Artificial Intelligence
From Turing's 1950 imitation game and the Dartmouth workshop of 1956 to deep learning, large language models and the risk-based rules of the EU AI Act — the history, methods and limits of AI, told through dated primary sources
Artificial intelligence (AI) is the project of building computer systems that perform tasks associated with intelligent beings — systems that infer, from the input they receive, how to generate outputs such as predictions, content, recommendations or decisions. The term was coined for a 1956 summer workshop at Dartmouth College, proposed in 1955 by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon on the conjecture that every feature of intelligence can be described precisely enough for a machine to simulate it. The field's first test of machine thinking was Alan Turing's 1950 imitation game. Early work on symbolic programs, robots such as SRI's Shakey and expert systems such as DENDRAL (1965) gave way to a first funding winter after the Lighthill report of 1973; the learning turn arrived with the back-propagation paper by Rumelhart, Hinton and Williams in Nature in 1986, and accelerated in 2012 when Krizhevsky, Sutskever and Hinton's AlexNet cut the ImageNet contest error to 15.3% against 26.2% for the runner-up. Deep Blue beat Garry Kasparov in 1997, AlphaGo beat Lee Sedol 4-1 in March 2016, and the Transformer architecture of 2017 led to large language models: GPT-3 (175 billion parameters, 2020), ChatGPT (November 2022) and GPT-4 (2023), which scored around the top 10% of test takers on a simulated bar exam. Governance moved in parallel — the OECD AI Principles (2019, updated 2024), the NIST AI RMF (January 2023) and the EU AI Act (Regulation (EU) 2024/1689), in force since 1 August 2024. The systems remain narrow and fallible: NIST's Generative AI Profile names confabulation — confidently presented false content — as a core risk, and Stanford's 2025 AI Index recorded record investment, 78% organizational adoption and sharply rising incidents.

In this article
- 1What artificial intelligence is
- 2Before the name: Turing's imitation game
- 3Dartmouth 1956: the field gets its name
- 4Symbolic AI, expert systems and the first AI winter
- 5The learning turn: neural networks and back-propagation
- 6Deep learning: ImageNet and AlexNet
- 7Machines that beat champions: Deep Blue and AlphaGo
- 8Milestones at a glance
- 9Transformers and large language models
- 10What "AI" does and does not mean
- 11Governance: OECD principles, the EU AI Act and NIST
- 12Today: adoption, investment and open questions
- 13Timeline
- 14Frequently asked questions
- 15Knowledge graph
What artificial intelligence is
Artificial intelligence (AI) is the project of building computer systems that perform tasks commonly associated with intelligent beings. Britannica defines it as the ability of a digital computer or computer-controlled robot to perform such tasks. Regulators describe the same object more operationally: under the OECD definition, an AI system is a machine-based system that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions. The EU AI Act defines an AI system in the same terms, adding that it is designed to operate with varying levels of autonomy and may exhibit adaptiveness after deployment. NIST refers to an AI system as an engineered or machine-based system that can, for a given set of objectives, generate outputs such as predictions.
The label covers a wide working range of deployed systems. The OECD formulation speaks of outputs such as predictions, content, recommendations or decisions, and such outputs now appear in services from machine translation to large language models. Medical applications are already counted: in 2023 the U.S. FDA approved 223 AI-enabled medical devices, up from just six in 2015. None of this amounts to general intelligence — no AI yet matches full human flexibility over wider domains or in tasks requiring much everyday knowledge, though some perform specific tasks as well as humans.
The name itself is dated. The term was coined for a summer 1956 research project at Dartmouth College, proposed in 1955 by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon. What followed — symbolic programs and expert systems, a funding winter, a learning turn, deep networks and large language models — is the substance of this article. Governance came later: OECD principles in 2019, the NIST risk framework in 2023, and the EU AI Act, in force since 1 August 2024.
