Many tiny knobs, turned together
A single artificial neuron does something trivial: it adds up its inputs, each scaled by a weight, and passes the result through a simple function. The power comes from wiring millions of these together in layers and adjusting every weight at once. Training nudges those weights — guided by gradient descent and backpropagation — so that each pass through the data makes the network's answers a little less wrong.
Stack enough layers and the network learns to represent things in stages: early layers might pick out edges in an image, later ones whole shapes, the last ones the answer. That stacking is exactly what gives deep learning its name, and it is why neural networks underpin most of modern machine learning.
Why the brain analogy is loose
The name borrows from biology, but an artificial neuron is a maths function, not a cell. The comparison is inspiration, not a claim about how brains actually work.