Conceptual illustration. Shapes, colors, and motion are not measured data.
A working vocabulary is part of the model. Energy is a quantitative property expressed in joules; force is expressed in newtons; their relationship depends on a physical process. Geometry might mean distances, curvature, boundary shape or network connectivity. Information needs a specified variable and uncertainty measure. For this program, memory means that a present readout contains recoverable information about a controlled earlier input. Learning is a stronger proposed criterion: experience changes future performance on a declared task. These definitions can be refined, but changing them changes the claim being tested and must be recorded.
Separate state from amount
Two systems can have equal total energy while occupying different physical states. A single energy value therefore does not identify a stored message, shape, phase distribution or history. To investigate Hill’s intuition, first specify candidate state variables that might carry the trace: structural arrangement, field configuration, correlations or chemical composition. Then ask whether a readout distinguishes histories when energy and relevant environmental conditions are matched. This is a proposed discriminating experiment. It does not assume that all such variables can be controlled perfectly or that unexplained variation implies a universal record.
Give geometry a representation
Choose the representation before looking for similarity. A point cloud supplies distances and density correlations; a surface supplies curvature and topology; a graph supplies edges and connectivity. These objects admit different statistics. A spiral photograph and an atom’s probability distribution cannot be compared simply by rotating both images until they look alike. The comparison must state its projection, normalization and uncertainty. If the intended shared property is a dimensionless ratio, calculate that ratio in each system using documented measurements, and identify whether the same physical assumptions apply.
Require a decoder for memory
A candidate memory observable should have a written encoding and readout task. For example, two randomized training sequences can be followed by the same probe; a held-out decoder tries to identify which sequence occurred. The decoder must not receive labels through filenames, run order or instrument settings. Retention time, disturbance tolerance and erasure behavior are separate measurements. A successful decoding result establishes a history-dependent state within the experiment. It does not establish consciousness, personal memory transfer or a record that survives arbitrary transformations of the system.
The mathematical lens
Energy and force have different dimensions. Naming a new concept does not remove the need to define its units and readout.
Turn it into a question.
- Question
- Does the term identify an observable rather than a metaphor?
- Method
- Write the unit, instrument output, processing rule and alternative explanation for every variable.
- What would count against it?
- If different physical quantities can satisfy the same wording, the definition is not yet discriminating.
Keep the intuition broad, but make every measured quantity precise.
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.