The History of AI: From Turing’s Question to ChatGPT and Beyond

Artificial intelligence feels brand new. But the field behind today’s chatbots is more than seventy years old — and it has already lived through booms, crashes and long “winters” where funding and interest almost disappeared.

That history is not just trivia. It explains today’s headlines. Every generation of AI has followed a similar pattern: a real breakthrough, huge promises, disappointment when the promises outrun the technology, and then quiet work that sets up the next breakthrough. If you understand that pattern, you can read the current AI boom with calmer eyes.

Here is the full story, era by era — what actually happened, why it mattered, and what people believed at the time that turned out to be wrong.

Key takeaways

• AI began as a question, not a product: in 1950 Alan Turing asked whether a machine could imitate human conversation well enough to fool a judge.
• The field got its name in a 1955 proposal for the 1956 Dartmouth workshop, where researchers expected rapid progress — and were far too optimistic.
• AI crashed twice (the “AI winters”) when expert systems and early promises failed to deliver, and funding collapsed.
• The modern era rests on three foundations: big data, powerful graphics processors (GPUs), and the 2017 Transformer architecture.
• ChatGPT’s launch on 30 November 2022 turned AI from a research topic into a household tool almost overnight.

1. The question: Turing and the birth of an idea (1950–1956)

In 1950, the British mathematician Alan Turing published a paper in the philosophy journal Mind called “Computing Machinery and Intelligence”. Instead of trying to define thinking, Turing proposed a practical test: if a person, communicating only through text, cannot reliably tell whether they are talking to a machine or a human, then the machine can be said to behave intelligently. This became known as the Turing Test.

Five years later, in 1955, four researchers — John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon — wrote a proposal for a summer research project at Dartmouth College. That proposal contained the first use of the term “artificial intelligence”, which McCarthy is credited with coining. The workshop itself took place in the summer of 1956, and it is generally treated as the founding event of AI as a field.

What people wrongly believed at the time: the Dartmouth generation expected machines to reach human-level intelligence within roughly a generation — some spoke of significant progress within two months of the workshop itself. The gap between that confidence and reality would define the next fifty years.

2. Early optimism and symbolic AI (1956–1979)

The first decades of AI were dominated by symbolic AI: the idea that intelligence could be built from rules, logic and symbols written by hand. Early programs proved geometry theorems, played checkers and solved algebra problems word by word.

The most famous program of the era was ELIZA, created by Joseph Weizenbaum at MIT in 1966. ELIZA imitated a psychotherapist: it matched patterns in what the user typed and reflected their own words back as questions. Type “I am sad” and ELIZA might reply, “Why do you say you are sad?” The program understood nothing — yet users confided in it as if it did. This tendency to project understanding onto a machine is still called the ELIZA effect.

In the 1970s, researchers at Stanford built MYCIN, an early expert system that helped diagnose blood infections and recommend antibiotics. MYCIN performed impressively in tests — in some evaluations, comparably to human specialists — but because of legal and ethical concerns about a computer making medical decisions, it was never used in real clinical practice.

What people wrongly believed at the time: that success in tiny, tidy worlds — a therapy script, a chess board, a narrow medical rulebook — meant the same approach would scale to the messy real world. It did not. Hand-written rules break when the world refuses to follow them.

3. The AI winters (1974–1993)

In 1973, the British mathematician Sir James Lighthill delivered a report to the UK’s science research authorities evaluating the state of AI. His conclusion was blunt: AI had failed to achieve its “grandiose objectives”. He pointed to a deep problem — combinatorial explosion: the number of possibilities a program must consider grows so fast that methods which work on toy problems grind to a halt on real ones. UK funding for AI research was cut sharply, and similar scepticism spread. The first AI winter had begun.

The field thawed briefly in the early 1980s with the commercial rise of expert systems — programs that captured a human expert’s rules of thumb for one narrow job. The star example was XCON, written at Carnegie Mellon for the Digital Equipment Corporation, which configured computer orders automatically and went into routine commercial use in 1980. By the mid-1980s expert systems were a real industry, and XCON was credited with saving DEC tens of millions of dollars a year.

But expert systems were brittle. They could not learn, could not handle situations outside their rules, and became enormously expensive to maintain — every new rule risked breaking old ones. From around 1987 the market collapsed, producing the second AI winter.

What people wrongly believed at the time: that the winters meant AI itself was a dead end. In fact, during the quiet years researchers kept developing the methods that would later power the modern boom — without headlines or money.

4. Machine learning rises (1993–2011)

Through the 1990s and 2000s, the centre of gravity shifted from hand-written rules to machine learning: instead of programming rules, researchers showed computers many examples and let algorithms find the patterns. Statistical methods quietly took over tasks like spam filtering, speech recognition and recommendation.