1. Before the name: Turing's imitation game
Alan Turing proposed the field's first test of machine thinking before it had a name. His paper “Computing Machinery and Intelligence” appeared in Mind in October 1950. Turing opened with the question “Can machines think?” and judged it too meaningless to deserve discussion, replacing it with a game he called the imitation game. The game is played with three people — a man, a woman and an interrogator — and the interrogator must determine which of the two unseen players is the man and which is the woman, knowing them only as X and Y. So that tone of voice could not help, the answers were to be written, or better still typewritten.
Turing made his expectations explicit. In about fifty years' time, he predicted, it would be possible to programme computers to play the game so well that an average interrogator would have no more than a 70 per cent chance of making the right identification after five minutes of questioning. He hoped machines would eventually compete with men in all purely intellectual fields, and noted that many people thought a very abstract activity, like chess, the best place to start. The original question, he added, was not one to argue but to outgrow: he expected the growth of machine capability to settle it in practice, with educated opinion eventually speaking of machines thinking without expecting contradiction.

The paper set a pattern that would recur. Five years after Turing wrote, the Dartmouth proposal turned the simulation of intelligence into a research programme: every aspect of learning or any other feature of intelligence, it conjectured, can be described precisely enough for a machine to be made to simulate it. How far that programme has run, and where it has stalled, is measured in the milestones that follow.
2. Dartmouth 1956: the field gets its name
The name arrived by proposal. On 31 August 1955, John McCarthy of Dartmouth College, Marvin Minsky of Harvard, Nathaniel Rochester of IBM and Claude Shannon of Bell Telephone Laboratories put forward a plan for the following summer. It proposed a 2-month, 10-person study of artificial intelligence to be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire — the first appearance of the term. The study was to proceed on a single founding conjecture: that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.
The proposal was ambitious about method as well as subject. An attempt would be made, it said, to find how to make machines use language, form abstractions and concepts, solve problems then reserved for humans, and improve themselves. The authors judged that a significant advance could be made on one or more of these problems if a carefully selected group of scientists worked on them together for a summer. The term coined for that workshop has named the field ever since.
We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, 31 August 1955
The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, 31 August 1955
An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, 31 August 1955
3. Symbolic AI, expert systems and the first AI winter
Early artificial intelligence was largely symbol manipulation: the Dartmouth plan itself aimed at language, abstraction and concept formation. Its most durable early product was the expert system — a computer program that uses artificial-intelligence methods to solve problems within a specialised domain that ordinarily requires human expertise. The first expert system was developed in 1965 by Edward Feigenbaum and Joshua Lederberg of Stanford University and became known as Dendral; it was designed to analyse chemical compounds. Such systems rest on two components: a knowledge base, an organised collection of facts about the domain, and an inference engine that interprets and evaluates those facts to provide an answer.

Knowledge for the base was acquired from human experts through interviews and observations, usually represented as if-then production rules: if some condition is true, then a certain inference can be made or action taken. Expert systems incorporated heuristic rules of thumb and, increasingly, the ability to learn from experience; they remain aids to, rather than replacements for, human experts. The approach nevertheless met a structural limit: the combinatorial explosion, in which the number of paths through a network grows with the alternatives open at one time and the number of steps.
The reckoning came in 1973, when the UK Science Research Council published James Lighthill's survey, commissioned to give an unbiased view of the state of AI research primarily in the UK. The report supported work related to automation and to computer simulation of neurophysiological and psychological processes, but was highly critical of basic research in foundation areas such as robotics and language processing. Its central finding was damaging: in no part of the field, Lighthill judged, had the discoveries made so far produced the major impact that was then promised. The effect was a massive loss of confidence in AI by the UK academic establishment, including the funding body — a condition that persisted for almost a decade, the first AI winter. AI research continued, but the next attempt to mount a major activity in the area did not come until a September 1982 meeting on intelligent knowledge-based systems.
In no part of the field have the discoveries made so far produced the major impact that was then promised.
The Lighthill Report, Artificial Intelligence: A General Survey, 1973
Expert systems remain aids to, rather than replacements for, human experts.
