"Artificial intelligence" is an umbrella term covering decades of research and a wide range of techniques. Sorting out what the phrase actually refers to makes most modern AI coverage easier to follow.
The Term Predates the Modern Wave
The phrase "artificial intelligence" was coined in 1956 at a summer workshop at Dartmouth College, organized by John McCarthy and others. The original ambition was broad: to build machines that could perform tasks generally associated with human intelligence, such as reasoning, learning, planning, and language use.
For most of its history, AI progressed in fits and starts. Early enthusiasm in the 1950s and 1960s gave way to a long stretch known as the "AI winter," when progress slowed and funding dried up. Pockets of useful work continued — in expert systems, search algorithms, computer vision, and speech recognition — but the field did not deliver on its broadest promises.
Machine Learning, the Workhorse Underneath
Most of what is currently called AI is, more specifically, machine learning. Machine learning is a family of techniques in which a computer program improves its performance on a task by being exposed to data, rather than by being given an explicit rule set.
A traditional program might recognize spam email by following hand-coded rules: if it contains certain phrases, it is spam. A machine-learning system, by contrast, is given a large set of emails labeled "spam" or "not spam" and learns the statistical patterns associated with each category. Once trained, it can classify new emails it has never seen.
Machine learning has been used commercially for decades — in fraud detection, search ranking, recommendation systems, and dozens of other applications most users never see.
Deep Learning and Neural Networks
The current wave of AI is built on deep learning, a subset of machine learning that uses artificial neural networks with many layers. Neural networks are inspired loosely by the structure of biological brains, but they are not models of how brains actually work. They are mathematical functions with millions or billions of adjustable parameters that get tuned during training.
Three things came together to make deep learning practical at large scale: enormous datasets, much faster computer hardware (especially graphics processing units), and incremental algorithmic improvements. By the early 2010s, deep neural networks were setting records on tasks like image classification and speech recognition. By the late 2010s and early 2020s, they were producing usable systems for translation, image generation, and natural-language interaction.
What Today's Systems Can and Can't Do
Modern AI systems are very good at certain pattern-recognition and pattern-generation tasks: classifying images, transcribing speech, translating between languages, and producing fluent text and images on demand. They are also notoriously good at producing confident-sounding output that is wrong, a behavior often described as "hallucination" in the context of language models.
It is worth being precise about what they are not. They do not have understanding in the way humans do; they do not have goals, preferences, or beliefs in any meaningful sense; and they do not "know" things the way a person knows them. They produce statistically plausible outputs given their training data and the input they receive. That capability is genuinely powerful and genuinely useful — but it is not the same as general intelligence.
Why It Matters Now
The reason AI has become a public topic in the last few years is straightforward. The same techniques that quietly powered backend systems for years now drive consumer-facing products that anyone can use through a web browser. The underlying ideas have been developed over decades. The accessibility, speed, and quality of the user experience are what changed.
Understanding the distinction between the broad term ("artificial intelligence"), the underlying methodology ("machine learning"), and the current dominant technique ("deep learning with neural networks") makes it considerably easier to read past the marketing and follow what is actually being claimed.
This article is for general informational and educational purposes only.