Concept study / Information & adaptationConceptual illustration. Shapes, colors, and motion are not measured data.
A prediction is useful when it can be compared with an outcome. In predictive-processing models, a system uses an internal model to anticipate inputs and updates relevant states when observations differ. Rao and Ballard developed a computational account of aspects of visual cortical responses using predictive coding. Friston’s free-energy framework offers a broader theoretical language connecting inference, action, and learning. These approaches are relevant to Hill’s emphasis on retained information shaping future behavior. Their existence does not show that a planet or universe possesses a predictive model in the same sense as a nervous system. That extension remains a separate research proposal.
An equation needs a process theory
A Bayesian expression can describe a relationship among probabilities without specifying the physical algorithm that realizes it. To connect a predictive theory with a device or brain, researchers need variables, update rules, and observations. Variational free energy is an inference-related mathematical quantity; it must not be casually identified with an energy supply measured in joules. We propose showing those definitions beside every interactive model. If a project uses the word prediction metaphorically, that should be clear before a reader mistakes the illustration for a mechanism.
What a strong comparison looks like
Build two models that predict the same measured behavior: one with the proposed predictive update and one with a simpler response rule. Fit both using the same training data, then compare new conditions that make their predictions diverge. A model that merely redescribes familiar behavior has not earned a broader claim. The Hill program can make a practical contribution by publishing these comparisons, including the conditions under which the simpler account works just as well. The decisive output is predictive accuracy under a fair test.
Testing the cosmic extension
To call the universe predictive, first identify what plays the role of internal state, predicted variable, and update signal. If the universe is defined as everything, an external environment cannot be assumed without specifying a larger framework such as Alpha 1. Even then, Alpha 1 is a hypothesis, not an observed input channel. We propose beginning with bounded physical systems where those roles can be measured. Success there would establish a local mechanism and motivate, rather than prove, a broader extension.
The mathematical lens
Bayesian updating relates probabilities; it does not specify a unique neural or cosmic implementation.
Turn it into a question.
- Question
- Does a predictive model outperform a simpler response model?
- Method
- Compare preregistered predictions on held-out perturbations that distinguish the models.
- What would count against it?
- Equal performance with fewer assumptions favors the simpler account for that task.
Prediction becomes a mechanism only when its physical implementation is specified.
Follow the original work.
These sources support the mechanisms and limits discussed here. They do not endorse Hill’s framework. Notes distinguish research proposals, reviews, and observations.
Continue the investigation
Edition 02 · 18 October 2026 · AI-assisted synthesis. See editorial method.