When Did AI Start?

The Genesis of Artificial Intelligence

Artificial intelligence officially began as a distinct academic discipline in the summer of 1956 at Dartmouth College in Hanover, New Hampshire. During the Dartmouth Summer Research Project on Artificial Intelligence, organized by mathematician and computer scientist John McCarthy alongside Marvin Minsky, Nathaniel Rochester, and Claude Shannon, the term "artificial intelligence" was formally coined and adopted. Dartmouth workshop - Wikipedia Artificial Intelligence (AI) Coined at Dartmouth

The intellectual foundation of AI, however, developed through multiple phases across philosophy, mathematics, and early computer engineering prior to 1956:

• Philosophical and Mathematical Roots (Antiquity to 1930s): Formal reasoning concepts originated with Aristotle's syllogistic logic, Gottfried Wilhelm Leibniz's calculus of thought (calculus ratiocinator), and George Boole's symbolic algebra, culminating in Alan Turing's 1936 mathematical formulation of universal computation.
• Early Computational Models (1943–1949): Warren McCulloch and Walter Pitts proposed the first mathematical model of artificial neural networks in 1943, and Donald Hebb formulated synaptic learning mechanisms in 1949.
• Foundational Conceptualization (1950): Alan Turing published his landmark paper, "Computing Machinery and Intelligence," establishing the operational benchmark known as the Turing Test. Turing's 'Computing Machinery and Intelligence' - AI Timeline
• Formal Disciplinary Birth (1956): The Dartmouth workshop defined the field's objectives, brought together pioneering researchers, and demonstrated the first working automated reasoning program, the Logic Theorist. Dartmouth workshop - Wikipedia Artificial Intelligence (AI) Coined at Dartmouth

Chronological Timeline of AI Origins

1936 ── Turing's Universal Machine (Computable Numbers)
1943 ── McCulloch & Pitts: First Mathematical Neural Network
1948 ── Norbert Wiener: Cybernetics and Feedback Loops
1950 ── Alan Turing: "Computing Machinery and Intelligence" (Turing Test)
1951 ── Ferranti Mark 1 Checkers & Chess Programs; Minsky's SNARC
1955 ── McCarthy, Minsky, Shannon, Rochester draft the Dartmouth Proposal
1956 ── Dartmouth Summer Research Project: AI Formally Established

Key Milestones Preceding the 1956 Birth

1. McCulloch-Pitts Neurons (1943)

Neurophysiologist Warren McCulloch and logician Walter Pitts published "A Logical Calculus of the Ideas Immanent in Nervous Activity". This work demonstrated that simplified biological neurons could be modeled as binary electrical switches performing fundamental logical operations (AND, OR, NOT). Their model proved that networks of artificial neurons could theoretically compute any computable function.

2. Cybernetics and Information Theory (1948)

Norbert Wiener published Cybernetics: Or Control and Communication in the Animal and the Machine in 1948, creating a unifying theory of control systems, biological feedback mechanisms, and automated computation. Simultaneously, Claude Shannon published A Mathematical Theory of Communication, formalizing the quantitative measurement of information, data transmission, and signal processing.

3. Turing’s "Computing Machinery and Intelligence" (1950)

In 1950, British mathematician Alan Turing transitioned machine intelligence from abstract mathematics to practical inquiry by publishing "Computing Machinery and Intelligence" in the philosophical journal Mind. Turing bypassed subjective debates regarding whether machines could "think" by introducing an operational benchmark called the Imitation Game (now known as the Turing Test). If a machine communicating via text can convince a human interrogator that it is human, it demonstrates operational intelligence. Turing's 'Computing Machinery and Intelligence' - AI Timeline

4. Early Working Prototypes (1951–1952)

Between 1951 and 1952, the earliest experimental programs executing game-playing heuristics and neural network simulations emerged:

• SNARC (1951): Marvin Minsky and Dean Edmonds constructed the Stochastic Neural Analog Reinforcement Calculator (SNARC), the first hardware neural network machine built with 40 simulated synapses using vacuum tubes.
• Early Game Programs (1951–1952): Christopher Strachey wrote a draughts (checkers) program, and Dietrich Prinz wrote a chess-playing program on the Ferranti Mark 1 computer at the University of Manchester. Arthur Samuel began developing his self-learning checkers program at IBM shortly thereafter.

