What Is Artificial Intelligence?

• The world
• A computer
• A goal

What you will learn

• Say what artificial intelligence is in your own words.
• Compare machine intelligence with human intelligence.
• Explain what machine learning is and spot AI around us.

Before you start

No prior knowledge needed. Knowing how to use a phone or a computer is enough.

What does intelligence mean for a machine?

Intelligence is the computational part of the ability to achieve goals. John McCarthy gave this definition. A goal could be, for example, winning a game of chess.

People, many animals and some machines have this ability, in different kinds and degrees.

A Stanford study panel quotes Nils Nilsson, who describes intelligence as the quality that lets an entity function appropriately and with foresight in its environment. The same report notes that AI has no definition that is both precise and accepted by everyone.

• A person
• An animal
• A machine
• A goal in the world

Is AI the same as human intelligence?

AI sometimes simulates human intelligence, but not always, and not even usually. That is McCarthy's answer. AI researchers are also free to use methods that are not observed in people.

McCarthy answers no when asked whether intelligence is a single thing so that one can ask a yes-or-no question like "is this machine intelligent or not?". Intelligence involves mechanisms, and AI research has made computers carry out some of them and not others.

The Stanford panel places intelligence on a multi-dimensional spectrum. An arithmetic calculator and a human brain both sit on it, and they differ in scale, speed, degree of autonomy and generality.

• An arithmetic calculator
• A human brain
• Four scales: scale, speed, degree of autonomy, generality

How do computers learn from data?

An algorithm is a clearly specified mathematical process for computation, with rules that give a prescribed result when followed.

One way computers learn from data is machine learning: people train an algorithm to create a model that makes predictions or decisions based on data.

Before the generative AI boom, talk of AI typically meant machine-learning models that can learn to make a prediction from data.

Generative AI can be thought of as a machine-learning model trained to create new data, rather than to make a prediction about a specific dataset.

That distinction is about the task: a generative model aims to produce new content. It can still use next-word prediction as part of that process, as the Markov example below shows.

• Training data
• A machine-learning model
• A prediction or a decision

Where do we meet AI in daily life?

We meet AI in applications and devices that can see and identify objects, or understand and respond to human language. That is how IBM describes them.

The suggested next word while you type an email is a familiar case. According to MIT News, Markov models, a much simpler early example of generative AI, have long been used for next-word prediction such as autocomplete in an email program.

• Seeing and identifying objects
• Understanding and answering human language
• Predicting the next word as you type

Real-world example

Try this with an adult. Open a new email as a blank draft that you will not send to anyone, and type a few imaginary words without personal information. If the app suggests a next word, you are watching next-word prediction at work on your screen. Change the last word and see how the suggestion changes.

Common misunderstandings

AI is only whatever is new. The Stanford panel's report describes the "AI effect": once people get used to a technology, it stops being considered AI, and newer technology emerges.

A machine with a high IQ score is a useful machine. McCarthy writes that making computers that score high on IQ tests would be weakly correlated with their usefulness.

Generative AI is a completely separate kind of AI. MIT associate professor Phillip Isola says the same algorithms can often be used for both generative AI and other types of AI.

Frequently asked questions

How is machine learning different from artificial intelligence?

Machine learning sits inside AI. IBM describes machine learning as directly underneath AI in a series of nested concepts.

Where does the Stanford report place an arithmetic calculator?

The Stanford panel places the arithmetic calculator on the intelligence spectrum. It also says such simple devices bear little resemblance to today's AI.

When did artificial intelligence research begin?

After the Second World War, a number of people independently started to work on intelligent machines. The English mathematician Alan Turing may have been the first.

Summary

• AI is the science and engineering of making intelligent machines, and intelligence is the computational part of the ability to achieve goals.
• AI sometimes simulates human intelligence, but not always.
• Machine learning creates models by training an algorithm. Some AI applications can identify objects, understand language or predict content.