History of Artificial Intelligence

The history of artificial intelligence spans more than seventy years, from early work on thinking machines in the 1940s and 1950s through alternating periods of rapid progress, public excitement and funding withdrawal, to the deep learning and generative AI systems of the 2020s. The field's name was coined for a 1956 research workshop at Dartmouth College, which is conventionally treated as its founding event, and its development since has been shaped by the interplay of algorithmic ideas, computing power and data.^[1][3][4]

Early Foundations

The intellectual groundwork was laid in the 1940s. In 1943 the neurophysiologist Warren McCulloch and the logician Walter Pitts described a mathematical model of an artificial neuron, and mid-century cybernetics explored feedback and control in machines and organisms. In 1950 the British mathematician Alan Turing published "Computing Machinery and Intelligence", proposing what became known as the Turing test: if a machine's responses were indistinguishable from a human's, it should be regarded as thinking. Turing predicted that machines would reach human-level conversational performance within decades.^[2][3]

In August 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon circulated a proposal for a two-month summer research project at Dartmouth College in 1956, based on the conjecture that every aspect of learning or any other feature of intelligence could in principle be described precisely enough for a machine to simulate it. The workshop is widely credited with establishing artificial intelligence — the term McCarthy chose — as a research field, and its optimism that a generation of researchers could make substantial progress set the tone for the following two decades.^[1][3]

Symbolic AI and the First AI Winter

Early research centred on symbolic methods: programs that manipulated logical expressions, search trees and formal rules. The Logic Theorist and the General Problem Solver, developed in the mid-1950s and early 1960s, proved theorems and solved puzzles, while Joseph Weizenbaum's ELIZA (1966) demonstrated rudimentary conversation through simple pattern matching. Funding flowed from agencies impressed by such demonstrations, accompanied by ambitious predictions that machine translation, robotics and general intelligence were within reach.^[4]

Reality intervened in the 1970s. Machine translation projects failed to meet expectations, neural network research stalled after critiques of the perceptron, and the 1973 Lighthill report for the British government concluded that AI had failed to deliver on its promises. Research funding was cut in the United Kingdom and by United States defence research agencies — a period now known as the first "AI winter".^[4]

Expert Systems and the Second AI Winter

The 1980s brought a commercial revival built on expert systems: rule-based programs that encoded the knowledge of human specialists. Systems such as XCON, used to configure computer orders, and medical diagnosis tools such as MYCIN demonstrated practical value, and Japan's Fifth Generation Computer Systems project, launched in 1982, galvanised industrial policy in the United States and Europe. Companies built businesses around specialised LISP machines and expert-system tooling, and investment boomed.^[4]

The boom again overshot. Expert systems proved expensive to maintain and brittle outside narrow domains, and the specialised hardware market collapsed as general-purpose workstations became more powerful. By the late 1980s and early 1990s the second AI winter had set in. The period nevertheless produced landmarks: in 1997 IBM's Deep Blue defeated world chess champion Garry Kasparov in a six-game match, becoming the first computer to beat a reigning champion under tournament conditions.^[4][5]

Machine Learning and Deep Learning

From the 1990s the field shifted toward statistical machine learning, in which systems learn patterns from data rather than encoding rules by hand. Techniques such as support vector machines and ensemble methods advanced handwriting recognition, spam filtering, search ranking and recommendation. In parallel, a small group of researchers continued developing neural networks, and in 2012 a deep convolutional network called AlexNet won the ImageNet image-recognition competition by a wide margin — a result that triggered the modern deep learning era.^[4][9]

Progress accelerated with data and computing power. In 2016 Google DeepMind's AlphaGo defeated the world champion of Go, a game long considered beyond machines. In 2017 the Transformer architecture, introduced in the paper "Attention Is All You Need", provided the design that would underpin large language models, and successive GPT models demonstrated that scaling transformer training produced steadily more capable systems. By the early 2020s language models could converse fluently, write software and answer professional-level examination questions, drawing comparisons with human performance on standardised tests.^[6][7][9]

Generative AI and the Frontier Era

OpenAI's release of ChatGPT in November 2022 brought generative AI to mainstream attention, becoming one of the fastest-adopted consumer applications in history, and releases from Google, Anthropic, Meta and Chinese developers followed in rapid succession. By 2026 the Stanford AI Index reported that several frontier models met or exceeded human baselines on PhD-level science questions and competition mathematics, that organisational adoption of AI had reached 88 per cent, and that industry produced more than 90 per cent of notable frontier models — a consolidation of research inside private companies that earlier generations of the field would not have recognised. The same period brought AI agents that operate software, reasoning models that spend more computation on deliberation, and the first commercial wave of autonomous transport.^[8][9]

>See Also

• Artificial Intelligence
• Turing Test
• Perceptron
• Transformer Architecture
• Generative AI
• MDEC