A Brief History of Artificial Intelligence

The question that started it all (1950)

In 1950 the British mathematician Alan Turing published a paper called Computing Machinery and Intelligence. Instead of asking whether a machine could think, which is hard to define, he proposed a practical test: if a person chatting through text cannot reliably tell whether they are talking to a human or a machine, then we may as well say the machine is thinking. This is now known as the Turing test.

Turing wrote this before the word for the field even existed. He was working with room-sized computers that had a tiny fraction of the memory in a modern phone. What makes the paper remarkable is how many objections to machine intelligence he anticipated, many of which people still raise today.

A name and a field (1956)

The field got its name in 1956 at a summer workshop held at Dartmouth College in the United States. The proposal was put together by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, and the term artificial intelligence is credited to McCarthy. The gathering set the agenda for decades of research.

Early results looked promising. The Logic Theorist, built by Allen Newell and Herbert Simon around 1955 and 1956, could prove mathematical theorems. A few years later, in 1959, Arthur Samuel described a checkers-playing program that improved with experience, an early example of what we now call machine learning. Optimism ran high, and some researchers predicted human-level intelligence within a generation.

The AI winters

The early predictions did not come true. The problems turned out to be far harder than expected, and computers of the era were simply too slow and too small. In 1973 a report by the mathematician James Lighthill criticised the lack of progress, and funding in the United Kingdom and the United States was sharply cut. This period of reduced interest and money became known as the first AI winter.

Interest revived in the 1980s with expert systems, programs that captured the knowledge of human specialists as long lists of if-then rules. Companies invested heavily, and the AI industry grew from a few million dollars at the start of the decade to billions by 1988. But these systems were expensive to maintain and brittle outside their narrow domains, and a second slowdown followed around the early 1990s.

Learning from data instead of rules

The lasting shift was a change in approach. Rather than hand-coding rules, researchers leaned into machine learning, where a system finds patterns in large amounts of data. This was quietly powerful, but it needed two things that arrived in the 2000s: huge datasets from the internet, and fast graphics processors that could do the heavy maths cheaply.

The turning point came in 2012. A neural network called AlexNet, built by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton, won the ImageNet image-recognition contest with a top-5 error rate of 15.3 percent, far ahead of the runner-up at around 26 percent. It showed that deep neural networks, given enough data and computing power, could beat older methods by a wide margin. The modern deep learning era had begun.

Transformers and the generative wave (2017 onward)

In 2017 a team of researchers at Google published a paper titled Attention Is All You Need, which introduced the transformer architecture. Transformers process whole sequences at once using an attention mechanism that weighs how much each part of the input relates to every other part. This made models far easier to train at scale.

Transformers became the foundation for large language models and image generators, the technology behind the generative AI tools that reached the wider public from late 2022. It is worth keeping perspective: these systems are very capable pattern learners trained on enormous datasets, not conscious minds. Understanding where they came from helps you judge what they can and cannot reliably do.

That same family of vision and image models now powers practical, everyday tools. Renderivo, for example, uses visual AI to clean up product photos for online sellers, removing busy backgrounds and producing tidy white-background, square images. It is a small, concrete result of a seventy-year arc that started with one mathematician asking whether a machine could think.