B1 · Introduction to Artificial Intelligence

Definitions of AI by Its Pioneers

There's no single agreed definition of AI. The main ones fall along two questions: is the goal to think or to act, and should the standard be human behaviour or rational behaviour? Russell and Norvig's textbook, the standard reference, defines AI as the study of agents that act rationally.

10 min Beginner Jump to the activity

What you need: Nothing. This page starts from the beginning, with no programming and no formulas.

Definition

Ask ten experts what AI is and you may get ten answers. They're all looking at the same elephant from different sides.

There's no single agreed definition of AI. The main ones fall along two questions: is the goal to think or to act, and should the standard be human behaviour or rational behaviour? Russell and Norvig's textbook, the standard reference, defines AI as the study of agents that act rationally.

In simple words

One group says AI should copy how people think. Another says just make it behave like a person from the outside. A third says forget people, make it reason correctly. The fourth, which most engineers follow today, says it should simply take the best action available to reach its goal.

Why it matters

Definitions decide what gets built, what counts as success and what the law covers. If someone claims their product is AI, or that AI has reached human level, you can ask: by which definition?

Key terms

Definition
A precise statement of what a word means and what it does not mean.
Thinking humanly
The approach that tries to make machines think the way people actually think, studied with psychology and brain science.
Acting humanly
The approach that judges a machine by whether its behaviour looks human, as in the Turing Test.
Thinking rationally
The approach that builds machines which reason by correct logic.
Acting rationally
The approach that builds agents which act to achieve the best expected result. Most AI today follows this view.
Rational agent
Something that perceives its surroundings and acts to do as well as possible towards its goal.

Explained in detail

Why there are so many definitions

Ask ten experts what AI is and you will hear several answers. This is not because they are confused. It is because "intelligence" itself is hard to pin down, and because each researcher cares about a different goal. A psychologist wants to understand human thinking. An engineer wants machines that work well. A lawmaker needs a definition precise enough to write rules. Each definition is a lens that shows one side of the subject clearly.

The founders

Alan Turing, in 1950, avoided defining thinking at all. He asked instead whether a machine could hold a conversation so well that a judge could not tell it from a person. John McCarthy, who coined the term in 1955, later described AI as "the science and engineering of making intelligent machines, especially intelligent computer programs". Marvin Minsky described it as making machines do things that would need intelligence if people did them. The computer scientist Elaine Rich, in 1983, gave a playful and honest version: AI is the study of how to make computers do things at which, at the moment, people are better.

The four approaches

The most widely used AI textbook, by Stuart Russell and Peter Norvig, sorts definitions using two questions. First, should AI be judged by how it thinks or by how it acts? Second, should the standard be human behaviour, with all its quirks, or ideal, correct behaviour (rationality)? The answers give four approaches: thinking humanly, acting humanly, thinking rationally and acting rationally.

Most modern AI follows the last one, acting rationally. Engineers build an agent with a goal and judge it by results: does the route planner find a fast route, does the spam filter catch spam, does the robot reach the shelf safely? It does not matter whether the machine thinks like a person, only whether it acts well.

Definitions written for laws

Governments now need definitions they can use in law. The European Union's AI Act of 2024, and the OECD, an international organisation of countries, describe an AI system as a machine-based system that, from the input it receives, works out how to produce outputs such as predictions, content, recommendations or decisions, which can affect the physical or digital world. Notice that this definition says nothing about thinking or feelings. It focuses on what the system does and what effects it can have, because that is what laws need to control.

How it works

The four approaches (Russell & Norvig)

Human standardRational standard
ThinkingThink like humans: model how people think (cognitive science).Think rationally: follow correct logic.
ActingAct like humans: pass as a person (the Turing Test).Act rationally: take the best action for the goal. This is the modern view.

Definitions from the founders and regulators

WhoDefinitionFocus
Alan Turing, 1950If a machine can pass as human in conversation, we should grant that it thinks.Behaviour
John McCarthy, 1955The science and engineering of making intelligent machines.Engineering
Marvin MinskyMaking machines do things that would require intelligence if done by people.Tasks
EU AI Act, 2024A machine-based system that infers from its input how to produce outputs such as predictions, content, recommendations or decisions.Law
OECD, 2023Close to the EU wording. Systems vary in how autonomous and adaptive they are.Policy

Try it yourself

Four ways to define AI

The experts asked different questions, so they got different definitions. Travel through them.

Who said it?

Match each idea about AI to the person or book it comes from.

  • "Can machines think?" Let's test it with an imitation game.
  • AI is "the science and engineering of making intelligent machines".
  • AI is making machines do things that would need intelligence if people did them.
  • AI is the study of agents that sense their environment and act to do the best they can.

The four ways of looking at AI

Textbooks group definitions into four views. Put each goal in the right view.

  • Chat so well that people can't tell it's a machine

  • Copy the way people solve puzzles, mistakes included

  • Reach only conclusions that follow by strict logic

  • A delivery robot choosing the fastest safe route

Common mistakes

  • Thinking only one definition is right. They answer different questions.
  • Treating passing the Turing Test as proof of real understanding. It tests behaviour only.

Check yourself

Question 1

1. Which definition of AI do most modern textbooks use?

Question 2

2. The Turing Test belongs to which approach?

Summary

  • There's no single definition of AI.
  • Four approaches: think or act, like humans or rationally.
  • Engineers and textbooks mostly use the rational-agent definition, and laws use the broad system definition.

Finished reading?

Sources: Russell & Norvig, Artificial Intelligence: A Modern Approach · EU AI Act, Article 3