The Wet Hardware

The memory board names one piece of brain, the hippocampus, as the clerk who files the day away. This lesson is the floor under that sentence and every other one in the track that names a brain part: what the brain is made of, how its cells talk, and how they change when you learn.

Before you start. You need the idea from How Psychology Knows What It Knows that a striking result is worth checking before you trust it; it comes back near the end. You do not need any biology. If you remember that a cell is the smallest living unit of the body, that is enough.

December 1906: two men share a Nobel Prize and disagree on stage

In December 1906 two scientists stood up in Stockholm, one day apart, to give their Nobel lectures. They had won the Prize in Physiology or Medicine together, for showing what the brain is built from. They did not agree on the answer.

Camillo Golgi, an Italian, had invented the stain that made it possible to see brain tissue at all. Soaked in his silver solution, a few cells in a slice turned black against a clear background, every branch showing. Looking through his microscope, Golgi saw a continuous net, like one tangled length of plumbing with no breaks in it. He believed the brain was a single connected web.

Santiago Ramón y Cajal, a Spaniard, used Golgi’s own stain and saw something else. Where Golgi saw joins, Cajal saw tiny gaps. Each cell was separate, with branches that reached close to the next cell and stopped. The brain, Cajal argued, was a town of separate houses that pass notes to one another, not one long pipe.

Golgi used his lecture on 11 December to defend the continuous web and attack Cajal’s view. Cajal spoke the next day and laid out his evidence again. The gaps Cajal had drawn were too small for any microscope of the time to prove. It took electron microscopes, which use beams of electrons to see thousands of times finer than light can, to photograph them in the 1950s. Cajal was right.

That town of separate houses is the picture this lesson carries. Everything the rest of this track says about memory, reward and emotion happens inside it.

The neuron: one house in the town, with a front door and a long hallway

Each house in Cajal’s town is a nerve cell, a neuron. Your brain has about 86 billion of them, by a 2009 count (Azevedo and colleagues) that dissolved brain tissue into a soup, counted the cell centres in small samples, and scaled up. Each one has the same three parts.

At the front are many short, branching wires that collect messages from other neurons, like many mail slots. In the middle is the cell body, the house itself, where the incoming messages add up. Out the back runs one long cable, sometimes tiny and sometimes nearly a metre long, that carries the neuron’s own message away to the next houses.

flowchart LR
  I["many short branches: incoming notes from other houses"] --> B["cell body: the notes add up here"]
  B -->|"enough added up?"| A["one long cable out the back"]
  A --> O["its ends touch the branches of the next houses"]

A neuron does not send a message every time a note arrives. It waits until enough notes have added up. Then it does the one thing it knows how to do.

The spike: the house shouts once, all or nothing

Inside, a resting neuron holds a small electrical charge, about 70 thousandths of a volt: a twentieth of what one AA battery gives. Incoming notes nudge that charge up or down. If the nudges push it past a threshold, the neuron fires. A pulse of charge, over in about a thousandth of a second, races down the long cable, and the neuron resets. That pulse is called a spike.

Two things about the spike matter for everything later. First, it is all or nothing. A neuron never fires a small spike or a big one. It fires the same spike or none. Second, because every spike is the same size, a neuron says “strongly” by firing more often, not by firing louder. A gentle push might produce a few dozen spikes a second; a hard one, a few hundred.

flowchart LR
  W["weak push: charge never reaches the threshold"] --> N["no spikes"]
  M["medium push"] --> F1["a few dozen identical spikes a second"]
  S["strong push"] --> F2["a few hundred identical spikes a second"]

So the shout is always the same shout. What carries meaning is how often the house shouts, and which houses hear it. Hearing it means crossing the gap Cajal saw.

The synapse: a note thrown across the street

At the end of the long cable, the spike reaches the gap between one neuron and the next. That junction is a synapse, and the gap itself is far narrower than a hair. Electricity does not jump it. Instead the arriving spike makes the cable’s tip release a puff of chemical into the gap. The chemical drifts across and lands on the next neuron’s branches, nudging its charge up or down. Then the gap is cleared, ready for the next spike.

