Artificial intelligence
In brief
Artificial intelligence refers to programs able to perform tasks that seemed to require human intelligence, most often by learning from examples rather than following hand-written rules.
Definition
Artificial intelligence (AI) is a field of computer science that aims to have machines perform tasks such as recognising an image, understanding a sentence, translating, playing games or making a decision. The term was already used in 1955 in the Dartmouth conference proposal (McCarthy, Minsky, Rochester, Shannon), but its consumer uses took off from 2022, with the release of ChatGPT in November.
The main families
- Rule-based AI: a human writes the rules (“if… then…”). Reliable but rigid.
- Machine learning: the program learns the rules itself from examples. It is the basis of most current systems.
- Deep learning: a form of machine learning that uses neural networks made of several successive layers. It brought a leap in image and speech recognition in the 2010s.
- Generative AI: deep learning models that produce content (text, images, sound, video). ChatGPT, Midjourney and Suno are examples.
How an AI “learns”
The model is shown a very large number of examples, and its internal parameters are gradually adjusted so that it makes fewer and fewer mistakes. Once trained, it applies what it has learned to new situations. It does not understand the world as a human does: it exploits statistical patterns, which explains both its performance and its errors.
Not to be confused with
- Generative AI and AI in general: AI has been around for a long time (recommendations, spam filters, route planning). Generative AI is a recent branch of it.
- Today’s AI and “general AI”: current systems are very good at certain tasks, but the idea of an AI as versatile as a human (AGI) remains a debated goal, not an established reality.
Spotted a mistake or have an idea about this entry? Let us know