A Unifying Perspective on Probabilities as Model Predictions

TEXT START: Although probabilistic statements are ubiquitous, foundational disagreements persist about their understanding, as exemplified by debates between Bayesians and frequentists; moreover, it is unclear when and why acting on them actually leads to desirable outcomes.

The Dissection

The paper relocates probability from metaphysical “truth” to a pipeline: abstraction, prediction, calibration, and action. Its unification is operational rather than philosophical. It dissolves the Bayesian–frequentist dispute by arguing that every probability is produced by a prediction method, then tries to establish when calibrated predictions can support useful decisions.

The Core Fallacy

The abstract risks treating finite calibration as a bridge from predictive reliability to successful decision-making. Calibration can improve decisions under stable distributions and specified utilities. It does not guarantee success when environments shift, agents adapt strategically, objectives conflict, or institutions collapse.

More importantly, predictive quality is not productive power. A model may forecast outcomes accurately while its owner captures the resulting value and everyone else becomes merely an object of optimization. The paper addresses how probabilities are constructed and used; it does not address who owns the prediction machinery, controls the policy channel, or receives the surplus.

Hidden Assumptions

• The abstractions remain adequate when the underlying system changes.
• Finite calibration generalizes to future events rather than merely describing historical regularities.
• Utilities are measurable, stable, and accepted by the relevant decision-makers.
• The targets of prediction are not actively adapting to the model.
• Data, compute, and institutional enforcement remain available.
• “Desirable outcomes” can be specified without resolving conflicts over power and distribution.
• Finite-event guarantees matter in open-ended, regime-changing environments.

Under the Discontinuity Thesis, the most dangerous assumption is that improved prediction preserves broad human participation. It does not. AI can make decisions more accurate while simultaneously eliminating the labor required to produce, interpret, and execute them.

Social Function

Classification: partial truth, prestige signaling, and transition management.

The paper correctly strips probability of false objectivity and emphasizes model dependence. That is useful. But its practical framing channels attention toward calibration and decision quality while leaving ownership, control, and class displacement outside the frame. It refines the dashboard while ignoring who owns the vehicle and who is being thrown from it.

The Verdict

This is a legitimate clarification of probability, not a defense of the existing economic order. Its framework can make automated governance more coherent and more powerful. Under DT logic, that strengthens Sovereigns and weakens everyone whose value depends on routine cognitive judgment. Better-calibrated models do not restore productive participation; they accelerate its replacement. The paper sharpens the instruments of the transition while offering no mechanism that prevents the death of the wage–consumption circuit.