The Dartmouth workshop, formally the Dartmouth Summer Research Project on Artificial Intelligence, was a research gathering held in the summer of 1956 at Dartmouth College in Hanover, New Hampshire, United States. Organized around a proposal by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, it brought together researchers investigating whether machines could perform activities associated with human intelligence. It is widely regarded as a founding event of artificial intelligence as a distinct research field, although its constituent ideas and experimental programs had earlier origins. (ojs.aaai.org)
Background and organization
The workshop emerged from overlapping investigations into computation, intelligent behavior, and mathematical models of machines. Before 1956, Shannon had proposed computer chess and constructed experimental devices, while Minsky had investigated neural computation. McCarthy and Shannon also edited Automata Studies, published in 1956. McCarthy later recalled that too few contributions addressed his central interest: making machines behave intelligently. The workshop was intended to focus attention directly on that problem rather than on the broader study of automata. (www-formal.stanford.edu)
McCarthy joined Dartmouth in 1954 and spent the summer of 1955 with Rochester’s research group at IBM. During that visit, they enlisted Minsky and Shannon in preparing a funding proposal. The four represented Dartmouth, Harvard University, IBM, and Bell Telephone Laboratories respectively, linking university research with industrial computing and communications laboratories. (www-formal.stanford.edu)
Their proposal, dated August 31, 1955, requested a two-month study involving ten researchers during the following summer. The Rockefeller Foundation awarded $7,500 toward the project. This distinction between the 1955 proposal and the 1956 gathering is important: the expression “artificial intelligence” appeared in the proposal before the workshop took place. (ojs.aaai.org)
Research agenda
The proposal’s central conjecture was that learning and other features of intelligence could, in principle, be described precisely enough for a machine to simulate them. It proposed investigating machines that could use language, construct abstractions and concepts, solve problems associated with human intelligence, and improve their own performance. These were research ambitions, not results already demonstrated by the organizers. (www-formal.stanford.edu)
Seven topics structured the proposed investigation: automatic computers; machine use of language; neural networks; the size of a calculation; self-improvement; abstractions; and randomness and creativity. The discussion of calculation size concerned the resources required to solve problems, anticipating questions associated with computational complexity. The neural-network section asked how networks could be arranged to form concepts, while the creativity section considered whether controlled randomness could contribute to inventive behavior. (www-formal.stanford.edu)
The agenda was therefore broader than symbolic artificial intelligence alone. It combined questions about formal reasoning and language with learning, neural organization, and adaptive behavior. Its unifying commitment was to investigate intelligence through explicit mechanisms that machines could implement, rather than to prescribe one exclusive technical method. (www-formal.stanford.edu)
Participants and working format
Despite the common designation “conference,” the gathering was an informal summer workshop rather than a tightly scheduled sequence of presentations. Researchers arrived at different times and stayed for different periods. McCarthy later explained that the intended collective effort was limited both by reduced funding and by participants’ existing research agendas. It consequently did not operate as a single team working continuously toward one agreed solution. (www-formal.stanford.edu)
Researchers associated with the workshop included Ray Solomonoff, Oliver Selfridge, Trenchard More, Allen Newell, and Herbert A. Simon. Discussions took place in Dartmouth Hall. Later recollections describe small-group exchanges about terminology and problem-solving methods, including the meaning and usefulness of heuristics—guides that help direct a search without necessarily guaranteeing success. The varying attendance makes a fixed participant count less informative than identifying who contributed and for how long. (spectrum.ieee.org)
Logic Theorist and computational reasoning
One prominent contribution was Newell and Simon’s presentation of the Logic Theorist, developed with Cliff Shaw. The program sought proofs of statements in propositional logic, demonstrating that computation could manipulate symbolic expressions rather than merely carry out numerical calculations. Its work connected mathematical proof with the study of human problem solving. (ocs.aaai.org)
The program used heuristic guidance to select promising steps during proof search. Newell and Simon also described list structures for representing formulas and their Information Processing Language. According to McCarthy’s recollection, these representations impressed him even though he preferred a different programming approach. The presentation thus contributed both a working example of machine reasoning and techniques for organizing symbolic information. (www-formal.stanford.edu)
Historical significance and documentation
Dartmouth’s importance lies chiefly in establishing a recognizable name and research agenda, not in completing a general theory of intelligence during one summer. McCarthy’s retrospective account distinguishes the ambitious plan for collective progress from the more fragmented gathering that occurred. Earlier researchers, including Alan Turing, had already investigated machine intelligence; Dartmouth provided a focal event for researchers pursuing related problems. (www-formal.stanford.edu)
The original proposal survives in archival copies at Dartmouth and Stanford and was republished in AI Magazine in 2006. On July 13–15, 2006, Dartmouth held a fiftieth-anniversary conference, AI@50, bringing back several original participants and examining subsequent developments and future research questions. These records and retrospective discussions document both the workshop’s ambitions and the differences between its planned organization and actual proceedings. (ojs.aaai.org)