What is AI, really?

When people say Artificial Intelligence (AI) today, they usually mean chatbots like ChatGPT and Claude, image generators, or deepfakes. But AI existed long before these tools went mainstream, and you have almost certainly been using it without calling it AI.

Many spam filters, the films Netflix suggests to you, the route Google Maps predicts, the face scan to unlock your phone, an automatic translation of a foreign website. All of these are examples of AI systems.

Try it

A quick gut check

Which of these statements do you think is most accurate?

A simple definition

Artificial intelligence is a broad term for computer systems designed to perform tasks that require skills we associate with intelligence. Like recognizing patterns, making predictions, choosing actions, or generating content.

It is also important to know what the definition does not say. It does not mean that the system thinks like a human, understands everything it processes, is conscious, or is equally capable across tasks. A system can be extremely good at spotting whether an email is spam, but completely unable to hold a simple conversation.

One umbrella, many forms

Under the AI umbrella live very different approaches. Rule-based AI systems follow hand-written logic, while machine-learning systems learn patterns from examples you give the machine. Within machine learning there are even more branches. Systems for vision, for language, for prediction, and for generation.

Artificial intelligencean umbrella over everything below
- Rule-based systemslogic written by people
- Machine learningpatterns learned from examples
  - Supervisedlearns from labelled examples
  - Unsupervisedfinds structure without labels
  - Reinforcementlearns by trial and feedback

Deep learning doesn’t sit on one branch: it is a set of techniques using layered neural networks, used across machine learning, and what most modern vision, language and generative systems are built on.

Every box here gets called “AI”. That is why the label on its own tells you so little about what you are actually looking at.

You don’t need to understand the full family tree. The one idea worth keeping is that different AI systems are built for different jobs. The question, rather than “is it AI”, is “what job is this AI system built to do?”

Curious?Curious about the deeper family tree?

Rule-based systems follow logic written directly by people: “if the email contains these words, flag it.” Machine-learning systems instead learn patterns from many examples. Three major approaches to machine learning are supervised learning, unsupervised learning, and reinforcement learning. Supervised learning learns from labelled examples (this email is spam, this one isn’t), unsupervised learning finds structure without labels, and reinforcement learning learns from trials and feedback.

Deep learning is a family of machine-learning techniques using layered neural networks; it powers many of today’s leading vision and language systems, including the generative models behind chatbots and image tools.

Try it

Two ways to catch spam

The same job, done twice. First with rules a person wrote, then by a system that only ever saw examples.

A person wrote these rules. The computer follows them exactly.

Congratulations! You are our WINNER. CLICK HERE to claim your FREE prize.

Contains “FREE” → +1 spam point
Contains “WINNER” → +1 spam point
Contains “CLICK HERE” → +1 spam point
2 or more points → call it spam

Try it

What is this AI built to do?

Group AI by purpose rather than technology. One system at a time: pick the job you think it was built to do, then try again or move on.

1 of 8

Your phone unlocking when it sees your face

The line that matters most right now: generative AI vs everything else

Most AI you have used for years (spam filters, recommendation systems, fraud detection) classify, predict, rank, or detect. These systems typically analyze inputs to make a prediction or decision rather than generate open-ended new content. Generative AI is the branch that has become much more prominent recently. It produces new content by learning patterns from existing data: text, images, audio, video, code, and more.

New content does not mean independence. Generated content is built from patterns learned from often very large amounts of pre-existing data.

The similar underlying technology, different consequences

Is AI good or bad? That question is like asking whether engines are good or bad. Similar image-recognition technology can be used in the medical field to detect signs of skin cancer, or for facial recognition and tracking people without their consent. Generative AI can give a voice to people who can’t speak, and can manufacture convincing misinformation. Recommendation systems can help you discover music you love, but can be tuned to keep you scrolling until late into the night.

So judging “AI” as a whole is too ambiguous. The useful questions are about each specific system: What is its goal? Who controls the system? What data does it use? Who benefits, and who can be harmed by the system? What happens when the AI is wrong, and can anyone challenge the outcome?

Try it

Generative or not?

One more pass with your new lens, one system at a time: does this create new content, or work with what already exists?

1 of 6

A tool that writes product descriptions from bullet points