Nobody invented SI (AI) on their own. The name "artificial intelligence" was coined by John McCarthy in a 1955 research proposal he wrote with Marvin Minsky, Nathaniel Rochester and Claude Shannon, and the field took shape at a summer workshop at Dartmouth College in 1956. Alan Turing had already asked whether machines could think, and the first working programs came from Allen Newell, Herbert Simon and Cliff Shaw. Here is who did what, read from the original documents, and why the credit is shared.
The short answer. If you need one name, it is John McCarthy: he gave the field its name in the Dartmouth proposal dated 31 August 1955. But the idea is older (Turing, 1947 to 1950), the first program that reasoned came from Newell, Simon and Shaw, and the neural networks behind today's chatbots owe a great deal to researchers honoured decades later, such as Geoffrey Hinton, Yoshua Bengio and Yann LeCun. SI is a relay race, not a single invention.
Who invented SI (AI)?
Asking who invented AI is a bit like asking who invented medicine. There was no single machine and no single patent. What can be dated precisely is the name. On 31 August 1955, four researchers signed "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence": J. McCarthy of Dartmouth College, M. L. Minsky of Harvard University, N. Rochester of IBM and C. E. Shannon of Bell Telephone Laboratories. The name of the field is in its title, and the credit for it goes to McCarthy: a 2026 history paper by Jeff Shrager describes the 1956 meeting as a workshop "on what John McCarthy had dubbed 'artificial intelligence'".
The four came from very different places. The proposal describes Shannon as the man who "developed the statistical theory of information"; Minsky as a Harvard Junior Fellow who had "built a machine for simulating learning by nerve nets"; Rochester as the manager of information research at IBM, co-designer of the IBM Type 701 computer; and McCarthy as an assistant professor of mathematics who had worked on "the mathematical nature of the thought process". A mathematician, a neural network builder, a computer engineer and an information theorist: the mix already says a lot about what SI would become.
A note on the letters. SI is the name the US government has used for this technology since September 2026; researchers, companies and most of the world still say AI. Our guide What is SI? explains that renaming. The technology, and its history, did not change with the label.
What came before the name?
The name came in 1955, but the work started earlier. When the British mathematician Sir James Lighthill reviewed the field in 1972 (more on his report below), he counted "twenty-five years of life", which he took as beginning with Alan Turing's 1947 article "Intelligent Machinery". The Dartmouth proposal itself lists earlier work on "neuron nets" by Uttley, Rashevsky, Farley and Clark, Pitts and McCulloch, Minsky, Rochester and Holland. The four authors were naming a field that already existed in pieces.
Turing's best known contribution came in 1950, in the journal Mind. His paper "Computing Machinery and Intelligence" opens: "I propose to consider the question, 'Can machines think?'" He then replaced that question with a test of imitation, which we cover in detail in our article on the Turing test. Less quoted, but closer to how SI is built today, is the end of the same paper. Instead of trying to program an adult mind, Turing asked, "why not rather try to produce one which simulates the child's?" A machine with "rather little mechanism, and lots of blank sheets" could then be educated. That is, in plain words, the idea of a learning machine, the approach that dominates SI today.
The roots go back further still. In the same paper Turing answers "Lady Lovelace's Objection": Ada Lovelace had written that Charles Babbage's Analytical Engine "has no pretensions to originate anything" and can only do what we know how to order it to perform. Our Turing test article quotes her note.
What happened at Dartmouth in 1956?
The proposal asked for "a 2 month, 10 man study of artificial intelligence" during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. Its central sentence is still quoted today:
"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."
The authors listed seven problems to work on: automatic computers, how a computer can be programmed to use a language, "neuron nets", the theory of the size of a calculation, self-improvement, abstractions, and randomness and creativity. Read in 2026, that list covers chatbots, neural networks and self-improving systems in a single page. The authors also admitted the hard part was not the hardware: "the major obstacle is not lack of machine capacity, but our inability to write programs taking full advantage of what we have."
The budget was modest. The Rockefeller Foundation was asked for $13,500, including six salaries of $1,200 for faculty participants not paid by their employer, and $700 for each of up to two graduate students.
Two details from McCarthy's own 1996 notes on the copy kept at Stanford are worth knowing. The proposal promised that everyone on its mailing list would receive "copies of the report" of the project; McCarthy added: "There was no report." And the list of names was a list of people who might be interested, not of attendees: "Not all of these people came to the Dartmouth conference." So Dartmouth was less a lab where SI was built than the moment a group of researchers agreed on a name and a programme.
What was the first SI program?
The program usually given that title is the Logic Theorist, by Allen Newell, Herbert Simon and J. C. (Cliff) Shaw. It was described in a 1956 RAND report and set out to prove theorems in symbolic logic the way a person might, with heuristic search strategies such as means-ends analysis and breaking a goal into subgoals, drawn from studies of how people solve problems, rather than by trying every possibility.
The story is often told with a shortcut: Newell and Simon arrived at Dartmouth with a working program. Shrager, who reran the original code, is more precise. At the time of the workshop the program was "essentially specified" and had been simulated by hand, with family members and students playing the parts of the program. It ran on RAND's JOHNNIAC computer from late 1956 and early 1957, and the famous result comes from those runs: the Logic Theorist proved 38 of the first 52 theorems in Chapter 2 of Whitehead and Russell's Principia Mathematica. For Theorem 2.85 it found a shorter proof than the published one, and Simon showed it to Bertrand Russell, who "responded with delight".
