Deep learning: the technique behind today's AI
Deep learning is a type of machine learning that uses neural networks with many layers. It powers AI assistants, the voice recognition on your phone and image generators.
Data as of October 22, 2026
Key figures
| Level | Beginner |
|---|---|
| Origin | 2012 |
| Introduced by | Geoffrey Hinton et ses étudiants (compétition ImageNet) ; prix Turing 2018 à Bengio, Hinton et LeCun |
Layers that learn to see and read
Deep learning is a branch of machine learning. It relies on neural networks organized into many layers. In a photo, the first layers pick up edges, the next ones shapes, then whole objects. With text, it works in a similar way: successive layers capture words, their meaning in context, then the intent of the sentence.
The big change is that nobody writes these steps by hand. The network figures out for itself what to look for, from examples.
The 2012 turning point
The theory goes back to the 1980s. The public breakthrough came in 2012: according to the ACM, Geoffrey Hinton and his students nearly halved the error rate for object recognition in the ImageNet competition. In 2019, the ACM gave the 2018 Turing Award to Yoshua Bengio, Geoffrey Hinton and Yann LeCun, calling them the fathers of the deep learning revolution (ACM announcement).
Since then, the same approach has taken over translation, voice, and then text with the Transformer architecture, which gave rise to large language models.
What it means for you, day to day
- Your phone transcribes a voice message into text.
- A photo app finds all your beach pictures.
- An AI assistant drafts a complaint letter from three lines of notes.
- Your video-call software blurs the background behind you.
All of these rely on deep learning. Knowing this helps you keep realistic expectations: these systems are excellent at recognizing patterns, and less reliable when a situation falls outside what they saw during training.
Common misconceptions
“Deep learning means the AI understands things deeply.” No, “deep” describes the stack of layers, not the depth of understanding.
“It’s a brand-new technology.” The principles are old; what is recent is the computing power (the GPUs) and the volume of data that make them effective.
“Deeper is always better.” Adding layers takes more data and energy, with no guarantee of better results.
To see these dates in the wider history of AI, open the AI timeline.
Related terms
- Neural network · Software made of many small connected computing units that learns by adjusting the strength of the links between them.
- Machine learning · A branch of AI where software learns patterns from examples instead of following rules written one by one by a person.
- GPU (graphics processing unit) · A chip built to run thousands of calculations in parallel, which became the engine for training and running AI models.
- Transformer · A neural network architecture built on attention, introduced in 2017, that underpins almost every large language model today.
- Large language model (LLM) · A program trained on huge amounts of text to predict what comes next in a text; it is the engine behind AI assistants.
- Computer vision · The branch of AI that lets machines analyze images and video to recognize objects, text, faces or scenes.
Try it
Full glossary
Frequently asked questions
Why is it called “deep”?
The word refers to the number of layers in the neural network. A “deep” network stacks many layers, each building a slightly more abstract representation than the one before. It has nothing to do with “deep” thinking in the human sense.
Who are the pioneers of deep learning?
Yoshua Bengio, Geoffrey Hinton and Yann LeCun received the 2018 ACM Turing Award for breakthroughs that made deep neural networks a critical part of computing. The award was announced in March 2019.
Why did deep learning take off so late?
The ideas had been around for decades, but two ingredients were missing: very large amounts of digital data and enough computing power, notably thanks to GPUs. Together, they finally made these networks work well.
Is deep learning the only kind of AI?
No. Other machine learning methods are still widely used, for example on spreadsheets of numbers in finance or logistics. But for language, images and voice, deep learning dominates.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.