This evergreen history article uses authoritative archives and official records. Exact dates are used when documented; gradual inventions and rollouts are described as periods rather than being assigned a misleading single birthday.
Quick facts
- The term “artificial intelligence” was proposed for the 1956 Dartmouth summer research project.
- Alan Turing’s 1950 paper asked whether machines could think and proposed an imitation-game test.
- AI developed through symbolic reasoning, probabilistic methods, neural networks, robotics and machine learning.
- The field experienced repeated “AI winters” when promises exceeded technical and commercial results.
- Generative AI grew from decades of research in language models, representation learning and large-scale computing.
Ideas before the field had a name
Mechanical automata, formal logic and programmable machines created the intellectual background for AI. In the 1940s Warren McCulloch and Walter Pitts described a mathematical model of neural activity. During the same period, wartime computing and information theory showed that machines could manipulate symbols and signals at unprecedented speed.
Alan Turing’s 1950 paper “Computing Machinery and Intelligence” reframed the question of machine thought as an observable conversational test. The imitation game did not define all intelligence, but it became a durable reference point for public and academic discussion.
Dartmouth and symbolic AI
John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon proposed the 1956 Dartmouth workshop in Hanover, New Hampshire. The proposal used the term artificial intelligence and argued that aspects of learning and intelligence could be described precisely enough for machines to simulate them.
Early programs proved theorems, solved puzzles and manipulated symbols. Researchers believed that general intelligence might arrive quickly, but programs worked mainly in restricted environments. The gap between demonstrations and messy real-world problems became a recurring feature of AI history.
Expert systems and the first commercial wave
From the 1960s through the 1980s, expert systems encoded specialist knowledge as rules. Systems in medicine, chemistry and corporate decision support showed that AI could be valuable when the domain was narrow and knowledge could be carefully represented.
Maintaining thousands of rules was difficult, hardware was expensive and systems often failed outside expected conditions. Funding and enthusiasm declined during periods later called AI winters. The lessons were important: intelligence required data, uncertainty management, continuous learning and scalable engineering, not only hand-written logic.
Machine learning and statistical methods
Statistical pattern recognition, decision trees, support-vector machines, probabilistic models and reinforcement learning shifted attention from explicitly programming every rule to learning from examples. Larger digital datasets and faster computers made these approaches increasingly practical.
IBM’s Deep Blue defeated world chess champion Garry Kasparov in New York in 1997, while speech recognition, search ranking and recommendation systems became commercial forms of AI. These systems were specialised, but they influenced everyday life long before the generative-AI boom.
Deep learning, transformers and generative AI
Neural networks returned to prominence as GPUs, data and improved training techniques allowed deeper models. ImageNet competition results in 2012 demonstrated major gains in computer vision. Deep learning then advanced speech, translation, game playing and scientific prediction.
The Transformer architecture introduced in 2017 enabled highly parallel language modelling. Large pretrained models could be adapted through prompting and fine-tuning. From 2020 onward, systems for text, image, audio and video generation reached mass audiences. This created new tools and new concerns about bias, misinformation, labour, safety, copyright, concentration of computing power and environmental cost.
What AI history teaches
AI progress has not followed a smooth line toward human-like intelligence. Different methods rise and fall as hardware, data, evaluation and commercial needs change. Success in one benchmark does not imply general reasoning, reliability or social understanding.
The future will likely combine neural models, tools, retrieval, robotics, symbolic constraints and human oversight. The central question is no longer only whether machines can perform intelligent tasks, but how their power, failures and benefits will be governed.
Common misconceptions
- AI was not invented by one person at one moment.
- The Dartmouth workshop named and organised a field; it did not instantly create intelligent machines.
- Modern AI is broader than chatbots and generative models.
- High benchmark performance does not guarantee truthfulness, safety or general intelligence.
Timeline: key years and locations
| Year | Location | Event | Why it mattered |
|---|---|---|---|
| 1943 | Chicago, USA | McCulloch and Pitts publish a mathematical neuron model | Connected neural activity with computation. |
| 1950 | Manchester/London research context, United Kingdom | Alan Turing publishes the imitation-game paper | Created a foundational question and evaluation idea. |
| 1956 | Hanover, New Hampshire, USA | Dartmouth summer project convenes | Established the name artificial intelligence. |
| 1970s–1980s | United States, Europe and Japan | Expert systems expand | Brought rule-based AI into commercial organisations. |
| 1997 | New York, USA | IBM Deep Blue defeats Garry Kasparov | Demonstrated specialised machine performance at world-champion level. |
| 2012 | Toronto-led research / ImageNet competition | AlexNet transforms computer vision | Showed the power of deep neural networks and GPUs. |
| 2017 | Google research, United States | Transformer architecture is published | Created the foundation for many modern language and multimodal models. |
| 2022 onward | Global | Generative AI reaches mass consumer use | Moved AI into everyday creative and professional workflows. |
Frequently asked questions
Who coined the term artificial intelligence?
John McCarthy is widely credited with the term in the 1955 proposal for the 1956 Dartmouth research project.
What is an AI winter?
A period when funding and enthusiasm decline because systems fail to meet inflated expectations.
Is generative AI the same as all AI?
No. AI also includes perception, planning, robotics, optimisation, recommendations, forecasting and many other methods and applications.