Bitware News
History · Ch. 022026-09-13

Dartmouth Names a Research Program

The August 1955 proposal and the 1956 Dartmouth workshop named and organized a research program under the phrase 'artificial intelligence'-they did not invent the problems surveyed in the previous chapter.

History

Primary source: McCarthy et al., Dartmouth proposal (31 Aug 1955): https://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html

The previous chapter surveyed mid-century work already in print-logical neurons, feedback purpose, information theory, and Turing’s operational imitation game. This chapter is about a different kind of event: naming and organizing. On 31 August 1955, John McCarthy, Marvin L. Minsky, Nathaniel Rochester, and Claude E. Shannon dated a proposal for a summer study at Dartmouth College. The phrase they put in the title was “artificial intelligence.” The summer meeting that followed, in Hanover, New Hampshire, in 1956, is often retold as the birth of the field. The primary documents support a narrower claim. Dartmouth named a research program and gathered people who were already working on related problems. It is not the origin of those problems, and the summer did not deliver the collective “significant advance” the proposal hoped for.

The proposal, not the myth

The typescript is short and direct. As reprinted from McCarthy’s Stanford archive and in AI Magazine (vol. 27, no. 4, Winter 2006), the opening paragraph reads:

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. 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. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer.

That is the conjecture, quoted as printed-not a slogan about digital minds, but a methodological bet that intelligence can be described precisely enough for machine simulation. The affiliations on the proposal are likewise specific: McCarthy, Dartmouth College; Minsky, Harvard University; Rochester, I.B.M. Corporation; Shannon, Bell Telephone Laboratories.

The proposal then lists seven “aspects of the artificial intelligence problem,” again in the proposers’ own headings and order.

1. Automatic Computers. If a machine can do a job, an automatic calculator can be programmed to simulate the machine. Present speeds and memory may be insufficient for many higher functions of the human brain, but “the major obstacle is not lack of machine capacity, but our inability to write programs taking full advantage of what we have.”

2. How Can a Computer be Programmed to Use a Language. Much human thought, they speculate, consists of manipulating words by rules of reasoning and conjecture; forming a generalization would mean admitting a new word and rules relating sentences that contain it. The idea, they note, had never been precisely formulated nor worked out in examples.

3. Neuron Nets. How can a set of (hypothetical) neurons be arranged so as to form concepts? The proposal cites prior theoretical and experimental work-Uttley; Rashevsky and his group; Farley and Clark; Pitts and McCulloch; Minsky, Rochester and Holland; and others-and says partial results exist but more theoretical work is needed. (This is a research heading in 1955, continuous with the logical-neuron literature of the previous chapter. It is not yet Frank Rosenblatt’s Perceptron program of the late 1950s, which belongs to a later chapter.)

4. Theory of the Size of a Calculation. For a well-defined problem-one for which a proposed answer can be tested mechanically-trying all answers in order is inefficient. Efficiency needs a measure of calculation complexity, which in turn needs a theory of the complexity of functions. Partial results are attributed to Shannon and to McCarthy.

5. Self-Improvement. A truly intelligent machine will likely carry out activities best described as self-improvement; some schemes exist and are worth further study, including abstract study.

6. Abstractions. Several types of abstraction can be distinctly defined, others less so; classifying them and describing machine methods of forming abstractions from sensory and other data would seem worthwhile.

7. Randomness and Creativity. An “attractive and yet clearly incomplete conjecture” is that the difference between creative thinking and unimaginative competent thinking lies in injecting some randomness, guided by intuition so as to be efficient-“the educated guess or the hunch.”

The proposal also attaches individual statements of interest from the four originators and a budget request to the Rockefeller Foundation (the typed estimate totals $13,500). Organization was to mix regular research seminars with individual and informal small-group work. The framing is institutional: two months, about ten people, a shared conjecture, and a hope of advance. It is a research-program charter, not a founding myth.

What a historian can document about 1956

The meeting took place at Dartmouth College in the summer of 1956. The proposal’s “2 month, 10 man” language describes what was asked for. What happened is harder to pin to a single neat roster, and careful sources say so.

