Detailed · 14 events
A History of Artificial Intelligence
1950s
Over roughly eight weeks in the summer of 1956 at Dartmouth College in New Hampshire, ten researchers including John McCarthy, Marvin Minsky, Claude Shannon, and Nathaniel Rochester proposed and adopted the term 'artificial intelligence'. Its historical weight lies less in any specific technical result than in establishing the field's own name and the researcher network that would carry it for decades.
Frank Rosenblatt at the Cornell Aeronautical Laboratory introduced the perceptron, a simple learning device that weighted inputs and applied a threshold. In 1960 the 'Mark I Perceptron', a software implementation on the IBM 704 connected to a 400-pixel sensor array, was demonstrated to the US Navy. It marks the start of the neural-network lineage and remains the structural skeleton beneath the MLP, the convolutional network, and modern deep learning.
1960s
Joseph Weizenbaum at MIT built ELIZA, a natural-language conversation program. Its most famous script, 'DOCTOR', imitated a Rogerian psychotherapist by reflecting user input back as questions through simple pattern-matching. Many subjects in early sessions reported feeling that the machine 'understood' them—an effect Weizenbaum found unsettling enough that he later wrote *Computer Power and Human Reason* (1976) in part to caution against it.
1970s
Commissioned by the British Science Research Council, James Lighthill produced a 1973 report assessing the state of AI research and concluded that 'in no part of the field have the discoveries made so far produced the major impact that was then promised.' Funding was sharply cut in Britain; DARPA in the US made similar moves. The decade that followed is known as the first AI winter, the first sustained rebuttal of the optimistic schedules set out by Newell, Minsky, and others.
1980s
Stanford's MYCIN (bacterial-infection diagnosis) and DEC's XCON (automated VAX configuration) showed the appeal of encoding domain expertise as explicit 'if-then' rules. The approach spread rapidly into industry in the early 1980s; Japan's MITI launched the Fifth Generation Computer Project (1982–92). At the 1985 peak, worldwide AI spending reached roughly US$1 billion. The maintenance burden and brittleness of rule systems, however, collapsed the boom by the end of the decade—the second AI winter.
1990s
IBM's chess-specific machine Deep Blue defeated the reigning world champion Garry Kasparov 3.5–2.5 in their six-game rematch. Kasparov had won the 1996 series; IBM substantially upgraded both hardware and evaluation function over the following year. It was the first decisive defeat of a top human chess player by a machine, and is often cited in AI history as a symbolic threshold—though its substance was game-tree search plus a specialised evaluation function, with no modern learning involved.
2010s
At ImageNet 2012, Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton of the University of Toronto reached a top-5 error rate of 15.3%—more than ten points ahead of the runner-up's 26.2% obtained by conventional methods. Their convolutional neural network, 'AlexNet', trained on two NVIDIA GTX 580 GPUs, proved the practical viability of deep learning overnight. Computer vision shifted, almost completely, from hand-engineered features to deep learning from that point forward.
DeepMind's AlphaGo won its five-game series against Lee Sedol, one of the world's top Go players, 4–1. The combinatorial explosion of Go had long been thought to put a machine victory at least a decade away; the combination of Monte Carlo tree search and deep reinforcement learning broke that wall. Move 37 in Game 2—'the divine move'—introduced a placement so unlike anything a human player would have chosen that the professional Go community discussed it for weeks.
Ashish Vaswani and colleagues at Google Brain and Google Research proposed the Transformer—a sequence-to-sequence architecture built solely on self-attention. It displaced the RNN and LSTM, the prevailing NLP architectures, with a structure that parallelised easily and trained efficiently. By 2026 the paper had been cited more than 150,000 times; every modern large language model (BERT, the GPT family, Claude, Gemini) is a descendant.
Dual-core A5, an eight-megapixel camera, and the Siri voice assistant. Steve Jobs died on 5 October, the day after the unveiling—the 4S effectively became his last product. Siri, derived from a startup spun out of SRI International and acquired by Apple, pushed the idea of operating a smartphone through natural language into the mainstream.
2020s
OpenAI's 175-billion-parameter language model. A more-than-hundredfold scale-up from GPT-2 (1.5B), it demonstrated 'few-shot' competence across text generation, summarisation, translation, and code—handling many tasks from a few in-prompt examples. The result pushed the research conversation toward scaling laws for large language models and reshaped the direction of AI investment.
OpenAI made ChatGPT publicly available—GPT-3.5 fine-tuned for dialogue and accessible free through any browser. One million users within five days; one hundred million within two months (the fastest growth of any consumer internet service to that point). Overnight, generative AI moved from a specialist research conversation into homes, schools, and workplaces, and prompted strategic pivots at Google, Meta, Anthropic, and Microsoft.
OpenAI announced GPT-4. The first major LLM to accept image input, it posted top-decile scores on the US Bar exam and AP tests. Parameter counts, training data, and architectural details were withheld—marking OpenAI's shift from the 'publish-and-research' posture of GPT-3 toward a more commercially closed research organisation.
The generation built around Apple Intelligence—Apple's own generative-AI suite. Summarisation, rewriting, image generation, and a strengthened Siri were designed to run on-device wherever possible, with selected workloads offloaded to dedicated Apple servers (Private Cloud Compute). A hardware-and-OS co-design intended to keep the AI surface on the device from being ceded to OpenAI or Google.