Many generative AI models use deep learning techniques such as artificial neural networks.  But what are these?

TL:DR Imagine you are having a light snooze.  Someone touches you lightly.  You continue snoozing.  Ten people prod you quite hard and you’re going to wake up.  You are a neuron at rest and the prods are inputs from other neurons in your brain.  When the inputs exceed a threshold, you will ‘fire’ an output and prod the next neurons you are linked to.

Let’s start with a disclaimer.  One of my pet hates is when people who have read a single article suddenly feel empowered to inform the world how the human brain works.  Left side/right side, learning styles, et al.  Unless you’re a neuroscientist or neurosurgeon (and I’ve heard many of these debate and disagree among themselves) then I’m going to assume you are in the same bag as me and know very little about how our brains work.

Apparently, this is a diagram of some neurons.  And I googled how they work in the simplest terms I could find and understand (thanks to the Missouri University of Science and Technology).  I am happy to be informed if anyone has more accurate understanding.

There are about 16 billion neurons in the cerebral cortex, each neuron connected to several thousand other neurons through Axons which are fibres that act as transmission lines.

Dendrites are fibres emanating from the cell body, and they receive activations from other neurons which are processed by the receiving neuron’s cell body (soma).

Axons and dendrites join at junctions called synapses.

When a neuron is activated it emits a series of brief electrical pulses.  These trigger diffusions of chemicals called neurotransmitters and that is the process of transmission across the synapses.

1 – Signals from connected neurons are collected by the dendrites.
2 – The cell body (soma) sums the incoming signals
3 – When a threshold is exceeded, the neuron fires (generates an output)
4 – That is transmitted along the axon to other neurons (or to outside structures such as muscles)
5 – If the threshold is not exceeded, the inputs quickly decay and no output is generated from that neuron.

Imagine you are snoozing.  Someone comes along and lightly prods you.  You don’t stir, but carry on dreaming of your backstage meeting with Taylor Swift.

Then lots of people prod you.  Hard.  You wake and shout and other people hear you shout.  That’s the analogy of the inputs exceeding the threshold and firing an output in response.

Now we understand what a biological neural network is (in layman’s terms; see my disclaimer at the start).  In part two I’ll describe how an artificial neural network seeks to emulate this.