Dartmouth 1956

A four-page funding proposal, a half-empty summer workshop and a new name: how “artificial intelligence” became a field. By the end you will know what really happened at Dartmouth, and why it still matters.

Every field likes a founding story. AI’s is set on the top floor of the mathematics department at Dartmouth College, a small university in the woods of New Hampshire, in the summer of 1956. It is often told as a grand conference where the giants of a new science gathered and planned the future. The truth is messier and more interesting: a loosely organised summer of conversations, with people drifting in and out, no shared project and no breakthrough. And yet it gave the field its name, its founding members and its first great ambition.

The proposal (1955)

In 1955 John McCarthy was a 27-year-old assistant professor of mathematics at Dartmouth. He wanted to bring together the scattered researchers working on machines that might think, and he recruited three co-signatories with more standing: Marvin Minsky, then a junior fellow at Harvard; Nathaniel Rochester, who led information research at IBM and had designed its 701 computer; and Claude Shannon of Bell Labs, already famous for information theory. Their proposal to the Rockefeller Foundation, dated 31 August 1955, opened plainly:

“We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College …”McCarthy, Minsky, Rochester and Shannon, Dartmouth proposal (1955)

Then came the sentence that has been quoted ever since — the founding bet of the field. The study, it said, 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.”Dartmouth proposal (1955)

The proposal was also strikingly optimistic. It suggested that “a significant advance” could be made on one or more of its problems if a carefully selected group worked on them together for a summer. The Rockefeller Foundation granted $7,500, a little over half the $13,500 the proposal had budgeted.

It listed seven problem areas. Tap each to see how the 1955 agenda maps on to today:

Why “artificial intelligence”?

The name in the proposal’s first line was new, and McCarthy chose it on purpose. Existing labels each carried baggage. “Automata studies” had disappointed him: when he and Shannon edited a collection of that name (published in 1956), most contributors sent work on the mathematical theory of automata rather than on intelligence. “Cybernetics” was taken, and closely identified with Norbert Wiener and with analogue feedback rather than digital computers. McCarthy later said he wanted to avoid having to accept Wiener as a guru or argue with him.

“Artificial intelligence” was a neutral flag under which anyone could gather. Not everyone liked it. Allen Newell and Herbert Simon, two of the most productive researchers of the period, preferred “complex information processing” for years. Critics have pointed out ever since that the word “intelligence” promises a great deal, and that the name itself may have fed the cycles of hype described in AI History at a Glance.

Who actually came

Popular myth The workshop is often pictured as ten founders meeting for a summer. In reality attendance was patchy. The workshop ran for roughly eight weeks, from about mid-June to mid-August 1956, and only three people stayed for the whole time: McCarthy, Minsky and Ray Solomonoff, a mathematician interested in how machines could learn by induction. Solomonoff’s own notes, kept during the summer, are our best record of who came.

Others came for days or weeks. Shannon and Rochester attended part of the time; Trenchard More, a young mathematician, replaced Rochester for three weeks. Oliver Selfridge, a pioneer of pattern recognition, and Arthur Samuel of IBM, whose checkers-playing program learned from experience, both took part, as did Alex Bernstein, who was writing a chess program at IBM. Solomonoff’s list also includes the cybernetician W. Ross Ashby, Warren McCulloch, the computer engineer Julian Bigelow, the logician Abraham Robinson and the mathematician John Nash. Newell and Simon were invited for the first two weeks, but how long they actually stayed is not well recorded. On a typical day there were between three and about eight people in the room.

There were no women among the participants, and nearly all came from a handful of American institutions — a narrowness that historians now note as part of the story.

Logic Theorist: the program that worked

The most concrete achievement on display at Dartmouth came from two of the short-stay visitors. At the RAND Corporation and the Carnegie Institute of Technology, Allen Newell and Herbert Simon, with the programmer Cliff Shaw, had built the Logic Theorist, often described as the first AI program. It proved theorems in symbolic logic by searching for chains of reasoning, guided by heuristics — rules of thumb that cut down the number of possibilities to try. Before it ran on a computer, they tested it by hand, with Simon’s family and students each acting out part of the program.

Logic Theorist eventually proved 38 of the first 52 theorems in chapter 2 of Whitehead and Russell’s Principia Mathematica, finding a shorter proof than the original for one of them. In January 1956 Simon told a class: “Over Christmas, Al Newell and I invented a thinking machine.” At Dartmouth, by most accounts, the reception was cooler than they expected: others were still talking about ideas, while Newell and Simon arrived with a working program and were not much interested in anyone else’s plans.

What it did and did not achieve

  • A name — artificial intelligence — and with it a recognisable field to fund and join.
  • A community: the people who would lead AI for the next twenty years met, argued and stayed in touch.
  • A shared ambition, set out in the conjecture, that learning and intelligence could be described precisely enough to be programmed.

Debated Was Dartmouth really the birth of AI?

Other views
  • Yes, as a field: the ideas were older, but a field needs a name, a community and funding, and Dartmouth supplied all three. This is the standard account.
  • No, it was a christening: Turing’s 1950 paper, the cybernetics movement, Shannon’s chess paper, Samuel’s checkers program and the 1955 Western Joint Computer Conference session on learning machines all came first. Dartmouth named a field that already existed.
  • The story is too American: British work — Turing, the Ratio Club, early learning machines — is often written out of a founding myth centred on one US campus.
Quick check

Who stayed at Dartmouth for the whole summer of 1956?

Attendance was patchy; according to Solomonoff’s notes, only these three were there throughout.

Why did McCarthy choose the name “artificial intelligence”?

Turing’s paper was “Computing Machinery and Intelligence”. McCarthy wanted a fresh, neutral banner.

What did Logic Theorist do?

It searched for proofs guided by rules of thumb, proving 38 of the first 52 theorems of chapter 2.

// Pause and reflect

  1. The Dartmouth conjecture says every feature of intelligence can “in principle” be precisely described. Do today’s learning systems support that claim, or quietly abandon it?
  2. How might the field have developed differently if McCarthy had chosen a more modest name?
  3. Why do fields need founding stories, and what gets lost when a messy summer becomes a legend?

Your notes are saved privately in this browser.

What happened next

The years after Dartmouth were a time of astonishing early programs and even more astonishing predictions: LISP, the perceptron, ELIZA and a robot called Shakey, all paid for by generous government grants. Continue with The Golden Years.

Key terms on this page

Sources & further reading