Intelligence From Simple Systems: Seven Months Inside Cell Signalling

DRAFT · biology · seven months of reading, written up in one go

I am not a biologist. I want that on the record before anything else, because everything below is a summary of what I have read, not what I have proven, and a real biologist would put it differently. What I am is someone who is more and more fascinated by how intelligence emerges from simple systems. Not intelligence in the brain sense. The other kind. The kind where nothing in the system is clever and the system is clever anyway.

I have been going down this rabbit hole for about seven months now, reading about cell signalling. This is my summary of what I have learned. I have tried to keep it high level without making it stupid, which is harder than it sounds, because the honest version of a lot of this is a chain of acronyms and the useless version is "cells talk to each other."

The thing that keeps pulling me back is that I find the similarities with mechanical engineering very close, and I enjoy that more than I expected to. I should be upfront: that is my lens, not biology's claim. A cell is not a machine that someone designed. But every time I read one of these mechanisms I recognise the shape of it from work, and the recognition is not vague. It is specific enough to be useful.

The thing that got me

Here is the fact that started it. Take one signal molecule, let it spread out from a source, and let it fade with distance. Every cell reads how much of it is present right where it sits, compares that number to a couple of fixed levels, and becomes a different type of cell depending on which side of those levels it lands on. High means one fate, medium another, low another. That is it. That is how a spinal cord gets its regions.

No cell is told what to become. No cell knows where it is. There is a concentration, and there are thresholds, and out the other side comes a body plan. The first time I understood that properly I sat there for a while, because I have built that circuit. Everyone who has ever wired a sensor has built that circuit. It is a comparator. You take an analog value, you set your trip points, and you quantise a smooth signal into discrete states. A cell is doing an analog to digital conversion and the output is what kind of cell it is going to be for the rest of its life.

What kept happening

Then it kept happening. Not once, everywhere.

A receptor is a lock on the outside of a cell, and the signal is the key. But the interesting part is that a single receptor catching a single molecule does almost nothing. Two of them have to pair up first. Pairing is the switch. That is an interlock: two things have to line up before the circuit closes, so a stray key cannot fire the machine on its own.

Once it fires, the signal runs down a relay chain and multiplies about a thousand times on the way. That is a gain stage. It is the same reason you put a servo between a joystick and a hydraulic ram. The input is too weak to do work, so you spend energy to amplify it, and the price you pay for the gain is that now you desperately need feedback, because an amplifier with no feedback is just a way to turn a small mistake into a large one. The cell has the feedback. The last stage in the chain reaches back and dampens itself.

Genes get switched on and off by proteins, and those switches wire into each other as logic. Two inputs required means an AND gate, and it acts as a coincidence detector. Either input will do means an OR gate, and it buys you redundancy, a backup path. This is not a metaphor someone laid on top afterwards. The behaviour is the gate.

And then the one I like most. A gene can switch itself on. Once it is on, it keeps itself on, so when the original signal fades the state stays. That is a latch. It is an over centre toggle, a relay holding its own coil. It is how a signal that lasts minutes turns into an identity that lasts decades. A neuron stays a neuron for life because something flipped and nothing unflipped it. Memory, built out of nothing but a loop.

The parts I did not expect

Some of it I would never have guessed, and those are the ones I enjoyed the most.

Noise. A gradient is a fuzzy thing. Molecules arrive at random. If a cell sampled it for an instant it would get garbage. So the cell listens for hours and averages. That is a low pass filter. It is debouncing a noisy sensor, and it is the first thing you learn to do when a limit switch chatters.

Ratios. Reading one gradient in absolute terms is fragile, because the whole signal can drift. So the cell reads two opposite gradients and takes position from the ratio of the two. That is ratiometric measurement. It is why you use a bridge instead of a single strain gauge: the common error cancels. It also means the pattern still works if the embryo is a different size, which is scaling by a dimensionless ratio, and I do not think I have ever seen it done more elegantly.

Sharpening. The cells do not just read the signal, they eat it. Each one pulls the molecule in and destroys it. That steepens the gradient and makes the boundary crisp instead of smeared. It took me a while to see that destroying your own input is a design decision and not a bug.

Isolation. Neighbouring cells wire their insides together with real channels, so small molecules and current pass straight through. That is a shared bus, and it is why a heart beats as one piece instead of as a billion arguments. But when a cell is injured, the channels slam shut and seal it off. A fuse. You isolate the fault so it does not take the rest of the system with it.

Subtraction. This is my favourite. You did not grow fingers. You grew a paddle, and then the cells in between were told to die, on cue, and the gaps are what is left. Your hand is a machined part. The shape came from removing material, not from adding it.

Where it breaks, and why that convinced me

What finally sold me on the engineering reading is what happens when it fails, because it fails the way machines fail rather than the way organisms are supposed to fail.

The master switch in that relay chain can get stuck in the ON position. About a quarter of all cancers are that one switch jammed. The grow signal never turns off. And separately, the brakes on the system can be lost. Stuck accelerator, no brakes. That is not a poetic description of cancer, it is close to the mechanism. When I read that, I stopped thinking of the analogy as an analogy.

Break the build instructions instead of the growth controls and you do not get a tumour, you get a body assembled wrong. Same system, different failure, and which failure you get depends on which signal you corrupted. That is a fault tree.

So what is the intelligence

Here is where I land, for now, knowing I am out of my depth.

There is no controller. There is no cell holding the drawing. Every cell is running a small, local, almost dumb rule: read the concentration here, compare it to a level, add up the inputs, flip a switch, hold the switch. The body is what those rules do when there are thirty seven trillion of them and they can talk. The intelligence is not in the parts. It is in the communication, and that is the whole point. At this scale, life is not a substance, it is a conversation.

That is the thing I keep chewing on. In mechanical engineering we get intelligent behaviour by making the controller smarter. Biology got it by making the parts talk and letting the behaviour fall out. I do not know what to do with that yet, but I am fairly sure it is the more interesting way round.

Below is the whole thing as a flowchart. I did not want to give you a PDF to pinch and scroll, so it is built into the page: eight phases, read the teal arrows down. Every block has a plain English line under it, and the lettered circles are jumps, so click one and its twin lights up. It is dense on purpose. That density is the point.

How cells communicate to build a body

A plain language flowchart. Read the teal arrows top to bottom through 8 numbered phases. Every block also has a simple one line explanation in grey italic. A circled letter marks a jump, so find the other circle with the same letter to see where it connects. Black arrows are the steps inside one phase.

← Back to the blog

Last meaningful edit: 15 July 2026.