How long has AI been around? 70 years of booms, winters and breakthroughs (2026)

In the summer of 1956, a handful of researchers gathered at Dartmouth College in New Hampshire to spend several weeks on a question nobody had a name for until the year before. So, how long has AI been around? As a named field of research, artificial intelligence is about 70 years old: the term was coined in a 1955 proposal and the field formally began at that 1956 Dartmouth workshop. AI only became popular with the general public much more recently, after ChatGPT launched in late November 2022.

That gap between "invented" and "popular" is the most interesting part of the story. For most of its life, AI was a research topic that went through bursts of excitement and long, cold stretches of disappointment. What you use today is the latest chapter of a very long book, and it helps to know how the earlier chapters went.

Short answer: how old is AI?

AI as a field dates to 1955–1956, which makes it roughly 70 years old in 2026. The ideas behind it are a little older: Alan Turing asked "Can machines think?" in a 1950 paper in the journal Mind, and the Open University notes that serious research into thinking machines began in the mid-1940s. AI "came out" for everyday users in stages: voice assistants and recommendations in the 2010s, then chatbots like ChatGPT from 2022.

If you want a single date to remember, use 1956. If you want the date AI became part of ordinary life, use 2022. Both are true, and they answer slightly different questions.

How it works: the background and the eras of AI

Before the timeline, a quick definition. Artificial intelligence means building computer systems that do things we normally associate with human thinking, such as understanding language, recognising images, making decisions or learning from examples. If you want the longer version, our explainer on what AI is covers it in plain English.

The people who started the field, including John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, are covered in detail in our guide to who created AI technology. Here we will focus on something different: how long each era lasted, and why progress came in waves rather than a straight line.

The founding era (1950s to late 1960s)

The Dartmouth proposal, written in 1955, made a bold claim: that "every aspect of learning or any other feature of intelligence can in principle be so precisely described" that a machine could simulate it. McCarthy picked the name "artificial intelligence" for it. The workshop itself ran in the summer of 1956, probably for around eight weeks rather than the six usually quoted.

Early results were genuinely impressive for the time. According to the Open University, within a few years computers were playing draughts and solving word and logic problems, and researchers had started on translation, speech-to-text and face recognition. In 1966, Joseph Weizenbaum published his work on ELIZA, a simple chatbot that imitated a therapist using pattern-matching rules. The Open University's summary puts it well: "Wherever ELIZA was demonstrated, it caused a sensation."

The first AI winter (1970s)

Optimism ran ahead of the hardware. Computers were slow, memory was tiny and the problems turned out to be far harder than the founders expected. In the UK, the government asked mathematician James Lighthill to review the field. His 1973 report formed the basis for ending support for AI research at most British universities, which contributed to what became known as the first "AI winter": a period of shrinking funding and fading interest.

The expert systems boom and the second winter (1980s to early 1990s)

AI came back in the 1980s in a more practical form called expert systems, which encoded human expertise as long lists of if-then rules. XCON, often called the first successful expert system, helped configure computer orders. But, as Giuliano Giacaglia explains in Making Things Think, these systems were expensive to maintain because they were hard to update and could not learn. When the market for specialised AI hardware collapsed in 1987, a second winter set in, lasting roughly until 1993.

The quiet machine learning years (1990s to 2011)

This is the era people tend to skip, and it matters. Researchers shifted from hand-written rules to statistics and probability, letting systems learn patterns from data. The public moment of the period came in 1997, when IBM's Deep Blue beat world chess champion Garry Kasparov in a six-game rematch, a year after losing to him. IEEE Spectrum reports that the upgraded machine could search up to 200 million options per second. Deep Blue was brute-force calculation, not learning, but it showed the world that computers could beat humans at a task once seen as a test of intellect.

The deep learning era (2012 to 2021)

The modern boom has a fairly clear starting point. In 2012, University of Toronto researchers Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton built AlexNet, an image-recognition system trained on the huge ImageNet dataset. The Computer History Museum describes it as the first system to combine deep neural networks, large datasets and GPU computing, and notes that before AlexNet almost none of the leading computer vision papers used neural networks, while afterwards almost all did.

Five years later, a Google team published "Attention Is All You Need" (June 2017), introducing the Transformer, a design "based solely on attention mechanisms." Nearly every large language model since is built on that idea. If the difference between these terms is fuzzy, our comparison of AI vs machine learning vs deep learning untangles them.

The generative AI era (2022 onwards)

ChatGPT was made free to the public in late November 2022. A UBS study, reported in February 2023, estimated it had reached 100 million monthly active users in January, about two months after launch. For comparison, the same report said TikTok took about nine months and Instagram about two and a half years. This is when most people would say AI "came out." By 2026, Stanford's AI Index found generative AI had reached 53% adoption within three years, faster than the internet or the personal computer at the same stage.

The dates at the edges are approximate. Historians argue about exactly when each winter began and ended, so treat the table as a map, not a contract.

Real examples: AI's 70 years squeezed into one calendar year

Long timelines are hard to feel. So here is an analogy we worked out ourselves. Imagine the whole history of AI, from the Dartmouth summer of 1956 to today in October 2026, squeezed into a single calendar year. Dartmouth happens at midnight on 1 January. Today is midnight on 31 December.

