Concept study / Information & adaptationConceptual illustration. Shapes, colors, and motion are not measured data.
Two clocks can agree because both track the same external reference, not because either controls the other. Likewise, two physical systems can share information through a common cause. Hill’s hypothesis specifically says prior interactions leave information that influences future behavior. The word influences makes this a causal claim. Its strongest test is to alter the proposed record while holding relevant alternatives under control, then ask whether the predicted future response changes. Observational analysis can contribute, but its assumptions must be explicit. A visually convincing relationship or a directed statistical measure is not a substitute for identifying the causal structure.
Prediction and intervention differ
Transfer entropy asks whether the past of one variable adds predictive information about another variable’s future beyond the latter’s selected history. This can help identify directional statistical relationships. Unobserved common causes, inappropriate sampling, and incomplete conditioning can still complicate the interpretation. Pearl’s causal framework distinguishes observation from intervention and shows why causal assumptions matter. For the project, every claimed influence should therefore include a diagram or verbal account of possible common causes and explain which of them the study can address.
A selective record disruption
We propose a two-stage design: first establish that a candidate state carries a history signal; then selectively disrupt that state and test the later behavior. The intervention must be checked for side effects. If erasing a record also damages the device or changes its energy supply, a worse outcome may have an ordinary alternative explanation. Include a sham intervention and a control that produces comparable disturbance without erasing the proposed content. This makes the test demanding but interpretable.
Information with a functional role
Kolchinsky and Wolpert proposed a formal way to connect information with a system’s continued viability using counterfactual changes to system–environment correlations. Their approach is a candidate definition, not a universal consensus about meaning. We propose using that distinction productively: state the chosen viability metric, show how scrambling a correlation changes it, and test alternative definitions. A result that depends entirely on one convenient metric should remain conditional. This is a route from stored correlation to functional relevance without introducing consciousness by assumption.
The mathematical lens
Intervening on X and merely observing X answer different questions in a causal model.
Turn it into a question.
- Question
- Does the retained record causally improve the later response?
- Method
- Disrupt the candidate record with sham and disturbance-matched controls, then measure the specified outcome.
- What would count against it?
- No selective effect, or an effect explained by intervention damage, weakens the causal-memory claim.
Show what changes when the proposed record is changed.
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.