What counts as artificial intelligence? One answer is that AI is machines and computer programs that can imitate smart human behavior, such as making predictions and decisions. John McCarthy wrote an early FAQ on the subject. In it he wrote that AI is the science and engineering of making intelligent machines, especially intelligent computer programs. He also wrote that AI does not have to confine itself to methods that are biologically observable. Today AI and machine learning are sometimes used as near-synonyms. Most of the current advances in AI have involved machine learning. An AI program does not think the way a person does.

A computer can be programmed with exact steps, the way a baker follows a recipe. Or it can be set up to learn from data. The term machine learning goes back to the 1950s, when it was defined as the field of study that gives computers the ability to learn without being explicitly programmed. Some tasks are easy for a person but hard to explain to a computer, such as telling pictures of different people apart. A program can be good at one narrow job and still be unable to do a different one. A chess program, for example, would struggle with a new variant of chess.

In 1950 the mathematician Alan Turing published a paper that asked, "Can machines think?" In the game he described, a person asking questions tries to tell which of two hidden players is the machine and which is the person. The answers are typed. Turing predicted that in about fifty years computers could play this game well enough that an average questioner would have no more than a 70 percent chance of picking the right answer after five minutes. The name "artificial intelligence" comes from a 1955 proposal for a summer project at Dartmouth College in New Hampshire. The proposal guessed that every part of learning could be described so exactly that a machine could simulate it.

The field of AI is said to have started officially in 1956, at a summer conference at Dartmouth College. Historians note that AI work was already going on before 1956. So the starting date is a matter of how the field is defined. The first trainable neural network, the Perceptron, was shown in 1957 by a Cornell psychologist. It had one layer of adjustable weights between its input and output. In July 1958 the U.S. Office of Naval Research unveiled it. A newspaper headline said it "learns by doing," and skeptics insisted it could not reshape the relationship between humans and machines.

In 1958 a room-sized computer taught itself to tell cards marked on the left from cards marked on the right. It needed 50 tries. A neural network is made of many simple parts, called nodes, joined together, and it is modeled loosely on the human brain. Most of today's neural networks are organized into layers of nodes. When a network is trained, the numbers that set how strongly its parts connect are adjusted until it gives the right answers for the training examples.

Machine learning is one way to make AI: the computer finds patterns in data. In supervised learning, a program is trained on examples that people have already labeled. In unsupervised learning, a program looks for patterns in data that has no labels. In reinforcement learning, a machine learns by trial and error, guided by a reward system.

Generative AI can be thought of as a machine-learning model trained to create new data. It does more than make a prediction about a given dataset. A large language model, the kind of AI behind many chatbots, is trained on text from sources across the internet, such as books and web forum discussions. It works by predicting which word is likely to come next, a bit like autocomplete but on a much larger scale. It breaks what you type into small pieces called tokens. Because it picks words by probability, it can produce a false statement that sounds right. AI mistakes are sometimes called hallucinations.

Spam can be flagged in two ways. Heuristics sort by common characteristics. Neural networks can flag messages that fit a general spam profile they have learned. A spam filter can wrongly flag a good email as spam, which is called a false positive. If the examples an AI learns from leave some people out, the AI can work worse for those people. AI can also be used to make decisions that affect people, such as whether a person is admitted to a school or approved for a loan.

A common way to check an AI's claims is to compare them with other reliable sources: look for facts that other pages confirm. AI can be wrong even when it sounds sure, so a confident answer is not proof. AI chatbots can make up facts and even make up the sources they name. If a chatbot names a book or article, a person should search to confirm that it really exists. An AI chatbot may not know about very recent events, because it can only use the information it learned from. An AI should not be asked to check its own work: a person should look for outside sources.

Words to know

• perceptron: first trainable network
• heuristics: rules of thumb
• generative: that creates
• simulate: act like
• hallucinations: AI mistakes