On that scale, ELIZA arrives around 19 February. The Lighthill report lands at the end of March. The second AI winter begins in mid-June. Deep Blue beats Kasparov on about 1 August. AlexNet does not show up until around 20 October, and the Transformer paper arrives in mid-November. ChatGPT launches around 12 December, less than three weeks before the year ends.

Put another way, almost everything you think of as "AI" today, the chatbots, the image generators, the coding assistants, happened in the last three weeks of a year-long story. That is why the field can feel brand new and very old at the same time. Both impressions are correct, depending on whether you are looking at the ideas or at the products.

The examples you meet every day also come from different eras. Spam filters and product recommendations are descendants of the statistical machine learning years. Face unlock on your phone and photo search grew out of the deep learning era that AlexNet started. Chatbots that write and summarise are products of the Transformer. Our overview of the types of AI shows how these families fit together.

Benefits and limitations of knowing AI's long history

Why does any of this matter if you just want to use the tools? We think the history offers two useful lessons, and they pull in opposite directions.

The first is reassurance. AI is not a fad that appeared from nowhere in 2022. It rests on seven decades of research, and many of its core ideas, such as neural networks and learning from data, have been tested, abandoned and revived several times. The current wave is built on more data, cheaper computing and better methods than any before it. Stanford's 2026 report, as summarised by Burges Salmon, puts organisational adoption at 88%, which is hard to dismiss as hype.

The second lesson is caution. Every previous boom came with promises that machines would match human intelligence within a decade or two, and every one was followed by disappointment. The founders in 1955 hoped a single summer of focused work could make real headway on problems that are still open today. The expert systems of the 1980s were sold as a revolution and ended up too costly to maintain. Today's systems are far more capable, but they still make confident mistakes, and the AI Index recorded a record 362 reported AI incidents in 2025.

Where do we land? The history suggests that the technology itself rarely disappears after a boom. What collapses is inflated expectations. Rule-based systems did not vanish after the second winter; they quietly became ordinary software. It is reasonable to expect something similar today: some of the excitement will fade, and much of the technology will simply become part of everything.

What it means for you: practical takeaways

The first practical point is to judge AI tools by what they do now, not by the label. Something marketed as "new AI" may be decades-old machine learning with a fresh coat of paint, and that is not necessarily bad. Mature techniques are often the most reliable.

The second is to keep a healthy margin of doubt with generative tools in particular. They belong to the youngest era of the field, the one with the least track record. Check facts, especially for anything involving health, money or law. If you write prompts regularly, it is worth understanding how large language models actually produce their answers, because it explains why they sometimes invent things.

The third is that the pace of change is real, but uneven. Generative AI spread faster than the PC or the internet according to Stanford, yet many workplaces are still working out where it fits. You do not need to adopt every new tool the week it launches. History rewards people who learn the fundamentals and pick tools that solve an actual problem.

What happens next

Here we need to separate what we know from what we are guessing.

What we know: investment and capability are both rising quickly. The 2026 AI Index reports US AI investment of about $285.9 billion in 2025, and shows AI systems on the OSWorld computer-use benchmark improving from about 5% accuracy in 2024 to 66.3% in 2025, close to the human baseline of 72.35%. That benchmark measures how well AI can operate a computer on its own, which is the foundation of what people now call agentic AI.

What we are guessing: whether this boom keeps going, levels off or runs into a third winter. The honest answer is that nobody knows. The conditions that caused earlier winters, mainly promises outrunning results and funding drying up, could return if expensive systems fail to pay their way. On the other hand, today's AI is already earning money and embedded in products used by hundreds of millions of people, which was never true in 1973 or 1987.

Our best reading of 70 years of history is this: expect the hype cycle to cool at some point, but do not expect the technology to go away. The field has survived two winters already. It is likely to outlast whatever comes next.

FAQ

When did AI become popular?

AI became popular with the general public after OpenAI released ChatGPT in late November 2022. A UBS study estimated it reached 100 million monthly users within about two months. Earlier peaks of public interest came with Deep Blue's chess win in 1997 and voice assistants in the 2010s, but nothing matched the 2022 surge.

When was AI created?

AI was created as a research field in 1955–1956. John McCarthy coined the term "artificial intelligence" in a 1955 proposal, and the Dartmouth Summer Research Project in 1956 is widely treated as the field's founding event. Alan Turing's 1950 paper "Computing Machinery and Intelligence" laid important groundwork a few years earlier.

When did AI come out?

There was no single release date. Early AI programs ran in research labs from the late 1950s, ELIZA appeared in 1966, and businesses used expert systems in the 1980s. For consumers, AI arrived gradually through search, recommendations and voice assistants, then went mainstream when ChatGPT launched to the public in November 2022.

How old is AI?

AI is about 70 years old in 2026, counting from the 1956 Dartmouth workshop where the field formally began. If you count from Alan Turing's 1950 paper, it is around 76 years old. The modern deep learning era is only about 14 years old, and generative AI chatbots have been widely available for about four years.

When was AI first invented?

Most historians point to 1955–1956 as the moment AI was first invented as a discipline, when McCarthy, Minsky, Rochester and Shannon proposed the Dartmouth project. The idea of thinking machines is older, with serious research starting in the mid-1940s, but 1956 is when AI got its name and its first dedicated research programme.