What does AI actually learn?
Everyone says AI is "trained". But when AI learns, what actually changes inside it? The answer: billions of tiny knobs.
EXAMPLE
Go out for coffee?
Picture a small decision: should I go out for coffee this afternoon? You think about three things: is it raining, did a friend invite you, and do you have money left?
WEIGHTS
One knob for each thing
Each thing has its own knob, called a weight. A friend's invite pushes hard toward "go", rain pulls the other way. Add them all up, and if it passes a threshold, you decide: go.
NEURON
An artificial neuron
That is an artificial neuron: it takes a few numbers, multiplies them by weights, adds them up, then decides. Surprisingly simple.
NEURAL NETWORK
Stacked into layers
Stack thousands of neurons into layers, each layer passing its results to the next, and you have a neural network. The first layer spots lines, the middle layers spot ears and eyes, and the last layer says: cat.
LEARNING
Learning means turning the knobs itself
Big chatbots have hundreds of billions of these knobs, and nobody turns them by hand. "Learning" means the machine turns each knob itself, a little at a time, until it gets the right answer.
PART 2
Which knob, which way?
But how does the machine know which knob to turn, and which way? First, it has to know how wrong it is. See you in part 2.
This article is based on the video Billions of tiny knobs from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.