Encyclopædia Britannica, “Expert system”
4. The learning turn: neural networks and back-propagation
The alternative to symbols was nearly as old as the field itself. In 1943 Warren McCulloch and Walter Pitts showed that, because nervous activity is all-or-none, neural events and their relations can be treated by means of propositional logic. They found that the behaviour of every network of neurons could be described in logical terms, and that for any logical expression meeting certain conditions one could find a network behaving as it describes. Their paper became the first mathematical model of an artificial neuron and the seed of neural-network research.
Learning returned to the centre in 1986, when David Rumelhart, Geoffrey Hinton and Ronald Williams published “Learning representations by back-propagating errors” in Nature, volume 323, pages 533-536. The procedure computes, for each weight of a network, a gradient indicating by what amount the error would increase or decrease if that weight were increased by a tiny amount. The network is then adjusted according to that gradient, and the network's representations are learned from data rather than written by hand.
The methods that resulted are now called deep learning: computational models composed of multiple processing layers that learn representations of data with multiple levels of abstraction. In these models the layers of features are not designed by human engineers but discovered from data using a general-purpose learning procedure. Their effect was felt well beyond vision: these methods dramatically improved the state of the art in speech recognition.
5. Deep learning: ImageNet and AlexNet
ImageNet was built to give the field something to measure. Presented in 2009, it uses the hierarchical structure of WordNet, giving the database a large-scale ontology of images. The team had constructed 12 subtrees containing 3.2 million images by the time of the paper, with the first release covering 5,247 synsets and 3.2 million images in total. The images were labelled by human workers on Amazon Mechanical Turk, with about 50 million images planned for the completed database.

The decisive contest was ILSVRC-2012. Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images of the ImageNet challenge. The network had 60 million parameters and 650,000 neurons, and consisted of five convolutional layers. On ILSVRC-2010 test data it reached top-1 and top-5 error rates of 37.5% and 17.0%, considerably better than the previous state of the art.
The winning entry was a variant entered in the ILSVRC-2012 competition, which achieved a top-5 test error rate of 15.3% against 26.2% for the second-best entry. Training took between five and six days on two GTX 580 3GB GPUs, and the authors expected results to improve simply by waiting for faster GPUs and bigger datasets. The deep-learning literature records that the success came from the efficient use of GPUs, and that progress on object recognition had been slow until the ImageNet competition of 2012.
6. Machines that beat champions: Deep Blue and AlphaGo
In May 1997, IBM's Deep Blue became the first computer system to defeat a reigning world chess champion in a match under standard tournament controls. Its six-game rematch against Garry Kasparov in New York ended 3.5-2.5 in the machine's favour. The system evaluated 200 million chess positions per second, a processing speed of 11.38 billion floating-point operations per second. The first encounter had ended differently: in February 1996 Deep Blue won the first game against Kasparov in Philadelphia — the first victory by a computer against a reigning world champion under regular time controls — but lost the match 4-2. The 1997 victory marked an inflection point in computing.

Go was the harder problem — long viewed as the most challenging of the classic games for artificial intelligence, owing to its enormous search space and the difficulty of evaluating board positions and moves. The program achieved a 99.8% winning rate against other Go programs, and defeated the human European champion by 5 games to 0 — the first time a computer program had beaten a human professional in the full-sized game.
Then came Lee Sedol. The player — winner of 18 world titles — was beaten 4-1 by AlphaGo in Seoul in March 2016, in a match watched by more than 200 million people worldwide. Twenty years apart, the two matches made the same point in public: on bounded domains, machines could beat the best human players.
