What it means
Deep learning is the branch of machine learning that stacks many layers of artificial neurons, so each layer transforms the output of the one before it. "Deep" simply refers to that depth.
The payoff is that the system learns its own features. Older approaches required experts to hand-specify what to look for — edges, shapes, keywords. A deep network discovers useful intermediate representations on its own: early layers might respond to edges, later ones to textures, later still to whole objects. Nobody programs that hierarchy; it emerges from training.
The ideas are decades old. What changed was the arrival of enough data and enough parallel compute to train large networks, plus architectural advances — most consequentially the transformer — that made depth pay off for language.
Why it matters
The self-taught-features property is the whole reason AI generalizes to messy real-world inputs where rule-writing failed. It is also the root of the interpretability problem: because nobody specified the intermediate representations, nobody can straightforwardly read them back out to explain a decision.
In practice
For most people deep learning is infrastructure rather than a choice — it's what the models you use are made of. It becomes a live decision only when you're building models rather than applying them.