4. Helmholtz, H. von. (1867). *Handbuch der Physiologischen Optik*. 5. Clark, A. (2013). Whatever next? *Behavioral and Brain Sciences*. 6. Halbwachs, M. (1925). *Les Cadres Sociaux de la Mémoire*. 7. Sutton, R. S., & Barto, A. G. (2018). *Reinforcement Learning: An Introduction*. MIT Press. 8. Rao, R. P., & Ballard, D. H. (1999). Predictive coding in the visual cortex. *Nature Neuroscience*. 9. Hohwy, J. (2013). *The Predictive Mind*. Oxford University Press. 10. James, W. (1907). *Pragmatism*.

*Written March 9, 2026, dedicated to those pursuing wisdom amid uncertainty.*