7. Milestones at a glance
The milestones below run from the McCulloch-Pitts neuron of 1943 to the entry into force of the EU AI Act in 2024, as dated by the sources cited in this article12A Logical Calculus of the Ideas Immanent in Nervous Activity (Bulletin of Mathematical Biophysics) · W. S. McCulloch & W. Pitts; research copy hosted by Carnegie Mellon University ↗3AI Act — Regulation (EU) 2024/1689; European Commission AI Act pages and the Official Journal text · European Commission / Publications Office of the European Union ↗.
| Year | Milestone | Why it matters |
|---|---|---|
| 1943 | McCulloch and Pitts publish their logical calculus of nervous activity | Because neural activity is all-or-none, neural events and their relations can be treated with propositional logic — the seed of neural-network research12. |
| 1950 | Turing's imitation game | “Computing Machinery and Intelligence” replaces “Can machines think?” with a game played by a man, a woman and an interrogator9Computing Machinery and Intelligence (Mind, vol. LIX, no. 236) · A. M. Turing — Mind (Oxford University Press); research copy hosted by UMBC ↗. |
| 1955/1956 | The Dartmouth proposal | A “2 month, 10 man study of artificial intelligence” for the summer of 1956 — the first use of the term8A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence · J. McCarthy, M. L. Minsky, N. Rochester, C. E. Shannon (Stanford University page) ↗. |
| 1965 | Dendral, the first expert system | Feigenbaum and Lederberg build a program to analyse chemical compounds at Stanford University10Expert system · Encyclopædia Britannica ↗. |
| 1973 | The Lighthill report | A UK survey finds the promised impact unrealised; confidence in AI falls for almost a decade11Artificial Intelligence: A General Survey (the Lighthill Report) · Sir James Lighthill — UK Science Research Council; archived by Chilton Computing (UKRI STFC) ↗. |
| 1986 | Back-propagation in Nature | Rumelhart, Hinton and Williams publish “Learning representations by back-propagating errors” (Nature 323, 533-536)13Deep learning (Nature 521, 436–444) · Y. LeCun, Y. Bengio & G. Hinton — Nature ↗. |
| 1997 | Deep Blue defeats Kasparov | The first computer system to beat a reigning world chess champion in a match under standard tournament controls16Deep Blue · IBM (IBM History) ↗. |
| 2009 | ImageNet released | 12 subtrees with 3.2 million images, built on the WordNet hierarchy for object recognition14ImageNet: A Large-Scale Hierarchical Image Database (CVPR 2009) · J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li & L. Fei-Fei — IEEE Conference on Computer Vision and Pattern Recognition ↗. |
| 2012 | AlexNet wins ILSVRC-2012 | A winning top-5 error of 15.3% against 26.2% for the runner-up, trained on two GTX 580 GPUs15ImageNet Classification with Deep Convolutional Neural Networks (NIPS 2012) · A. Krizhevsky, I. Sutskever & G. Hinton — Advances in Neural Information Processing Systems ↗. |
| 2016 | AlphaGo defeats Lee Sedol | A 4-1 victory in Seoul, watched by more than 200 million people worldwide18AlphaGo (research breakthrough page) · Google DeepMind ↗. |
| 2017 | The Transformer | “Attention Is All You Need” proposes an architecture based solely on attention mechanisms5Attention Is All You Need (arXiv:1706.03762) · A. Vaswani et al. — arXiv (NeurIPS 2017) ↗. |
| 2019 | OECD AI Principles | The first intergovernmental standard on AI, adopted on 22 May 20192OECD AI Principles; Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449) · Organisation for Economic Co-operation and Development (OECD) ↗. |
| 2020 | GPT-3 | An autoregressive language model with 175 billion parameters, applied without gradient updates or fine-tuning6Language Models are Few-Shot Learners (arXiv:2005.14165) · T. B. Brown et al. (OpenAI) — arXiv ↗. |
| 2022 | ChatGPT | Released on 30 November 2022 as a free research preview, interacting in a conversational way19Introducing ChatGPT · OpenAI (via Internet Archive) ↗. |
| 2023 | GPT-4 and the NIST AI RMF | GPT-4 passes a simulated bar exam scoring around the top 10% of test takers20GPT-4 Technical Report (arXiv:2303.08774) · OpenAI — arXiv ↗; NIST releases its AI Risk Management Framework on 26 January 20234AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1) and Generative AI Profile (NIST AI 600-1) · National Institute of Standards and Technology (NIST), U.S. Department of Commerce ↗. |
| 2024 | The EU AI Act enters into force | Regulation (EU) 2024/1689 — the first-ever comprehensive legal framework on AI worldwide3. |
8. Transformers and large language models
In June 2017, Ashish Vaswani and colleagues published “Attention Is All You Need”, proposing the Transformer — a network architecture based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. It was more parallelizable than earlier architectures and required significantly less time to train. On English-to-French translation it set a new single-model state-of-the-art BLEU score of 41.8 after 3.5 days on eight GPUs, a small fraction of the training costs of the best previous models.