The era’s symbolic moment came on 11 May 1997, when IBM’s Deep Blue defeated the reigning world chess champion, Garry Kasparov, in a six-game rematch in New York — winning the match 3½ to 2½. Kasparov had beaten the machine the year before; the upgraded Deep Blue, able to evaluate around 200 million chess positions per second, won the rematch, making front pages worldwide.

Yet Deep Blue did not “think” the way its audience imagined. It won through enormous calculation speed plus chess knowledge programmed by humans — brilliance at one narrow task, not general intelligence.

What people wrongly believed at the time: many concluded that because a machine had mastered chess, general intelligence must be close. Deep Blue itself showed the opposite: a system can be superhuman at one task and unable to do anything else.

5. The deep learning decade (2012–2019)

The modern era began in 2012. AlexNet, a deep neural network built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton at the University of Toronto, entered the ImageNet image-recognition competition and won with a top-5 error rate of 15.3% — against 26.2% for the second-place entry. It was not a small improvement; it was a rout.

Three ingredients came together in AlexNet: a deep neural network with many layers, a huge labelled dataset (ImageNet, containing millions of photos) and GPUs, whose parallel processing made training such networks practical. Within a few years, deep learning had transformed image recognition, speech recognition and translation, and every major technology company was investing heavily.

Then came the paper that quietly underpins almost everything you use today. In 2017, eight Google researchers published “Attention Is All You Need”, introducing the Transformer architecture. Transformers process text by weighing how much each word should “pay attention” to every other word, which makes them far better at understanding context — and far easier to train at large scale — than the designs before them.

In October 2018, Google released BERT, a Transformer-based model that reads text in both directions to understand context. The same year, OpenAI began its GPT series. The template for the modern AI race — pre-train a huge Transformer on enormous amounts of text, then adapt it — was now in place.

What people wrongly believed at the time: even many specialists assumed these models would remain invisible research tools for years. That assumption had about four years left.

6. The generative explosion (2020–2023)

In 2020, OpenAI released GPT-3, a language model with 175 billion parameters — far larger than anything before it. Its surprise was few-shot learning: give it a couple of examples inside your prompt and it could often perform a new task without any retraining. Researchers were astonished; the general public barely noticed.

That changed on 30 November 2022, when OpenAI released ChatGPT as a free research preview. Built on a model in the GPT-3.5 family, fine-tuned to follow instructions and hold a conversation, ChatGPT reached a reported one million users within five days. Suddenly, anyone could talk to an AI in plain language — no code, no API key, no expertise.

The race that followed reshaped the industry: in 2023, models like GPT-4 added the ability to understand images as well as text, and image and video generators spread the same technology into visual media. “Generative AI” — AI that creates text, images, code, audio and video — became the defining technology story of the decade. It also raised new versions of old questions, including what AI means for jobs.

What people wrongly believed at the time: some declared that human-level general intelligence had effectively arrived; others dismissed chatbots as a passing fad. The evidence fits neither extreme — these systems are genuinely powerful at language tasks, and genuinely unreliable in ways their fluency hides.

7. The agent era (2024–today)

The current phase moves beyond chat. AI agents are systems designed not just to answer a single prompt, but to carry out multi-step tasks: breaking a goal into steps, using tools such as web browsers, code interpreters and other software, checking their own results and iterating. Work that began as “ask a question, get an answer” is becoming “describe a goal, supervise the process”.

This era is young, and honesty matters here. Today’s agents are impressive on well-defined digital tasks and still unreliable on open-ended ones; they need supervision, and their mistakes can be confident ones.

If the pattern of seventy years teaches one lesson, it is this: AI progresses in steps, not straight lines, and the steps that last are the ones grounded in real capability — more data, better mathematics, faster hardware — rather than in press releases.

The history of AI at a glance: timeline

• 1950 — Alan Turing publishes “Computing Machinery and Intelligence” and proposes the imitation game (the Turing Test).
• 1955 — The Dartmouth workshop proposal coins the term “artificial intelligence” (John McCarthy).
• 1956 — The Dartmouth workshop is held; AI is founded as a research field.
• 1966 — ELIZA, the first famous chatbot, is created by Joseph Weizenbaum at MIT.
• 1972 — MYCIN, an early medical expert system, is developed at Stanford.
• 1973 — The Lighthill report criticises AI’s failures; UK funding collapses, beginning the first AI winter.
• 1980 — XCON enters commercial use at DEC, launching the expert-systems boom.
• ~1987 — The expert-systems market collapses; the second AI winter begins.
• 1997 — IBM’s Deep Blue defeats world chess champion Garry Kasparov in a match.
• 2012 — AlexNet wins the ImageNet competition by a wide margin, starting the deep learning era.
• 2017 — “Attention Is All You Need” introduces the Transformer architecture.
• 2018 — Google releases BERT; OpenAI begins the GPT series.
• 2020 — OpenAI releases GPT-3 with 175 billion parameters.
• 2022 — ChatGPT launches on 30 November and brings AI to the general public.
• 2023 — Multimodal models such as GPT-4 accept images as well as text.
• 2024–today — AI agents that use tools and complete multi-step tasks become the industry’s focus.

