drag to rotate · scroll to zoom
A proof you have already spoken with

A neural network and the brain

A neural network writes poems, solves problems and answers questions. Behind it stand eighty years of work, thousands of scientists and data centers the size of a factory. Nobody thinks it appeared by itself.

And in your head works the network they copied it from. It has hundreds of times more connections, it spends 20 watts, and it assembled itself. Let us compare honestly, point by point.

1 · How it is built

What is inside a neural network

Nothing mysterious. The dots are artificial neurons, the lines are the connections between them. A question comes in on the left, an answer comes out on the right.

There are 37 neurons and 252 connections here. A large language model has hundreds of billions of connections, but the principle is the same.

2 · Learning

Train it yourself

This is a real neural network, and it is running on this very page. The task is to separate the gold dots from the blue ones. Right now its connections are random, and the surface of the answer is a set of meaningless waves.

At every step the network checks its answer against the right one and turns each connection a little. After a hundred steps the gold dots stand on the hill and the blue ones in the low ground. Nobody wrote the answer into it: it found the answer itself.

A red dot is an error of the network. The network: 2 inputs, two layers of 12 neurons each, 1 output, 205 connections.

3 · By itself?

“It trained itself.” Let us check

The network really does learn without being prompted. But it was taught how to learn. Here are three things people gave it. Switch any one of them off.

And people also had to invent the network itself, the layers, the formula of a neuron, and to build the computer on which all of this is calculated.

4 · The price

What it took before a machine could speak

80 yearsfrom the first formula of a neuron (1943) to ChatGPT (2022)
A Nobel Prizein physics for 2024: Hopfield and Hinton, for the foundations of neural networks
175 billiontunable connections in the GPT-3 model (the year 2020)
hundreds of billionsof words written by people, read by such a network while it learned
80 billiontransistors in one chip, the chips it is trained on. Thousands of such chips are needed
≈ 1300 MWhof electricity went into one training of GPT-3, by the estimate of researchers
8 authorsof a paper in 2017 invented the architecture on which the chatbots of today run
0neural networks have arisen without any part played by people

Everything here was invented, calculated and tuned: the formula of the neuron, the layers, the goal, the learning rule, the chips, the power plants. The smarter the network, the more mind has been put into it.

Brown et al., 2020 (GPT-3). Vaswani et al., 2017 (“Attention Is All You Need”). Patterson et al., 2021 (the energy estimate).

5 · The likeness

The neural network was copied from the brain

In 1943 the neurophysiologist McCulloch and the mathematician Pitts described a living neuron by a formula. Every neural network began from that. So the likeness is a real one:

In both, a network of many simple elements
Knowledge is kept not in the elements but in the strength of the connections between them
It learns from examples: some connections grow stronger, others grow weaker

But this is the likeness of a copy to the original. The engineers did not invent the principle. They saw it in something that was already working.

6 · The difference

The copy and the original

Neural network
The human brain
The element
a formula: add up the inputs and compare them with a threshold
a living cell. By the calculations, one neuron of the cortex equals a whole network of 5 to 8 layers
Connections
175 billion in GPT-3
about 100 trillion
Energy
a data center, megawatts
20 watts, like a dim light bulb
Learning
hundreds of billions of words
for a child to begin to speak, thousands of times less is enough
Assembly
factories, chips, thousands of engineers
builds itself from a single cell, by the record in DNA
Repair
trained, and then frozen
rebuilds its connections all through life and works around damage
An “I”
picks the most probable next word
knows that it is. It loves, it chooses, it asks “why”

Azevedo et al., 2009 (86 billion neurons). Beniaguev et al., Neuron, 2021 (one neuron as a deep network).

7 · The building

The most complex network is assembled with no engineers

A neural network is assembled in a factory and trained in a data center. Your brain grew from a single cell in nine months. On average about 250 thousand neurons appeared every minute, and each one found its place and its neighbors.

There were no engineers nearby. There was a record in DNA: three billion letters. It sets the architecture, and the learning goes on for a whole life after that. Nobody programs a child to speak, but the ability to learn any language is already laid in the child.

People invented the architecture, the goal and the learning rule of a neural network. The brain receives them ready-made, even before birth. Who wrote them?

8 · Objections

Four objections and four answers

“Nobody wrote the connections of the network, it trained itself”

True, and you saw it. But you also saw the other thing: take away the goal, the examples or the rule, and what remains is noise. The network learns by itself only because it was taught how to learn. The ability to learn does not arise by itself.

“Evolution trained the brain the way an algorithm trains a network”

Any learning algorithm runs on a machine that is already built. Selection needs a system that is already alive and already reproducing, with a record in DNA. It does not create that system. And people, knowing the goal and looking at a finished sample, spent 80 years on a rough copy.

“Soon AI will surpass a human. So a mind is only computation”

A calculator has long counted faster than we do. If a machine ever overtakes a human in everything, that will show the same thing once again: a mind appears where a mind has worked on it. Nobody has ever watched the reverse happen, not once.

“Then who created the Supermind?”

A cause is needed for whatever had a beginning. The Universe had a beginning, life had a beginning, your mind had a beginning. But the chain “and who created him?” cannot go on without end: at the start must stand the One who has always been. God named Himself in just this way: “I AM WHO I AM”.

Conclusion
A calculator has→an engineer
A neural network has→a human mind
A human mind has→
What next

Still have questions? Ask them

“Questions for Life” is a free course for those who doubt or do not believe. Thirty days, one short lesson at a time in Telegram: meaning, conscience, God, Jesus, suffering. Honest and without pressure.

You can reply to any lesson, and a real person, a mentor, will write back.

If you have a mentor code, enter it on the course page.

Start the course for free →