GPT-3 followed in 2020: an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, tested in the few-shot setting. For all tasks it was applied without any gradient updates or fine-tuning, with tasks specified purely through text interaction with the model. It performed strongly on many natural-language datasets — translation, question answering and cloze tasks among them — as well as on tasks requiring on-the-fly reasoning or domain adaptation. The authors also identified datasets where few-shot learning still struggled, and others with methodological issues tied to training on large web corpora. Human evaluators had difficulty distinguishing its generated news articles from articles written by humans.
ChatGPT arrived on 30 November 2022 as a free research preview: a model that interacts in a conversational way, trained with reinforcement learning from human feedback — the same methods as InstructGPT, with slight differences in the data-collection setup. It could answer follow-up questions, admit mistakes, challenge incorrect premises and reject inappropriate requests. GPT-4 followed in March 2023, a large-scale multimodal model that accepts image and text inputs and produces text outputs.
GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. While less capable than humans in many real-world scenarios, it exhibited human-level performance on various professional and academic benchmarks, including a simulated bar exam passed with a score around the top 10% of test takers. Its developer recorded another kind of result: some aspects of GPT-4's performance could be predicted from models trained with no more than 1/1,000th of its compute.
9. What "AI" does and does not mean
An AI system is “machine-based”; it “infers” from its input; it generates outputs such as predictions, content, recommendations or decisions2.
These are definitions of behaviour — inference and outputs — rather than of consciousness, understanding or intent2.
Britannica warns against a common conflation: artificial intelligence and machine learning are not the same, and machine learning is the method that helps a computer to achieve AI1Artificial intelligence · Encyclopædia Britannica ↗.
The limits are as documented as the capabilities. No AI yet matches full human flexibility over wider domains or in tasks requiring much everyday knowledge, though some perform specific tasks as well as humans1. NIST's Generative AI Profile names confabulation — confidently presented erroneous or false content, colloquially called hallucination — as a phenomenon of generative systems, and a natural result of their design: they generate outputs that approximate the statistical distribution of their training data4. The NIST framework itself is intended to be voluntary, rights-preserving, non-sector-specific and use-case agnostic4. Even GPT-4, for all its examination results, was reported as less capable than humans in many real-world scenarios20.
An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.
OECD, Recommendation of the Council on Artificial Intelligence
“Confabulation” refers to a phenomenon in which GAI systems generate and confidently present erroneous or false content in response to prompts.
NIST, Generative AI Profile (NIST AI 600-1)
Machine learning helps a computer to achieve artificial intelligence.
Encyclopædia Britannica, “Artificial intelligence”
10. Governance: OECD principles, the EU AI Act and NIST
Governance began with principles. The OECD Recommendation on Artificial Intelligence — the first intergovernmental standard on AI — was adopted by the OECD Council at Ministerial level on 22 May 2019. It was revised on 8 November 2023 so that its definition of an AI system would stay technically accurate and reflect developments including generative AI, and the principles were updated in May 2024. They promote the use of AI that is innovative and trustworthy and that respects human rights and democratic values — from inclusive growth and human rights to transparency, robustness and accountability; 47 adherents have committed to them.
In the United States, the federal instrument was a framework rather than a statute. NIST released its AI Risk Management Framework on 26 January 2023, after a consensus-driven, open, transparent and collaborative development process. It is intended for voluntary use, to improve the ability to incorporate trustworthiness considerations into the design, development, use and evaluation of AI products, services and systems. The framework is intended to be voluntary, rights-preserving, non-sector-specific and use-case agnostic, with a core composed of four functions: GOVERN, MAP, MEASURE and MANAGE.
