Concept note · the lab vocabulary
Morphological computing
Letting the geometry and physics of a substrate carry the computation — no address space, no wiring fabric.
What it is
Computation carried by the shape and physics of a body or material rather than by a stored program that names its connections. The seventy-year habit of computer science is to draw lines from point A to point B — a fixed fabric of wires, or a digital address space standing in for one — and to make learning a matter of changing numbers stored on those lines. Morphological computing asks what happens when the lines are not given: when connectivity is something the substrate’s own geometry makes true or false, moment to moment, and strength is a physical quantity like overlap or contact area rather than a scalar in a table.
In the lab
- Fluid Morphic Computing →
- The substrate stated tightly — blank spheres as channel state, r₁ + r₂ > d as synapse creation, overlap depth as conductance, a single leak as plasticity — with a working technical note and a reference simulator.
Our position
Interconnect itself can be an emergent geometric predicate. Our formulation is deliberately minimal: state is volume, a connection exists exactly while two volumes span the distance between them, strength is the depth of the overlap, and one leak is both forgetting and structural pruning. We hold the line between a substrate and a cognitive architecture — the rules specify the first and make no claim to the second — and we state the sharpest objection to the model in our own words before anyone else has to: with no pairwise term in the update, what it binds is co-location, not correlation. A proposal that cannot name its own weakest joint is a metaphor wearing equations.
Honest limitations
It is a proposal on paper and in a simulator, not a device. Nothing here proves computational universality, an energy advantage or robustness; the materials that would embody it — gels, ionogels, emulsions, compressible conductors — exist only in pieces, and assembling them into a registered field of independently driven volumes is an unsolved transduction problem. The naive simulator is quadratic in the number of channels per tick. And the idea has neighbours — in-materio computation, Physarum processors, hydrogel neuromorphics, overlapping-kernel models — that already occupy the surrounding conceptual space; what is claimed is the specific combination, not the originality of every part.
The experiments are the proof
Every claim in this note is made concrete somewhere in the lab — the idea exists because the machines needed it.
The map of the lab →