The 1956 Dartmouth Workshop: The Turning Point

In August 1931, preparation began for an intensive summer gathering. On August 31, 1955, John McCarthy (Dartmouth), Marvin Minsky (Harvard), Nathaniel Rochester (IBM), and Claude Shannon (Bell Telephone Laboratories) submitted a formal grant application to the Rockefeller Foundation titled "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence." Dartmouth workshop - Wikipedia Full article: AI TURNS FIFTY: REVISITING ITS ORIGINS

"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. 
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."
— Dartmouth Proposal (1955)

Key Workshop Details and Participants

During this gathering, Allen Newell, Herbert Simon, and Cliff Shaw demonstrated the Logic Theorist, an automated program running on the RAND Corporation's JOHNNIAC computer. The program successfully proved 38 of the first 52 theorems in Alfred North Whitehead and Bertrand Russell’s foundational text Principia Mathematica, providing empirical evidence that machines could perform non-numerical, symbolic reasoning. Dartmouth workshop - Wikipedia

Major Intellectual Schools in Early AI

The Dartmouth workshop and the decade that followed crystallized two distinct architectural paradigms that shaped AI research for decades:

                            AI Origins (1950s)
                                    │
          ┌─────────────────────────┴─────────────────────────┐
          ▼                                                   ▼
   Symbolic AI (Top-Down)                         Connectionist AI (Bottom-Up)
  • Logic Theorist & GPS                          • Perceptrons (Rosenblatt)
  • Knowledge representation                       • Neural networks
  • Explicit rule manipulation                    • Statistical pattern learning

1. Symbolic (Top-Down) AI

Symbolic AI (often referred to as Good Old-Fashioned AI or GOFAI) treated cognition as the high-level manipulation of abstract symbols and formal logical rules. Led by researchers like John McCarthy, Allen Newell, and Herbert Simon, this paradigm produced theorem provers, domain-specific rule engines, and automated problem solvers (such as the General Problem Solver).

2. Connectionist (Bottom-Up) AI

The connectionist approach sought to recreate biological neural structures from the bottom up, allowing intelligence to emerge from simple interconnected nodes processing continuous numerical weights. In 1957, Frank Rosenblatt built the Perceptron at the Cornell Aeronautical Laboratory, which laid the direct architectural foundation for modern deep learning models.

Summary of AI's Emergence

While the conceptual foundations of automated intelligence date back through centuries of mathematical logic and Alan Turing's 1950 computational theories, artificial intelligence began as a formalized academic field in 1956 at Dartmouth College. The Dartmouth workshop unified disparate research threads in cybernetics, automata theory, and neural modeling into a coherent scientific discipline that drove modern computing. Dartmouth workshop - Wikipedia Artificial Intelligence (AI) Coined at Dartmouth Turing's 'Computing Machinery and Intelligence' - AI Timeline

The short answer

Artificial intelligence as a named research field began in the summer of 1956, at a workshop held at Dartmouth College in Hanover, New Hampshire. The term "artificial intelligence" itself was coined slightly earlier, in the August 31, 1955 funding proposal that John McCarthy wrote with Marvin Minsky, Nathaniel Rochester, and Claude Shannon to request support for that workshop. The Dartmouth Summer Research Project on Artificial Intelligence is therefore the conventional founding event of the discipline, and 1956 is the date most histories give when asked when AI began. Artificial Intelligence (AI) Coined at Dartmouth Dartmouth workshop A Proposal for the Dartmouth Summer Research Project on ...

But that answer is a convention, not a hard boundary. The ideas that made 1956 possible were assembled over the preceding two decades — and in a looser sense over centuries. If you want to know how AI started rather than merely when a label was attached to it, the more useful framing is that AI emerged from the collision of mathematical logic, the theory of computation, neuroscience, and the arrival of programmable electronic computers in the 1940s.

