Dartmouth Summer Research Project on Artificial Intelligence

The founding document of the field is a budget request. Dated 31 August 1955 and signed by John McCarthy of Dartmouth, Marvin Minsky of Harvard, Nathaniel Rochester of IBM and Claude Shannon of Bell Telephone Laboratories, it opens: "We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College." The sentence that follows is the field's charter, and it is a conjecture, stated as one: 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 audacity sits directly beside the line items. Total estimated expenses: 13,500 dollars — itemized down to 1,200 dollars in salary for each faculty participant not supported by his own organization, 700 dollars for each of up to two graduate students, and 850 dollars for secretarial expenses. The proposal names seven aspects of the problem: Automatic Computers; How Can a Computer be Programmed to Use a Language; Neuron Nets; Theory of the Size of a Calculation; Self-Improvement; Abstractions; and Randomness and Creativity. The ambition for the summer is bounded and plain: "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."

What actually happened in the summer of 1956 is best documented in James Moor's report on the fiftieth-anniversary conference, AI@50, held at Dartmouth on 13–15 July 2006 and attended by five of the original participants: Trenchard More, John McCarthy, Marvin Minsky, Oliver Selfridge and Ray Solomonoff. By their own account the workshop under-delivered as a meeting. "The attendees did not come at the same time and most kept to their own research agenda," and "it was not really a conference in the usual sense. There was no agreement on a general theory of the field and in particular on a general theory of learning." The most concrete research McCarthy could point to from that summer was brought in by visitors: Allen Newell, Cliff Shaw and Herbert Simon's Information Processing Language and the Logic Theory Machine.

What the summer did produce was the name. McCarthy, who is credited with coining "artificial intelligence," explained at AI@50 that he wanted "to nail the flag to the mast." The flag was contested even among the people under it: Trenchard More, who took some of the best notes of the 1956 project, "admitted that he never liked the use of 'artificial' or 'intelligence' as terms for the field." Moor's summary of the founding is the most honest one on record: "The field of AI was launched not by agreement on methodology or choice of problems or general theory, but by the shared vision that computers can be made to perform intelligent tasks."

The anniversary produced its own dated predictions, now themselves part of the record. Ray Kurzweil "claimed that we can be confident of Turing test–capable AI within a quarter century — a prediction with which many disagreed." McCarthy offered that human-level AI "is likely but not assured by 2056." Moor added the field's standing caution: Herbert Simon "once predicted a computer chess champion within 10 years. He was wrong about the 10 years, but it did happen within 40 years."

One item on the 2006 program reads differently now than it did then. Among the more than three dozen presentations, "Simon Osindero working with Geoffrey Hinton discussed more powerful networks" — neural networks being, at that anniversary, one ongoing debate among many about how useful they might ever be. Six years later a network from Hinton's lab won the 2012 ImageNet contest by ten percentage points, and the conjecture of 1955 stopped being a summer project.