Artificial intelligence feels new, but the ambition behind it is more than seventy years old. The history is a series of bursts of excitement followed by disappointment, and then progress when better data, algorithms and hardware arrived. Knowing the story helps put today's headlines in perspective.
The early idea: can machines think? (1940s and 1950s)
In 1950, the British mathematician Alan Turing published a paper asking whether machines can think. He proposed a practical test, now called the Turing test: if a person chatting through text cannot reliably tell a machine from a human, the machine could be said to show intelligent behaviour.
Around the same time, researchers built early models of artificial neurons, inspired by the brain, and early computers made it possible to try out ideas.
The birth of the field: Dartmouth, 1956
The term artificial intelligence was introduced for a summer workshop at Dartmouth College in 1956, organised by John McCarthy and colleagues including Marvin Minsky, Claude Shannon and Nathaniel Rochester. The proposal assumed that every feature of learning or intelligence could in principle be described precisely enough for a machine to simulate it.
Early programs of the period could prove simple theorems, play checkers and solve algebra word problems. Researchers were optimistic and predicted rapid progress.
Symbolic AI and the first AI winter
For decades, much of the field followed symbolic AI: encoding knowledge as rules and logic. This worked for narrow, well-defined problems but struggled with the messiness of the real world, including vision, speech and common sense.
By the mid-1970s, results had fallen short of promises, computers were too slow, and funding was cut. This period is known as the first AI winter.
Expert systems and a second winter
In the 1980s, expert systems brought AI to business. These programs captured the rules of human experts, for example for diagnosing equipment faults or configuring computer orders. Some were commercially useful.
But they were expensive to maintain, brittle outside their narrow domain and unable to learn. When specialised hardware markets collapsed and expectations were not met, interest dropped again in the late 1980s and early 1990s.
The quiet rise of machine learning (1990s and 2000s)
Meanwhile, a different approach grew: machine learning, where systems learn patterns from data. Statistical methods, support vector machines and decision-tree ensembles solved practical problems such as spam filtering, recommendation and credit scoring.
Notable moments include a computer chess system defeating the world champion in 1997 and the spread of web-scale data, which provided the raw material for learning.
The deep learning breakthrough (2010s)
Neural networks had existed since the 1950s, but three things came together in the early 2010s: very large labelled datasets, powerful GPUs and improved training techniques. In 2012, a deep neural network won a major image-recognition competition by a wide margin, and the field shifted rapidly toward deep learning.
Soon afterward, deep networks improved speech recognition, machine translation and game playing. In 2016, a system called AlphaGo defeated a top professional player at the board game Go, a milestone many had expected to be years away.
Transformers and the era of large language models
In 2017, researchers at Google published the paper "Attention Is All You Need", introducing the transformer architecture. Transformers handle sequences efficiently and scale well with more data and compute.
Training very large transformers on enormous text collections produced large language models that can write, summarise, translate and answer questions. Public chatbots launched in the early 2020s brought generative AI to hundreds of millions of people, and similar techniques extended to images, audio and code.
Patterns worth noticing
Looking across the timeline, several lessons repeat:
- Hype outruns reality, then reality catches up when new ingredients appear.
- Data and computing power matter as much as clever ideas.
- Narrow successes do not automatically lead to general intelligence.
- Society keeps adjusting: questions about jobs, fairness, safety and regulation have followed every wave.
Frequently Asked Questions
Who invented artificial intelligence?
No single person did. Alan Turing laid important foundations, and John McCarthy and colleagues named and organised the field in 1956.
What is an AI winter?
It is a period of reduced interest and funding after AI failed to meet inflated expectations. The field has had at least two.
Why did AI progress so fast after 2012?
Large datasets, GPU computing and better neural network methods combined, letting deep learning outperform older approaches on many tasks.
Is today's AI the same as what early researchers imagined?
Partly. Today's systems are far more capable in narrow tasks than early programs, but none has the broad, flexible intelligence the pioneers hoped to build.
Conclusion
AI's history runs from Turing's question through winters and revivals to deep learning and language models. The recurring lesson is to stay curious but realistic: real progress comes when ideas, data and computing power line up.