Start with what the term actually covers
Artificial intelligence is software that performs tasks we would normally say need human judgement: recognising a face, translating a sentence, predicting what you will want to watch next, or writing a paragraph that reads naturally.
AI does not think the way a person does. It studies huge amounts of data, finds patterns in it, and uses those patterns to produce an output such as a prediction, a classification, a piece of text or an image.
Rules versus learning
The easiest way to see what is new about modern AI is to compare it with ordinary software. A traditional program follows rules a person wrote: "if the email contains this phrase, mark it as spam." An AI system is shown thousands of examples of spam and not-spam and works out its own rules, including subtle ones nobody thought to write down. The next lesson, How AI Actually Learns, explains how.
AI, machine learning and deep learning aren't the same thing
The three terms are often used as if they meant the same thing, but they are nested, like "vehicle", "car" and "electric car".
- Artificial intelligence is the broad goal: building systems that behave intelligently.
- Machine learning is the main approach today: systems that improve by learning from examples instead of following hand-written rules.
- Deep learning is a technique within machine learning, built on layered networks loosely inspired by neurons. It powers most of the AI you have heard about in recent years, including chatbots and image generators.
A short timeline
- 1956: the term "artificial intelligence" is coined at a summer workshop at Dartmouth College.
- 2012: a deep neural network wins a major image-recognition competition by a wide margin, and deep learning takes off.
- 2017: researchers introduce the transformer design, the foundation of today's language models.
- Late 2022: ChatGPT is released and conversational AI reaches the general public.
How a system like this works, in four steps
Strip away the branding and most AI systems follow the same loop:
- Collect data. The system is given examples: text, images, numbers, past behaviour.
- Train the model. It studies those examples and picks up patterns and relationships.
- Make a prediction. It classifies, recommends or generates something new.
- Refine. Developers evaluate the results and improve the model over further rounds of training.
Not all AI is the same kind of AI
- Narrow AI is built for one job, such as spam filtering, translation or facial recognition. Essentially all AI in use today is narrow, even when it feels broad.
- General AI is a hypothetical system matching human ability across any task. It does not exist today.
- Generative AI creates new content such as text, images, audio and code. It has grown fastest in recent years. See Types of AI Models.
What AI is not good at
AI is strong at spotting patterns and producing fluent language. It is weak at knowing when it is wrong, at tasks needing up-to-the-minute facts it was never given, and at anything where the training data was thin or skewed. It can produce confident answers that are false, and it can reflect biases in the data it learned from. Treat its output as a strong first draft, not a verified fact. See Fact-Checking AI Output and Common AI Myths, Debunked.
Frequently asked
Is ChatGPT the same thing as artificial intelligence?
No — ChatGPT is one application built using AI. Artificial intelligence is the much broader field that also includes computer vision, robotics, and speech recognition.
Do I need to know how to code to understand AI?
Not to understand the concepts. Coding becomes useful later if you want to build your own AI applications, not to make sense of how the field works.
Can AI replace human jobs?
It can automate specific tasks within a job. Judgement, accountability, and context — deciding whether an AI output is actually right for the situation — still fall to a person.