All or Nothing

A neuron's membrane sits at about minus 70 millivolts relative to the outside, maintained by pumps moving ions against their gradients at considerable metabolic expense. It is a charged battery waiting to discharge.

When input pushes the voltage past a threshold (around minus 55) sodium channels open, sodium rushes in, the voltage spikes, and the change opens neighbouring channels, which regenerates the spike further along. The result is an action potential that travels the length of the axon arriving at full strength, because it is re-created at every point rather than propagating like a wave in water.

The consequence of this design is the foundational constraint on neural coding. Every spike is the same size. A neuron cannot signal a stronger stimulus with a bigger spike; it is all or nothing.

So intensity must be encoded some other way, and the answer is rate and timing: more spikes per second, or spikes arriving in particular patterns, or spikes synchronised across populations. Everything the nervous system represents is built out of that.

At the far end of the axon the signal becomes chemical. The spike triggers the release of neurotransmitter into a gap of a few tens of nanometres, where it binds receptors on the next cell and nudges its voltage up or down.

Introducing a chemical step into an electrical system looks like poor engineering, and it is what makes the whole thing modifiable. The strength of a synapse can be adjusted; a third neuron can intervene at the gap; hormones and drugs can bind the same receptors. A rigidly electrical nervous system would be fast and would be nearly impossible to tune.

A single neuron may receive input at thousands of synapses, some excitatory and some inhibitory, and it fires only when the sum crosses threshold. That summation is where the computation happens, and it is happening at eighty-six billion neurons simultaneously.

How experience changes the system is the next question, and the answer has been substantially the same since Donald Hebb proposed it in 1949.

Hebb's rule, usually compressed to "cells that fire together wire together", is actually causal in his formulation: if one cell repeatedly takes part in firing another, the connection between them strengthens. In the 1970s long-term potentiation was found (a lasting increase in synaptic strength following high-frequency stimulation) and it looks very much like the mechanism.

Learning, on this account, is not information being deposited somewhere. It is the pattern of connection strengths changing, so that the same input produces a different output. There is no file.

Which brings us to plasticity, and to correcting the popular version without denying the real one.

Synapses strengthen, weaken and are eliminated throughout life. Cortical maps reorganise after injury: lose a finger and the cortical territory that served it is annexed by its neighbours within weeks. Training changes structure measurably: London taxi drivers who have learned the city's layout show enlarged posterior hippocampi, and the effect scales with years of experience.

The limits are equally real. Critical periods exist: a child deprived of patterned visual input in early life does not develop normal vision later, and language acquisition after adolescence rarely reaches native fluency. Reorganisation after adult stroke is partial and effortful. The popular claim that the brain can be rewired at will considerably overstates a genuine and constrained phenomenon.

And two myths worth dispatching directly.

We do not use ten per cent of our brains. Imaging shows activity throughout across any period; damage to almost any region produces a deficit; and the brain consumes around a fifth of the body's energy while being a fiftieth of its mass. Evolution does not maintain that expense for tissue that does nothing.

Nor is any neurotransmitter "the" chemical of anything. Dopamine is the standard example: it operates in several pathways with different functions (reward learning, movement, prolactin regulation, executive control) through at least five receptor subtypes distributed differently across them.

This is why drugs acting on dopamine are so awkward. Antipsychotics that block it to reduce psychosis produce movement disorders, because the same transmitter runs the motor pathway. Parkinson's treatment that raises it to restore movement can produce impulse control problems, because the same transmitter runs reward learning. There is no dial marked dopamine; there are several circuits that happen to use the same molecule, and a drug in the bloodstream cannot tell them apart.