How did John McCarthy and Marvin Minsky start the MIT Artificial Intelligence Project?

Some of the most important research groups begin with a grand building and a large grant. The MIT Artificial Intelligence Project began with a conversation. Two young assistant professors told their lab director what they needed, he said yes, and that was that. There was not even a written proposal.

The two were John McCarthy and Marvin Minsky, both veterans of the 1956 Dartmouth workshop where the name "artificial intelligence" was first used. At MIT they turned that summer's ideas into a permanent home for AI research.

When it started

The founding date depends on whom you ask. McCarthy, in his own history of the Lisp language, says that he and Minsky started the project in the fall of 1958, the same season he joined MIT as an assistant professor of communication sciences in the Electrical Engineering Department. Minsky held the same junior rank, but in mathematics. Other accounts, including many short biographies of Minsky, say he joined MIT's staff in 1958 and started the project with McCarthy a year later, which is why 1959 often appears as the founding year. The difference probably reflects when the work began versus when the project became an established, named effort, but the sources do not settle it.

How it was set up

The project lived inside MIT's Research Laboratory of Electronics, known as RLE. RLE held a contract from the U.S. armed services that gave its director, Professor Jerome Wiesner, unusual freedom to launch projects he found scientifically interesting. That freedom is why no formal proposal was needed.

McCarthy later recalled the request in a few words. Here is the whole starting kit:

A keypunch deserves a word of explanation. Computers of the late 1950s did not have keyboards attached. Programmers wrote code on paper, typed it on a keypunch that cut holes in cardboard cards, and fed the stack of cards into the computer. Having your own keypunch meant not waiting in line for someone else's.

What the project worked on

The first big product was Lisp. McCarthy had been thinking about a list-processing language for AI since Dartmouth, and the implementation began at MIT in the fall of 1958. The team started by translating Lisp functions into machine instructions by hand. Then a programmer, Steve Russell, realized that a function McCarthy had written for theoretical purposes, called eval, could run Lisp programs directly, and hand-coded it. Lisp suddenly had an interpreter. Daniel Edwards, Timothy Hart, Michael Levin, Klim Maling and others contributed to the system, and Hart and Levin wrote the first Lisp compiler in Lisp itself.

Lisp was a tool for bigger goals. McCarthy wanted it for his Advice Taker, a program he proposed in 1958 that would accept statements about the world and draw common-sense conclusions from them. Early applications written in Lisp at MIT included a program by Nathaniel Rochester and Saul Goldberg for symbolic calculations on electrical circuits, James Slagle's thesis on solving calculus integrals symbolically, supervised by Minsky, and Paul Abrahams' thesis on checking mathematical proofs.

The group also explored a new way of using computers. Because Lisp let you define a function, test it and edit it without leaving the interpreter, it suited time-sharing, in which many users share one computer through terminals instead of waiting for batch runs of punched cards. McCarthy recalled that Lisp was shown running in an early, experimental time-sharing arrangement on the IBM 704 in 1960, or possibly 1961.

Why it mattered

The MIT project gave AI something it had lacked: a stable institution. Dartmouth had been one summer. MIT offered offices, students and continuing funding, which let a community form and pass its skills on. Its graduate students learned the new methods by doing real projects, such as Slagle's integration program.

It also linked AI tightly to programming tools. Lisp, born in the project, became the standard language of AI research for decades. And the group's early interest in interactive, shared computing pointed toward the way most people use computers now, at a screen and keyboard rather than through stacks of cards.

What happened to the MIT AI Project after its first years?

The founders soon went separate ways. McCarthy left for Stanford University in 1962 and continued his AI research there. Minsky stayed at MIT for the rest of his career, until his death in 2016.

In 1963 MIT launched Project MAC under director Robert Fano, funded by a grant of $2 million from ARPA, the Defense Department's Advanced Research Projects Agency. Minsky's AI group became part of it. The group focused on vision, the movement and manipulation of objects by machines, and language, which it saw as keys to more intelligent machines. Its programmers wrote their own time-sharing system, called ITS, for the PDP-6 computer the group shared, and then for its successor, the PDP-10.

In 1970 Minsky's group split off as the MIT Artificial Intelligence Laboratory. Over the following decades it was home to work on Lisp machines and to programmers such as Richard Stallman, whose experiences there led to the free-software movement. On July 1, 2003, the AI Lab merged with MIT's Laboratory for Computer Science to form CSAIL, the Computer Science and Artificial Intelligence Laboratory, the largest laboratory on the MIT campus, housed in the Stata Center. A project that began with one room and a keypunch became one of the biggest research groups at the Institute.