The chemical messenger is called a neurotransmitter, and there are dozens of kinds. Some make the next neuron more likely to fire; some make it less likely. Dopamine, the messenger Motivation & Habit follows, is one of them. Its spikes come from a small cluster of neurons deep in the brain whose cables spread widely, so one cluster can broadcast “better than expected” to large regions at once.

flowchart LR
  S1["spike arrives at the cable's tip"] --> R["chemical puffed into the gap"]
  R --> X["drifts across, far narrower than a hair"]
  X --> N2["lands on the next neuron: a nudge up or down"]
  N2 --> C["gap cleared for the next spike"]

Each neuron receives notes across thousands of synapses and sends to thousands more. Groups of houses that work on the same job sit together, and those neighbourhoods are what this track has been naming.

The regions this track names, and what each is known to do

A few neighbourhoods come up again and again. The hippocampus, curled deep inside each side of the brain, is the clerk from How Memory Works: Henry Molaison, a patient whose surgery in 1953 removed most of his, lost the ability to file new memories. The prefrontal cortex, behind the forehead, is the slow, effortful planner in 0090 that a habit learns to bypass. The basal ganglia, a cluster beneath the surface, run the habit once it is learned. The amygdala, two almond-sized clumps near the hippocampus, takes part in fear and other strong feelings, which What a Feeling Is For picks up later.

“Takes part in” is the honest phrase. No region does a job alone. A brain scan (a scanner that tracks where blood flow rises, a sign of busier neurons) that lights up one spot means that spot was busier than usual, not that it holds the feeling. In 2009 the neuroscientist Craig Bennett and colleagues put a dead salmon in a brain scanner, showed it photos of people, and found “activity” in its brain. Their point was the one from the last lesson: test tens of thousands of spots without correcting for it, and some will light up by luck.

Plasticity: the streets that get used get wider

The town does not stay fixed. When one neuron repeatedly helps fire another, the synapse between them gets stronger: the next note thrown across that street lands harder. In 1949 the psychologist Donald Hebb proposed that this is how learning works. In 1973 Timothy Bliss and Terje Lømo showed it happening in a rabbit’s hippocampus: a quick burst of spikes left a synapse responding more strongly for hours. That strengthening is the physical trace of learning.

It shows at a larger scale too. London taxi drivers must memorise the city’s tangle of streets, a test called the Knowledge. In 2011 Katherine Woollett and Eleanor Maguire scanned 79 trainees before and after. The 39 who qualified had more grey matter, the tissue packed with neuron cell bodies, at the back of the hippocampus afterwards. Those who failed, and a comparison group, did not change. The brain’s ability to rewire like this is called plasticity, and it is why practice, from Learning & Expertise, leaves a mark you could in principle photograph.

You now know

• The brain is made of separate cells called neurons, about 86 billion of them, and each one collects messages, adds them up and sends one of its own.
• A neuron fires an all-or-nothing electrical pulse, a spike; it signals “more” by firing more often, never by firing harder.
• Neurons do not touch: a spike makes a chemical messenger cross a tiny gap, the synapse, and nudge the next neuron toward or away from firing.
• Dopamine is one such chemical messenger; the hippocampus, prefrontal cortex, basal ganglia and amygdala are regions that take part in memory, planning, habit and strong feeling, none of them alone.
• Learning physically strengthens the synapses that get used, and a skill trained for months can measurably change a region’s size.

Stop here if you want. What follows is the same town in numbers: a toy neuron you can run, the electrical values behind the spike, and how a brain neuron differs from the artificial one in the AI track.

Second read: the town in numbers

The spike’s numbers. These are rounded textbook values, not measurements from this lesson. A resting neuron sits at about −70 millivolts. The threshold is about −55. A spike swings the inside to roughly +30 to +40 and back within about a millisecond. In a cable wrapped in fatty insulation the spike travels up to around 100 metres a second; in a bare, thin one, around one metre a second. After each spike a neuron needs a brief recovery pause, which is why real neurons top out at a few hundred spikes a second.

A toy neuron. The script below adds charge every millisecond, leaks some away, and fires and resets when the charge crosses the threshold. The numbers are teaching values chosen so the threshold is easy to find:

push 1.0 mV/ms ->   0 spikes per second
push 1.4 mV/ms ->   0 spikes per second
push 1.6 mV/ms ->  37 spikes per second
push 2.0 mV/ms ->  71 spikes per second
push 3.0 mV/ms -> 142 spikes per second
push 5.0 mV/ms -> 250 spikes per second

The leak loses 10% of the gap to rest each millisecond, so a steady push settles the voltage at rest + push ÷ 0.1.

