# What actually happened at the 1956 Dartmouth workshop on artificial intelligence?

Date: 2019-01-22 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-happened-at-the-dartmouth-workshop/

**Answer:** From about June 18 to August 17, 1956, a shifting group of roughly twenty researchers met at Dartmouth College to discuss how machines might learn and reason, with only three of them staying the whole summer. They never agreed on a joint project, but the meeting gave the new field its name, its first community and its early focus on symbolic reasoning and search.

The "Dartmouth Summer Research Project on Artificial Intelligence" is often described as the moment AI was born. That makes it sound like a grand conference with a program and a closing declaration. It was nothing of the kind. It was a long, loose summer of conversations in a few rooms at a small New England college, with people arriving and leaving all the time.

The plan came from a 1955 proposal signed by John McCarthy, Marvin Minsky, Claude Shannon and Nathaniel Rochester, which asked for "a 2 month, 10 man study of artificial intelligence." What actually happened in 1956 was messier, and in some ways more interesting, than that tidy plan.

## When and where it took place

The meetings lasted close to eight weeks, beginning around June 18, 1956, and ending around August 17. Some accounts say six weeks, but the notebooks of Ray Solomonoff, one of the participants, along with notes from Minsky and Trenchard More, point to the longer span. Solomonoff gave a closing talk on August 17.

The group had the whole top floor of Dartmouth's mathematics department in Hanover, New Hampshire. On most weekdays they met in the main math classroom, where one person would lead a discussion about his own ideas. Most visitors stayed at the Hanover Inn. McCarthy, then a young mathematics professor at Dartmouth, already lived in town.

## Who came

In May 1956, McCarthy told the Rockefeller Foundation, which was paying for the summer, that eleven people planned to attend. The reality was different. Only three people stayed for the full eight weeks: McCarthy, Minsky and Solomonoff. Others came for a few weeks, a few days, or a single afternoon.

Solomonoff's own list of those who actually appeared has twenty names. Among them:

| Participant | Background in 1956 | What he brought |
|---|---|---|
| John McCarthy | Dartmouth mathematician, organizer | Interest in reasoning programs and chess |
| Marvin Minsky | Harvard, co-organizer | Ideas for a machine to prove geometry theorems |
| Claude Shannon | Bell Labs, founder of information theory | Information theory applied to machines |
| Ray Solomonoff | Independent researcher | Machines that learn by induction and probability |
| Trenchard More | Student standing in for Nathaniel Rochester of IBM | Programs for deductive logic |
| Allen Newell and Herbert Simon | RAND and Carnegie Tech | The Logic Theorist, a working proof program |
| Arthur Samuel | IBM | A checkers program that learns |
| Oliver Selfridge, W. Ross Ashby, Warren McCulloch, John Nash, Alex Bernstein | Various | Pattern recognition, cybernetics, neurons, mathematics, chess |

Daily sessions usually drew between three and eight people. In one week in July, only four people came all week.

## What they did all summer

The sessions were discussions, not lectures. Ashby described his homeostat, an electromechanical device that adjusted itself to stay stable. Samuel talked about checkers and how to score board positions. Bernstein discussed chess. More presented his work on symbols in deductive logic. Newell and Simon described the Logic Theorist, the only program present that actually ran and proved things.

The group even tried to pin down its own vocabulary. One afternoon they looked up the word "heuristic" in a large dictionary in the math room. That word, meaning a rule of thumb that guides a search, would become one of the field's central ideas.

McCarthy hoped the summer would produce a shared project. On August 7 he sounded people out about working together on chess, perhaps on IBM's 704 computer. The answers were lukewarm. Shannon liked chess but doubted its relevance; Minsky strongly opposed it and preferred geometry; Rochester suggested probability problems; Samuel thought checkers would do as well. No group project came of it.

Meanwhile, individuals got real work done. Solomonoff filled about 175 pages on induction and wrote "An Inductive Inference Machine," his first report on learning machines. McCarthy, listening to Newell and Simon describe their list-processing language, began thinking about the kind of language he wanted for AI work.

> **Try it:** Pick a hard problem at your work or school and invite three people to discuss it for one hour, each leading for fifteen minutes. Afterward, write down one idea you got from someone else. That is the Dartmouth format in miniature.

## What came out of it

Measured against the proposal, the summer fell short. The Rockefeller Foundation granted only about half the money requested, people could not stay long, and the organizers' hope for a joint result never materialized. Solomonoff's private summary was blunt: the project had not been very suggestive, but he had met interesting people and written his report.

Measured another way, its effects were large. The name "artificial intelligence" stuck and gave scattered researchers a shared identity. The people who met that summer, especially McCarthy, Minsky, Newell and Simon, went on to lead the field's main laboratories for decades. The Logic Theorist, presented at a symposium at MIT that September, gave the new field a concrete model: intelligence as symbol manipulation and heuristic search. Several historians see the summer as the point where attention shifted from brain-like networks toward symbolic reasoning. Minsky himself moved in that direction after his talks with Solomonoff.

The workshop also fed specific inventions. McCarthy's thinking about list processing at Dartmouth led to his Lisp language in 1958. Samuel's checkers work became a landmark in machine learning.

## What did the Dartmouth workshop get right and wrong about AI?

The proposal rested on a bold guess: that every feature of learning or intelligence could, in principle, be described precisely enough for a machine to simulate it. Sixty-three years on, that guess has not been proven or disproven. Machines now translate languages, recognize faces and beat champions at Go, all goals that would have pleased the organizers. Yet no system matches the flexible, general intelligence they imagined.

Where they were most wrong was the timescale. The proposal suggested that real progress was possible in one summer with ten carefully chosen people. The problems turned out to be far harder, and the field has gone through cycles of high hopes and funding cuts, often called "AI winters."

What they got right was the list of problems. Language, abstraction, learning, self-improvement and neural networks all appear in the 1955 proposal, and all remain active research topics in early 2019. The biggest recent advances have come from neural networks trained on large amounts of data, an approach the workshop discussed but set aside in favor of symbols. In that sense the field has come back around to one of the ideas on the table that summer.

## Sources

- A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence: https://web.archive.org/web/2019/http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html
- Ray Solomonoff and the Dartmouth Summer Research Project in Artificial Intelligence, 1956: http://raysolomonoff.com/dartmouth/dartray.pdf
- Dartmouth workshop: https://en.wikipedia.org/w/index.php?title=Dartmouth_workshop&oldid=878151960