The AI Act — Regulation (EU) 2024/1689 — entered into force on 1 August 2024 as the first-ever legal framework on AI, addressing the risks of AI. It became applicable on 2 August 2026 with some exceptions: prohibited AI practices and AI literacy obligations applied from 2 February 2025, and the governance rules and obligations for general-purpose AI models became applicable on 2 August 2025. AI use cases that can pose serious risks to health, safety or fundamental rights are classified as high-risk under the regulation.
11. Today: adoption, investment and open questions
Adoption is now the measurable story. private AI investment grew to $109.1 billion — nearly 12 times China's $9.3 billion and 24 times the U.K.'s $4.5 billion. Business usage accelerated alongside it: 78% of organizations reported using AI in 2024, up from 55% the year before. Between November 2022 and October 2024, the inference cost of a system performing at the level of GPT-3.5 dropped over 280-fold, driven by increasingly capable small models.
Capability gains are dated, not hypothetical. Benchmarks introduced in 2023 to test advanced AI systems saw scores rise within a year by 18.8 points on MMMU, 48.9 on GPQA and 67.3 on SWE-bench. In 2023 the U.S. FDA approved 223 AI-enabled medical devices, up from just six in 2015. Two Nobel Prizes have recognised work related to deep learning.
The frictions are documented alongside the gains. AI-related incidents are rising sharply, yet standardized responsible-AI evaluations remain rare among major industrial model developers. Governments are responding: in 2024, U.S. federal agencies introduced 59 AI-related regulations — more than double the number in 2023 — issued by twice as many agencies. The gap the sources describe is one of assurance: what these systems can do is increasingly measured, and how reliably they can be trusted is not.
12. Timeline
- 1943A logical calculus for nervous activityWarren McCulloch and Walter Pitts publish "A Logical Calculus of the Ideas Immanent in Nervous Activity", showing that because neural activity is all-or-none, neural events and their relations can be treated with propositional logic — the first mathematical model of an artificial neuron and the seed of neural-network research.
- October 1950Turing's imitation gameAlan Turing publishes "Computing Machinery and Intelligence" in Mind. He replaces the question "Can machines think?" with the imitation game, in which an interrogator tries to tell a machine from a human by written answers alone, and predicts that by about fifty years' time an average interrogator would have no more than a 70 per cent chance of making the right identification after five minutes of questioning.
- Summer 1956 (proposal dated 31 August 1955)The Dartmouth workshop names the fieldJohn McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon propose "a 2 month, 10 man study of artificial intelligence" for the summer of 1956 at Dartmouth College — the first use of the term. The proposal's founding conjecture: every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it.
- 1965DENDRAL, the first expert systemEdward Feigenbaum and Joshua Lederberg at Stanford University develop the first expert system, later known as Dendral, to analyse chemical compounds. Expert systems encode human expertise in a knowledge base of if-then production rules that an inference engine evaluates to give an answer.
- Early 1973The Lighthill report and the first AI winterThe UK Science Research Council publishes James Lighthill's survey finding that "in no part of the field have the discoveries made so far produced the major impact that was then promised". The report provoked a massive loss of confidence in AI by the UK academic establishment, including its funding body — a freeze that persisted for almost a decade.
- 1986Back-propagation reaches NatureDavid Rumelhart, Geoffrey Hinton and Ronald Williams publish "Learning representations by back-propagating errors" in Nature (323, 533-536). The procedure computes, for each weight, a gradient indicating how the error would move if the weight changed, and adjusts the network accordingly — the method behind modern deep learning.
- May 1997Deep Blue defeats KasparovIBM's Deep Blue becomes the first computer system to defeat a reigning world chess champion in a match under standard tournament controls, beating Garry Kasparov 3.5-2.5 in a six-game rematch in New York. The machine evaluated 200 million chess positions per second at 11.38 billion floating-point operations per second.