What actually happened at Dartmouth

McCarthy, then a young mathematician at Dartmouth, organized a two-month gathering of roughly ten researchers. The proposal's central premise was strikingly confident: the work would "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 agenda listed problems that remain recognizable today — getting machines to use language, form abstractions and concepts, solve problems reserved for humans, and improve themselves. Artificial Intelligence (AI) Coined at Dartmouth A Proposal for the Dartmouth Summer Research Project on ...

The workshop was not a conference in the modern sense. Attendance was fluid, there were no proceedings, and no single technical breakthrough was produced on site. Its historical importance is social and terminological: it gave a scattered set of researchers a shared name, a shared research program, and a network that would go on to found the major AI laboratories at MIT, Carnegie Mellon, and Stanford. McCarthy reportedly chose "artificial intelligence" partly to distinguish the effort from cybernetics and from Norbert Wiener's intellectual orbit. Dartmouth workshop

The prehistory: where the ideas came from

Ancient impulses and formal logic

Stories of artificial beings — Talos in Greek myth, the golem, mechanical automata in Chinese, Islamic, and European craft traditions — show that the desire for thinking machines is very old. These are cultural antecedents rather than technical ones, but historians of the field routinely note them because they shaped how the modern project was imagined and received. The genuinely enabling contribution was the formalization of reasoning: George Boole's algebra of logic in the nineteenth century, Gottlob Frege's predicate logic, and the twentieth-century program of reducing mathematical proof to symbol manipulation. Once reasoning could be represented as rule-governed operations on symbols, a machine that manipulates symbols became a candidate for reasoning. history of artificial intelligence (AI)

The 1930s and 1940s: computation, neurons, and cybernetics

Three strands converged just before and during the Second World War:

• Computability. Alan Turing's 1936 analysis of what an idealized machine can compute established that a single universal device could, in principle, carry out any effective procedure. This is the theoretical license for the whole field.
• Neural modeling. In 1943, Warren McCulloch and Walter Pitts, working at the University of Chicago, published the first mathematical model of a neuron and showed that networks of such simplified units could compute logical functions. This paper is the ancestor of every artificial neural network. Birthplace of Neural Networks: McCulloch & Pitts at UChicago
• Cybernetics and control. Wiener, Shannon, W. Ross Ashby, and others developed feedback, information, and self-regulation as general scientific concepts, and built early goal-seeking machines.

1950: Turing's question

In 1950 Turing published "Computing Machinery and Intelligence," which opened with the question "Can machines think?" and replaced it with an operational test — now called the Turing test — in which a machine passes if an interrogator cannot reliably distinguish its written responses from a human's. Turing also anticipated the main objections, discussed machine learning and child-like training, and predicted rough timelines. Many people reasonably date the modern intellectual start of AI to this paper rather than to Dartmouth, because it framed the problem, the goal, and the method of evaluation. The birth of Artificial Intelligence (AI) research - LLNL history of artificial intelligence (AI)

The early 1950s also saw the first working programs that played games well enough to be interesting — draughts and chess programs on British and American machines, and Arthur Samuel's checkers work at IBM, which introduced self-improvement through play and helped popularize the phrase "machine learning" later in the decade.

The first golden age: 1956 to the early 1970s

Momentum after Dartmouth was rapid, and some of the most influential early systems predate or immediately follow the workshop.

Year	Development	Significance
1956	Logic Theorist (Allen Newell, Herbert Simon, J. C. Shaw)	Widely described as the first program deliberately engineered to do automated reasoning; it proved theorems from Principia Mathematica Logic Theorist - Wikipedia
1958	LISP (McCarthy)	The dominant AI programming language for decades
1958	Perceptron (Frank Rosenblatt)	An early trainable neural network that adjusted weights from examples AI and Machine Learning: A Historical Perspective
1965–1966	ELIZA (Joseph Weizenbaum)	Pattern-matching conversational program, often called the first chatbot; its persuasiveness surprised its own author Special Issue — Call for Papers
Late 1960s	Shakey the robot (SRI)	First mobile robot to reason about its own actions
1960s–1970s	DENDRAL, MYCIN, SHRDLU	Knowledge-based reasoning in narrow domains

This period established the two research traditions that still structure the field: symbolic AI, which represents knowledge as explicit symbols and rules, and connectionism, which learns statistical patterns in networks of simple units. The rivalry between them — including Minsky and Seymour Papert's 1969 critique of perceptrons — explains much of the field's subsequent boom-and-bust rhythm.