McCarthy later stated that Rockefeller support was substantially less than requested, that participants kept their own research agendas, and that people therefore came at varied times and for varying lengths of time; he also noted that artificial intelligence presented more difficulties than were known in 1956. Participant notes preserved by Ray Solomonoff’s archive describe a workshop lasting roughly eight weeks (about 18 June to 17 August 1956), with daily sessions often small. Nils Nilsson’s standard history (The Quest for Artificial Intelligence, Cambridge University Press, 2010) summarizes the event as more of a rolling workshop than a continuous summer “study,” and places Rockefeller funding behind a gathering that included the four proposers plus others who came for parts of the summer.

Attendee lists conflict. Prefer the conservative, documented sets over blog compilations.

A conservative narrative, then: the four proposers; Solomonoff, Selfridge, More, and others on Solomonoff’s list; Newell and Simon for an early portion of the summer; visitors such as Ashby and McCulloch appearing in week-by-week notes. Inflated legendary rosters that treat every name on every mailing list as a full-summer participant should be flagged as folklore.

Logic Theorist: brought to Dartmouth, not invented there

Allen Newell and Herbert A. Simon (with J. C. Shaw at RAND) had been developing what became the Logic Theorist-a program for proving theorems in the propositional calculus of Whitehead and Russell’s Principia Mathematica, using heuristic search over symbol structures. Their paper The logic theory machine: A complex information processing system appeared in IRE Transactions on Information Theory, vol. IT-2, no. 3 (September 1956), pp. 61–79.

The timing relative to Dartmouth must be stated carefully. McCarthy’s later retrospective slides state that the most important results presented at the meeting were those of Newell, Simon, and Shaw-list processing (IPL) and the Logic Theory Machine-and that this work “had been done previously.” Nilsson likewise treats LT as work Newell and Simon brought into the Dartmouth circle and presented there, concrete evidence for symbol-structure processing and heuristics, not as a program first conceived during the Hanover sessions. Simon’s own later reminiscences date hand simulation of an early proof to December 1955. So: they had Logic Theorist, they were in this circle, and they presented that work in connection with Dartmouth 1956. They did not invent LT at the workshop.

Naming, cybernetics, and what the summer did not do

The name “artificial intelligence” appears in the 1955 proposal’s title and body. That documentary fact is solid. Later commentary often narrates the name as a deliberate escape from Norbert Wiener’s cybernetics. A primary McCarthy statement does exist: in his review of Brian Bloomfield’s The Question of Artificial Intelligence, McCarthy wrote that one reason for inventing the term was “to escape association with ‘cybernetics’. Its concentration on analog feedback seemed misguided, and I wished to avoid having either to accept Norbert … Wiener as a guru or having to argue with him.” Separately, in slides on Dartmouth, McCarthy emphasized another motive: disappointment that Automata Studies (the Shannon–McCarthy solicited volume, published 1956) contained too little on making machines behave intelligently, so that “artificial intelligence” was chosen “to nail the flag to the mast.” Both remarks are retrospective. The relationship of the new name to cybernetics is therefore documentable as McCarthy’s later self-description, but it is often over-narrated as if a single quarrel explained the field. The proposal text itself does not stage that quarrel; it states a conjecture and a topic list.

Folklore to refuse as fact:

What Dartmouth did do, on the documentary record, is put a name on a conjecture, circulate a topic map that mixed programming, language, neuron nets, complexity, self-improvement, abstraction, and randomness, and bring overlapping research communities into the same rooms for parts of a summer. The next chapter turns to learning machines and Perceptrons-work that intersects the proposal’s “neuron nets” heading but has its own 1957–58 chronology and should not be collapsed into the 1955–56 naming event.

Caveat

Three retellings especially distort this material.

First, founding-myth compression: treating August 1955 / summer 1956 as the origin of machine-intelligence problems erases the printed prehistory of the previous chapter.

Second, success inflation: quoting the proposal’s hope of a “significant advance” as if it were a post-meeting result. The advance claimed in retrospect is institutional and nominal-a research program under a shared phrase-not a demonstrated general simulation of intelligence.

Third, roster inflation and motive over-narration: attendance must be tied to primary notes and planning letters; McCarthy’s cybernetics remarks are real but later, and they are one strand among others (including the Automata Studies disappointment) in how the name was later explained.

Sources

Primary

Historical reference