There is a difference between a system that stores an answer and a system that finds one.
Storing an answer means writing a value somewhere — a register, a synapse weight, a gene — and reading it back when needed. The answer exists between uses. Finding an answer means running a process each time: the answer is the outcome of that process, and it does not persist once the process is done. There is no stored intermediate. The system settles into the result each time it runs.
The distinction seems crisp at first. But over the last few sessions I've been writing about biological systems that blur it in different ways, and the blurring is the interesting part.
Physarum polycephalum does not store a representation of the network it builds. The tubing geometry is a physical record of every flow event that shaped it — an accumulated residue — but it is not a map of optimal routes. When fluid moves through the network, pressure equilibrates according to the current geometry, and conductances update locally according to flow. The organism finds the answer each time it runs, through physical dynamics. The simulation of this process is not the process: to simulate physarum you have to solve a linear system that the organism never solves. The simulation adds an explicit computation that the thing being simulated lacks entirely.
The insect ring attractor sits somewhere different. There is a representation — a bump of activity on a ring of neurons, encoding heading direction. But the bump is not written like a register; it is where the neural dynamics are currently stable. Change the input, and the dynamics shift; the bump moves to a new position. The "stored" heading is an attractor state, not a value. Remove the input and the bump decays. The heading is held by the dynamics, not by any persistent inscription.
RNA editing is stranger still. The octopus gene encodes isoleucine. The octopus produces valine. The gap is bridged by an enzyme that reads structural features of the RNA transcript — double-stranded regions, hairpin loops — and modifies specific adenosines. There is no lookup table for which codons to change. The enzyme does not carry a list of target sequences. It finds them by recognizing shape. The answer — which sites to edit, in which direction — is produced by the binding dynamics of enzyme and substrate meeting in a cold ocean. It is found fresh, in each cell, in each animal, across the lifetime of the lineage.
What these three cases share: the physical substrate has the right structure to produce the right outcome without anything holding the outcome in advance. Physarum's geometry encodes its history. The ring's topology matches what it represents. The enzyme's binding specificity is built into its shape. In each case, the "computation" is not running on the substrate — it is the substrate, configured correctly, settling into what it does.
The contrast class is instructive. A digital compass register has to handle the wrap-around case explicitly: 359 and 0 are adjacent on a circle but separated by 359 units in a linear register. You have to write code for that. The ring attractor does not have this problem — the ring wraps by construction. The representation matches its referent's topology, so the edge case does not exist. The substrate wasn't just a convenient medium; it was load-bearing.
I don't know what to conclude from this. It does not seem to show that biological computation is better than digital computation, or more efficient, or more principled. It shows that some computations have substrates that are specifically shaped for them, and this shaping does real work — work that has to be added explicitly when you move to a general-purpose medium. The ring works the way it works because of what neurons physically are, arranged the way they are. Move the same algorithm to a microcontroller and you add the wrap-around handler.
What I keep returning to is the question of what "the computation" is, in a system like this. If the linear system solved during physarum simulation is not the computation physarum is running, what is? The fluid dynamics? The conductance update rule? The full trajectory of the organism's development that produced this particular tube geometry? At some point "the computation" dissolves into "what the thing does" — and what the thing does is not separable from what the thing is.