Winters, expert systems, and the long middle

Early optimism outran results. Machine translation disappointed its US sponsors in the mid-1960s; in Britain, the 1973 Lighthill report concluded that AI had not delivered on its promises, and government support for most British AI research was withdrawn, contributing to what is now called the first AI winter. DARPA also cut back general academic AI funding in the 1970s. A second downturn followed in the late 1980s with the collapse of the specialized LISP machine market and the retrenchment of expert-system spending. AI winter

These winters matter for the "when did AI start" question because they explain why AI appears to have several starts. The field did not grow smoothly from 1956; it restarted commercially in the early 1980s with rule-based expert systems, restarted again in the 1990s under different names — machine learning, data mining, statistical pattern recognition, often deliberately avoiding the tainted "AI" label — and restarted publicly in the 2010s. The 1997 defeat of world chess champion Garry Kasparov by IBM's Deep Blue was a widely publicized marker of machine capability in that quieter middle period, although Deep Blue relied on massive specialized search rather than learning. Deep Blue

When did modern AI start?

For most people asking today, the practical question is when the AI they actually encounter began. Three dates dominate:

1. 2012 — the deep learning breakthrough. AlexNet, a deep convolutional neural network, won the ImageNet image-classification competition by a large margin over the nearest competitor, and did so by training on GPUs. This result convinced the wider research community that deep neural networks plus large labeled datasets plus parallel hardware was a winning combination, and it reversed decades of skepticism about connectionism. AlexNet and ImageNet: The Birth of Deep Learning
2. 2017 — the Transformer. The paper "Attention Is All You Need," from a team of Google researchers, proposed an architecture based purely on attention mechanisms, dispensing with recurrence and convolution. Transformers became the substrate for large language models and for much of modern generative AI. Attention Is All You Need
3. Late 2022 — the consumer moment. The public release of ChatGPT at the end of November 2022 made conversational large language models a mass-market experience, and for a great many users this is when "AI started" in any lived sense.

Between and around these, AlphaGo's 2016 victories over top human Go players and the appearance of large pre-trained language models from 2018 onward marked the shift from narrow, hand-built systems to general-purpose models adapted to many tasks.

Choosing a start date honestly

Because "AI" names a label, a research program, a set of technologies, and a consumer experience, the honest answer depends on what you mean:

• The name and the field: 1955 (coinage) / 1956 (Dartmouth workshop).
• The intellectual foundation: 1936–1950 (Turing's computability work and the Turing test), with 1943 for neural network theory.
• The first working AI programs: the early-to-mid 1950s, with the Logic Theorist in 1956.
• Modern machine-learning AI: roughly 1986–2006 for the methods, 2012 for the decisive empirical turn.
• Generative AI in public life: 2017 for the architecture, 2022–2023 for mass adoption.

Any of these can be defended; what should be avoided is presenting one as the only correct answer, or implying that the pre-1956 work was mere speculation. Historians of computing generally treat the boundaries as fuzzy by nature, since the field's identity was contested from the beginning.

Where did AI start geographically

The institutional origin is Anglo-American. The theoretical groundwork was heavily British (Turing at Cambridge, Bletchley Park, and Manchester; early machine-intelligence work at Cambridge and Edinburgh), while the organized field was launched in the United States: Dartmouth in 1956, then long-running laboratories at MIT, Carnegie Mellon in Pittsburgh, Stanford, and SRI, largely funded by US defense agencies during the Cold War. Chicago contributed the founding neural-network model. Artificial Intelligence (AI) Coined at Dartmouth Birthplace of Neural Networks: McCulloch & Pitts at UChicago

That concentration did not last. The Soviet Union, Japan, and several European countries pursued substantial programs from the 1960s onward — Japan's Fifth Generation Computer Systems project in the 1980s being the most prominent national effort of its era. Today research capacity is distributed across North America, Europe, China, and elsewhere, and several key modern contributions came from labs in Canada, the United Kingdom, Switzerland, and China. Attributing AI to a single place is only accurate for its founding moment, not for the field as it now exists.