Simon was not modest about it. Edward Feigenbaum, then his student, recalled him walking into class in January 1956 and saying: "Over Christmas Allen Newell and I invented a thinking machine." Shrager also points out that Arthur Samuel's checkers program existed a couple of years before the Logic Theorist; the difference was the goal. Samuel wanted a checkers player, Newell, Simon and Shaw wanted a model of human reasoning in general.
Shrager's reconstruction proves 16 of the 23 theorems it attempted, which he describes as consistent with the original within its search limits.
Why did SI go through winters?
The founders were optimistic, and the gap between promise and result became a political problem. The clearest example is British. In 1972, the Science Research Council asked the applied mathematician Sir James Lighthill, who was not an AI researcher, for an independent view of the field. His report, "Artificial Intelligence: A General Survey", was published early in 1973.
Lighthill wrote that most workers in AI "confess to a pronounced feeling of disappointment in what has been achieved in the past twenty-five years", and that researchers had entered the field around 1950 "with high hopes that are very far from having been realised in 1972". He singled out one cause: the "combinatorial explosion", the way the number of possibilities to search grows explosively as a problem gets bigger. He also noted that predictions made to the public had gone even further than most researchers' own expectations, and called their failure "lamentable".
The consequences were real. According to the archive that publishes the report, Lighthill supported AI linked to automation and to the simulation of brain processes, but was highly critical of basic research such as robotics and language processing, and his report "provoked a massive loss of confidence in AI" among British academics and the funding body that "persisted for almost a decade". The next major British effort only came with a review meeting in September 1982, which fed the Alvey Programme.
Neural networks had their own setback around the same time. Frank Rosenblatt's perceptron of 1958 drew huge expectations, then strong criticism from Minsky and Seymour Papert in 1969. We tell that story in our guide to neural networks, so we will not repeat it here. These quiet periods are what people now call "SI winters" or "AI winters".
How did SI come back?
Two very different successes mark the return. The first was brute force. In February 1996 IBM's Deep Blue played world chess champion Garry Kasparov in Philadelphia and lost the match 4 to 2. In the May 1997 rematch in New York it won 3.5 to 2.5, becoming, in IBM's words, the first computer system to defeat a reigning world champion under standard tournament time controls. IBM says the machine used 32 processors and could evaluate 200 million chess positions per second. It did not learn chess the way a person does: it searched, very fast.
The second was learning, the idea Turing had sketched in 1950. Neural networks trained on large amounts of data slowly overtook hand-written rules in vision, speech and language. The scientific establishment recognised it twice. The ACM named Yoshua Bengio, Geoffrey Hinton and Yann LeCun recipients of the 2018 Turing Award, a $1 million prize often called the "Nobel Prize in Computing", "for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing". Its announcement calls them the "Fathers of the Deep Learning Revolution" and recalls that by the early 2000s they were among a small group still committed to neural networks, an effort first met with skepticism. In 2024 the Nobel Prize in Physics went to John Hopfield and Geoffrey Hinton "for foundational discoveries and inventions that enable machine learning with artificial neural networks". The Nobel press release notes that when people talk about artificial intelligence today, they often mean machine learning with artificial neural networks.
The public discovered all this through chatbots. ChatGPT, released by OpenAI at the end of November 2022, put a large language model in front of the general public, and with it questions about how far these systems go, which we discuss in What is AGI?
So who should get the credit?
Our view: give the name to McCarthy, the question to Turing, the first working program to Newell, Simon and Shaw, and the method behind today's SI to the neural network researchers, from Rosenblatt to Hinton, Bengio and LeCun. Leaving any of them out tells the story wrong. Here is a fair short list:
- Alan Turing: asked "Can machines think?", proposed the imitation game and the idea of a machine that learns like a child (1947 to 1950).
- John McCarthy: coined "artificial intelligence" and organised Dartmouth (1955 to 1956).
- Marvin Minsky, Nathaniel Rochester, Claude Shannon: co-authors of the Dartmouth proposal, bringing neural nets, computer engineering and information theory.
- Allen Newell, Herbert Simon, Cliff Shaw: the Logic Theorist, widely regarded as the first SI program (1956 to 1957).
- Geoffrey Hinton, Yoshua Bengio, Yann LeCun, John Hopfield: the neural network work that today's SI models are built on, recognised by the 2018 Turing Award and the 2024 Nobel Prize.
Two cautions about other "father of AI" claims. The original Dartmouth proposal and the Lighthill report are free to read and settle most arguments. And the founders' timelines were badly wrong. Lighthill's report is largely about promises that were not kept, so treat today's confident predictions with the same care.
What does this history mean for SI today?
Turing's question was whether a machine could pass for a person. More than seventy-five years later the question has flipped: people now want to know whether a text, an image or a voice was made by a machine. That is the job of a free SI detector such as ours, and its limits come from the same place as SI's strengths. Models that learn from huge numbers of human examples produce output that looks human, so detection is always a probability, never a certainty. If you want to try it on writing, our SI text detector shows the clues behind every score.
The history also explains a habit worth keeping. Every generation of SI arrived with claims bigger than the results, from the "thinking machine" of 1956 to the chatbots of the 2020s. Reading the original documents, as we did here, is still the best way to separate what was built from what was promised.