- June 2009ImageNet arrivesJia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li and Li Fei-Fei present ImageNet, a WordNet-backed image database for object recognition: 12 subtrees with 3.2 million cleanly annotated images at first release, gathered with Amazon Mechanical Turk. It becomes the training ground that makes the 2012 deep-learning breakthrough measurable.
- September 2012AlexNet wins ILSVRC-2012Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton win the ImageNet contest with a deep convolutional network trained on 1.2 million images across 1000 classes — five to six days on two GTX 580 3GB GPUs — reaching a winning top-5 test error rate of 15.3%, compared with 26.2% for the second-best entry.
- March 2016AlphaGo beats Lee SedolGoogle DeepMind's AlphaGo beats Lee Sedol, winner of 18 world titles, 4-1 in Seoul — a match watched by more than 200 million people worldwide. Earlier, trained on human games and self-play with value and policy networks, it had won 99.8% of games against other Go programs and defeated European champion Fan Hui 5-0.
- June 2017The TransformerAshish Vaswani and colleagues publish "Attention Is All You Need", proposing the Transformer — a network architecture based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. It is more parallelizable and trains in a fraction of the time, reaching 28.4 BLEU on English-to-German translation and 41.8 on English-to-French after 3.5 days on eight GPUs.
- 22 May 2019OECD AI Principles adoptedThe OECD Council adopts the Recommendation on Artificial Intelligence — the first intergovernmental standard on AI — setting principles for trustworthy AI from inclusive growth and human rights to transparency, robustness and accountability. The principles are updated in May 2024, and 47 adherents now commit to them.
- June 2020GPT-3: scaling language modelsOpenAI publishes "Language Models are Few-Shot Learners": GPT-3, an autoregressive language model with 175 billion parameters, applied without any gradient updates or fine-tuning, with tasks specified purely by text. It performs strongly on translation, question answering and cloze tasks, and generates news articles that evaluators struggle to distinguish from human-written ones.
- 30 November 2022ChatGPT opens the chatbot eraOpenAI releases ChatGPT as a free research preview: a model that interacts in a conversational way, fine-tuned from the GPT-3.5 series using reinforcement learning from human feedback, able to answer follow-up questions, admit mistakes, challenge incorrect premises and reject inappropriate requests.
- March 2023GPT-4 passes professional examsOpenAI reports GPT-4, a large-scale multimodal model that accepts image and text inputs and produces text outputs. While less capable than humans in many real-world scenarios, it exhibits human-level performance on many professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers.
- 1 August 2024The EU AI Act enters into forceThe AI Act — Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence — enters into force, the first-ever comprehensive legal framework on AI worldwide. Its risk-based approach schedules prohibitions of certain practices from 2 February 2025, obligations for general-purpose AI models from 2 August 2025, and general application from 2 August 2026.
13. Frequently asked questions
What is artificial intelligence?
Britannica defines artificial intelligence as the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings1. Regulators use a more operational definition: an AI system is a machine-based system that, for explicit or implicit objectives, infers from its input how to generate outputs such as predictions, content, recommendations or decisions2.
What was Turing's imitation game?
In his 1950 paper, Alan Turing replaced the question of whether machines can think with a game played by three people: a man, a woman and an interrogator9. The interrogator had to work out which of the two unseen players was the man and which was the woman, knowing them only as X and Y, with answers written so that tone of voice could not help9.
When was the term “artificial intelligence” first used?
The term first appeared in the 1955 proposal for a 2-month, 10-person study of artificial intelligence to be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire8. It was put forward on 31 August 1955 by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon8.
What is an expert system?
An expert system is a computer program that uses artificial-intelligence methods to solve problems within a specialised domain that ordinarily requires human expertise10. It relies on a knowledge base — an organised collection of facts about its domain — and an inference engine that interprets and evaluates those facts to provide an answer10.
What was the first AI winter?
The UK Science Research Council's 1973 Lighthill report found that in no part of the field had the discoveries so far produced the major impact that was then promised11. It provoked a massive loss of confidence in AI by the UK academic establishment, including its funding body, a condition that persisted for almost a decade11.