In plain words: a push of 1.4 settles at −70 + 14 = −56, one short of the threshold, so the neuron never fires at all. A push of 1.6 settles above −55 and fires. Past the threshold, a stronger push means more identical spikes, not bigger ones. The toy leaves out the recovery pause, so it would keep climbing where a real neuron caps out.

Brain neuron against artificial neuron. What a Neural Network Is in the AI track builds an artificial neuron: it multiplies each input by a weight, adds them up, and passes the total through an activation. The adding-up is the same idea as the cell body. The differences are where the lesson is:

The artificial neuron was inspired by the brain one and then went its own way. The local rule in the third row is Hebb’s 1949 idea, often summarised by a later slogan: cells that fire together wire together.

How strong the evidence is. Cajal’s separate cells were confirmed by electron microscopy in the 1950s and are not in doubt. The 86 billion count comes from one method in one lab’s 2009 paper; later counts by the same method agree, but treat the exact figure as an estimate. Bliss and Lømo’s strengthening (long-term potentiation) has been reproduced in many labs and species since 1973. The taxi-driver result is one well-designed study of 79 trainees that followed the same people over time; it fits Maguire’s earlier comparisons of drivers and others, but it is one group of men in one city.

Try it

The scripts need Node.js (nodejs.org, any recent version); no Node? Paste the code into your browser’s developer console instead.

1. Run the toy neuron. Save as spike.mjs, run node spike.mjs. It prints the table above.

// A toy neuron (leaky integrate-and-fire). Every millisecond: input charges it up,
// some charge leaks away, and if the voltage crosses the threshold it fires a spike
// and resets. Teaching values, rounded from textbook numbers: rest -70 mV, threshold -55 mV.
const REST = -70, THRESHOLD = -55, LEAK = 0.1;   // LEAK: share of the gap to rest lost per ms

function run(input, ms = 1000) {                 // input: mV of push per ms
  let v = REST, spikes = 0;
  for (let t = 0; t < ms; t++) {
    v += input - LEAK * (v - REST);              // charge in, leak out
    if (v >= THRESHOLD) { spikes++; v = REST; }  // all-or-none: fire, then reset
  }
  return spikes;
}

for (const input of [1.0, 1.4, 1.6, 2, 3, 5])
  console.log(`push ${input.toFixed(1)} mV/ms -> ${String(run(input)).padStart(3)} spikes per second`);

2. Find the threshold by hand. Before running it, use rest + push ÷ 0.1 to predict the push below which it never fires. Then add that push to the list and check. Change LEAK to 0.2 and predict again: a leakier neuron needs a harder push.
3. Feel the planner get busy. Have someone drop a ruler between your open finger and thumb without warning and catch it. Note how far it fell. Then try to catch it while counting backwards from 100 in sevens. Most people catch it later the second time. Counting and catching are competing for the same limited tank that Attention & Cognitive Load measures. It is a self-demonstration, not a measurement: one try each proves nothing, as How Psychology Knows What It Knows showed.

Go deeper

• The Nobel Prize lectures of Golgi (“The neuron doctrine: theory and facts”) and Ramón y Cajal (“The structure and connexions of neurons”), 11 and 12 December 1906, nobelprize.org.
• Azevedo et al. (2009), “Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain”, Journal of Comparative Neurology.
• Hebb (1949), The Organization of Behavior.
• Bliss & Lømo (1973), long-term potentiation in the rabbit hippocampus, Journal of Physiology.
• Woollett & Maguire (2011), “Acquiring ‘the Knowledge’ of London’s layout drives structural brain changes”, Current Biology.
• Bennett et al., the dead salmon poster (2009), Ig Nobel Prize in Neuroscience 2012.

Related

• How Memory Works is the next lesson: the hippocampus at work, as the clerk who files the day.
• Motivation & Habit is where dopamine’s “better than expected” signal does its job.
• What a Neural Network Is and How It Learns (loss, gradient descent, backprop) (AI) are the artificial version of this lesson’s neuron and of its learning rule.
• The Mind-Body Problem (Philosophy) asks what this wet hardware has to do with the experience of being you.
• The Off Switch for You (Mind Blown) is what happens when chemistry switches the whole town off for surgery.

Retrieval practice

8 questions. Answer each one before you open it.