Frequently asked questions

When was artificial intelligence invented?
AI as a named field dates to 1955–1956: the term “artificial intelligence” appeared in the 1955 Dartmouth workshop proposal by John McCarthy and colleagues, and the workshop itself was held in the summer of 1956. The ideas are older — Alan Turing’s foundational paper appeared in 1950.

Who is considered the father of AI?
John McCarthy is most often given that title: he coined the term “artificial intelligence” and organised the Dartmouth workshop. Alan Turing is regarded as the field’s founding thinker for his 1950 paper and the Turing Test.

What was the first AI chatbot?
ELIZA, created by Joseph Weizenbaum at MIT in 1966. It simulated a psychotherapist using simple pattern matching, and users often treated it as if it understood them — a reaction now called the ELIZA effect.

What is an AI winter?
A period when funding, commercial interest and confidence in AI collapse after over-promising and under-delivering. The two major winters followed the 1973 Lighthill report and the late-1980s collapse of the expert-systems market.

Why did ChatGPT matter so much if GPT-3 already existed?
GPT-3 (2020) was powerful but accessible mainly to developers through an API. ChatGPT (30 November 2022) wrapped a fine-tuned model in a simple, free chat interface that anyone could use, reaching a reported one million users in five days.

Sources

1. The Telegraph India — “From Turing’s question in 1950 to fears that AI could wipe out humanity, a timeline of thinking machines” (Turing 1950; Dartmouth proposal 1955; MYCIN 1972; Lighthill 1973; second winter). telegraphindia.com
2. Cambridge University Press / European Psychiatry — “History and definitions of AI” (Turing 1950; Dartmouth 1956; AI winters overview). cambridge.org
3. Wikipedia — “AI winter” (Lighthill report, 1973, and its consequences). en.wikipedia.org
4. “A brief history of AI: how to prevent another winter” — critical review, arXiv (ELIZA, 1966). ar5iv.labs.arxiv.org
5. HISTORY — “Deep Blue defeats Garry Kasparov in chess match”, 11 May 1997. history.com
6. “AI History: Booms, Winters, and Revivals” (XCON commercial use from 1980; expert-systems collapse). dev.to
7. AIToday — “2012: ImageNet Classification with Deep Convolutional Neural Networks” (AlexNet 15.3% vs 26.2%). aitoday.dev
8. “Attention Is All You Need: The Paper That Revolutionized AI” (Transformer paper, 2017, arXiv:1706.03762). wasilzafar.com
9. Wikipedia — “Transformer (deep learning)” (BERT 2018; GPT series; ChatGPT based on fine-tuned GPT-3.5, end of 2022). en.wikipedia.org
10. Wikipedia — “GPT-3” (175 billion parameters; paper May 2020, API June 2020). en.wikipedia.org
11. “Artificial Intelligence OpenAI: Complete Technical Guide” (ChatGPT launch date, 30 November 2022). justoborn.com

Editorial note: dates in this article were checked against the sources above. Two common retellings were corrected: the term “artificial intelligence” appeared in the 1955 proposal (the workshop followed in 1956), and ChatGPT was built on the GPT-3.5 family, not the original GPT-3.

Explore more beginner-friendly explainers in our Artificial Intelligence category.

Frequently Asked Questions

When was artificial intelligence invented?

AI became a named field in 1955 and 1956. The term artificial intelligence appeared in the 1955 proposal for the Dartmouth workshop, written by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon. The workshop itself took place in the summer of 1956. The founding ideas are older, including Alan Turing’s 1950 paper on machine intelligence.

What was the first AI chatbot?

ELIZA, created by Joseph Weizenbaum at MIT in 1966, was the first famous chatbot. It imitated a psychotherapist by matching patterns in typed text and reflecting users’ words back as questions. The program understood nothing, yet users often confided in it as if it did, a reaction now known as the ELIZA effect.

What is an AI winter?

An AI winter is a period when funding, commercial interest and confidence in AI collapse after promises outrun results. The first followed the critical 1973 Lighthill report and sharp UK funding cuts. The second began around 1987 when expert systems proved brittle, hard to maintain and expensive, causing their commercial market to collapse.

Why was ChatGPT such a turning point in AI history?

ChatGPT launched on 30 November 2022 as a free research preview built on a fine-tuned GPT-3.5 family model. GPT-3 had been powerful but mainly reached developers through an API, while ChatGPT let anyone talk to AI in plain language without code or expertise. It reportedly reached one million users within five days.