🕸️ Neural Networks
6th Grade 🤖 Artificial Intelligence
An artificial neural network is a way of building software that learns from examples. The name comes from the brain, where nerve cells (neurons) connect to each other — but a neural network is really a huge number of simple calculations hooked together, not a real brain.
Inspired by the brain — but not a brain
Your brain has billions of nerve cells, and each one receives signals from other cells and passes signals on. Scientists got an idea from this: what if we built simple “cells” in a computer that take in numbers, calculate something and pass the result along?
That’s how the artificial neural network was born. But don’t get confused: an artificial neuron is much simpler than a real nerve cell, and the network doesn’t think like a person, doesn’t feel and isn’t aware of itself. It’s a very big math formula whose settings were set during training. The brain inspiration is mostly a general idea, not a copy.
Artificial neurons and weights
An artificial neuron takes in a few numbers, multiplies each one by a “weight” — a number that decides how important that input is — adds everything up and decides what number to pass on. Picture deciding whether to go outside to play: “It’s sunny” matters a lot, “A friend is waiting” matters even more, and “I have homework” lowers the chances. Each reason has a weight, and the total makes the decision.
A neural network often has millions of weights, and sometimes far more. Nobody writes them by hand — they’re set during training. The weights are really the network’s “memory” of what it learned from the data.
Layers: from tiny details to the whole picture
The neurons are arranged in layers. The input layer takes in the data — for example, the colors of all the tiny dots (pixels) in a picture. The output layer gives the answer — for example, “cat” or “dog.” In between are hidden layers that do most of the work.
In networks that recognize pictures, researchers found that the first layers tend to react to simple things like lines and edges, later layers to shapes like circles and corners, and even deeper layers to whole parts like an eye or an ear. Step by step, the network builds an answer. When there are many such layers, it’s called “deep learning” — “deep” because of the number of layers, not because the network thinks deep thoughts.
This is how things you know work: face recognition that unlocks a phone, speech recognition that turns talking into text, and translating a sentence from one language to another.
How a network learns from mistakes
At the start of training the weights are random, and the network answers almost by guessing. You show it a picture of a cat, and it says “dog.” Now you calculate how wrong it was, and then nudge every weight a tiny bit in the direction that would have made the mistake smaller. After millions of examples and tiny corrections, the network gets better and better.
Imagine an old radio with tons of tuning knobs: each time you turn every knob a smidge and listen to whether the sound got clearer. No single knob “knows” what a cat is — the knowledge is spread across all the weights together.
A box that’s hard to peek inside
One problem with neural networks is that it’s hard to explain why they gave a certain answer. With millions of weights, you can’t just read them and understand. That’s why they’re sometimes called a “black box,” and many researchers are working on ways to better understand what happens inside.
There’s another problem too: a neural network can learn the wrong patterns without anyone noticing. If, for example, all the training photos of wolves were taken in snow, the network might learn to recognize “snow” instead of “wolf.” Researchers only discover mistakes like this when they test the network carefully on different examples.
That’s one more reason not to blindly trust a computer’s answer — especially for important decisions. A person needs to check, ask questions and make sure.
Key facts
- An artificial neural network was inspired by the brain, but it’s a math calculation, not a brain.
- Each artificial neuron multiplies its inputs by weights and adds them up.
- The weights are set during training, not written by hand.
- Deep learning is learning with a network that has many layers.
- It’s hard to explain why a neural network answered the way it did — so it’s sometimes called a “black box.”
Quick quiz
Where did neural networks get their name?
What does a “weight” do in an artificial neuron?
Who sets the weights in a neural network?
What does “deep” mean in “deep learning”?
In a network that recognizes pictures, what do the first layers tend to react to?
What happens at the start of training a neural network?
Why is a neural network sometimes called a “black box”?
Which statement about neural networks is true?
💭 Think and talk about it
- Pick an everyday decision, like what to eat at lunch. What reasons affect it, and what “weight” would you give each one?
- Why do you think it matters to understand why a computer gave a certain answer, especially in medicine or in a courtroom?
- How is a neural network like your brain, and how is it very different?
Questions kids ask
Is a neural network an artificial brain?
No. It got a general idea from the brain, but it’s a much simpler math structure. It doesn’t think, feel or understand like a person — it calculates using weights it learned from data.
Where are neural networks used?
In lots of places: face recognition on phones, speech recognition, translation, spam filtering, video recommendations and chatbots that write text.
Why do you need such powerful computers to train a network?
Because training uses millions of examples, and for each one the computer has to calculate the answer and adjust millions of weights. That’s an enormous number of calculations, so special computers that do many calculations at the same time are used.