Jargon:Neural network
1 min read
Say this at the party
“A computer program made of lots of tiny connected maths units that pass numbers along, loosely inspired by how brain cells connect.”
In plain English
Picture thousands of tiny dials arranged in layers. A number goes in one end, each dial nudges it a little, and an answer comes out the other end.
During training, the network tries, gets it wrong, and the dials are adjusted a tiny bit each time. After millions of tries, the dials are set so the answers come out right.
Stack enough layers together and you get "deep learning", which is what powers chatbots, voice assistants and photo apps.
Where it came from
Warren McCulloch and Walter Pitts described a maths model of brain cells in 1943. In 1958 the psychologist Frank Rosenblatt built the perceptron, one of the first neural networks that actually ran.
See it in action
When you say "Hey Siri", a small neural network on your phone listens for that sound pattern and wakes the assistant only when it's fairly sure it heard it.
What people get wrong
- It isn't a digital brain. Real brain cells are far more complicated, and nobody fully understands how brains learn.
- People who build them can't always explain why a network gave a particular answer. That's an open problem, not a solved one.
Go deeper
- ArticleWikipediaNeural network (machine learning)The full story, from McCulloch and Pitts in 1943 to deep learning.
- Video3Blue1BrownBut what is a neural network? | Deep learning chapter 1The best visual explanation there is. Twenty minutes and you'll actually get it.
- ArticleWikipediaPerceptronRosenblatt's 1958 machine, one of the first working neural networks.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.