What are large language models, and where did they come from?
They descend from the Transformer, a 2017 network architecture based solely on attention mechanisms that dispensed with recurrence and convolutions entirely5. GPT-3 then showed in 2020 that an autoregressive language model with 175 billion parameters could perform many tasks without any gradient updates or fine-tuning, specified purely through text6.
What does the EU AI Act do?
The AI Act — Regulation (EU) 2024/1689 — entered into force on 1 August 2024 as the first-ever legal framework on AI, addressing its risks3. Prohibited AI practices applied from 2 February 2025, obligations for general-purpose AI models from 2 August 2025, and general application from 2 August 20263.
What is “confabulation” in AI systems?
NIST's Generative AI Profile defines confabulation as a phenomenon in which generative AI systems generate and confidently present erroneous or false content in response to prompts — colloquially referred to as hallucinations or fabrications4. Such outputs are a natural result of how generative models are designed: they generate outputs that approximate the statistical distribution of their training data4.
References (20)
- 1Artificial intelligence — Encyclopædia Britannica 2025 Archive↩View source
- 2OECD AI Principles; Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449) — Organisation for Economic Co-operation and Development (OECD) Web↩View source
- 3AI Act — Regulation (EU) 2024/1689; European Commission AI Act pages and the Official Journal text — European Commission / Publications Office of the European Union 2024 Web↩View source
- 4AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1) and Generative AI Profile (NIST AI 600-1) — National Institute of Standards and Technology (NIST), U.S. Department of Commerce Web↩View source
- 5Attention Is All You Need (arXiv:1706.03762) — A. Vaswani et al. — arXiv (NeurIPS 2017) 1706 Journal↩View source
- 6Language Models are Few-Shot Learners (arXiv:2005.14165) — T. B. Brown et al. (OpenAI) — arXiv 2005 Journal↩View source
- 7The 2025 AI Index Report — Top Takeaways — Stanford Institute for Human-Centered Artificial Intelligence (HAI), Stanford University 2025 WebView source
- 8A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence — J. McCarthy, M. L. Minsky, N. Rochester, C. E. Shannon (Stanford University page) Web↩View source
- 9Computing Machinery and Intelligence (Mind, vol. LIX, no. 236) — A. M. Turing — Mind (Oxford University Press); research copy hosted by UMBC Book↩View source
- 10Expert system — Encyclopædia Britannica 2025 Archive↩View source
- 11Artificial Intelligence: A General Survey (the Lighthill Report) — Sir James Lighthill — UK Science Research Council; archived by Chilton Computing (UKRI STFC) Web↩View source
- 12A Logical Calculus of the Ideas Immanent in Nervous Activity (Bulletin of Mathematical Biophysics) — W. S. McCulloch & W. Pitts; research copy hosted by Carnegie Mellon University Web↩View source
- 13Deep learning (Nature 521, 436–444) — Y. LeCun, Y. Bengio & G. Hinton — Nature Journal↩View source
- 14ImageNet: A Large-Scale Hierarchical Image Database (CVPR 2009) — J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li & L. Fei-Fei — IEEE Conference on Computer Vision and Pattern Recognition 2009 Web↩View source
- 15ImageNet Classification with Deep Convolutional Neural Networks (NIPS 2012) — A. Krizhevsky, I. Sutskever & G. Hinton — Advances in Neural Information Processing Systems 2012 Web↩View source
- 16Deep Blue — IBM (IBM History) Web↩View source
- 17Mastering the game of Go with deep neural networks and tree search (Nature 529, 484–489) — D. Silver et al. — Nature JournalView source
- 18AlphaGo (research breakthrough page) — Google DeepMind Web↩View source
- 19Introducing ChatGPT — OpenAI (via Internet Archive) 2023 Archive↩View source
- 20GPT-4 Technical Report (arXiv:2303.08774) — OpenAI — arXiv Journal↩View source
Muhammad Ali
Mahatma Gandhi
Ottoman Empire