House · MIT AI Laboratory
Artificial intelligence as a discipline was named here, studied here, and championed here; here, too, it stumbled and rose again, more than once.
The Birth of Project MAC
In July 1963, the U.S. Defense Advanced Research Projects Agency (ARPA, later DARPA) approved a 2.2 million dollar grant to the Massachusetts Institute of Technology (MIT) to launch a research program code-named Project MAC. The acronym admitted several readings—"Mathematics and Computation," "Machine-Aided Cognition," "Multiple Access Computer"—each of which mirrored a different ambition of the project: to fuse mathematics, computation, cognition, and shared computing into a single endeavor. The official director was Robert Fano, but the two who truly shaped the project's trajectory were younger: John McCarthy (1927–2011) and Marvin Minsky (1927–2016).
McCarthy arrived at MIT from Dartmouth in 1958 and brought with him the language he had invented: LISP. Minsky landed the same year. In 1959 the two co-founded MIT's "AI Group," which would become a core branch of Project MAC. Hardware development of the LISP machine began here in 1963.
McCarthy left for Stanford in 1962 to found SAIL, but the legacy he left behind—LISP, time-sharing, and the vision of an "Advice Taker," a machine that could be reasoned with—gave the MIT AI Project an unmistakable agenda from the very start: symbolic reasoning, time-shared computing, general intelligence.
From AI Project to AI Laboratory
In 1970, the AI Group formally split off from Project MAC and became the MIT AI Laboratory, with Minsky as its director. That moment marked the high noon of artificial intelligence as an independent discipline. Out of Minsky's circle of students, collaborators, and visitors came many of the names that would later define the field: Terry Winograd (1946–) of SHRDLU, Joseph Weizenbaum (1923–2008) of ELIZA, Seymour Papert (1928–2016) of the Logo language, Patrick Winston, Gerald Sussman, Hal Abelson, Carl Hewitt, and Richard Stallman.
The lab occupied the ninth floor of MIT's Tech Square—a building later mythologized as "the Mecca of AI." Everyone walking through its doors was working at the very edge of the field. On the roof hung "the world's first arm" (a manipulator designed by Minsky); in the basement sat early LISP machines; the corridors were papered with Jargon File slang and hand-written notes from the hackers who lived there.
The Citadel of Symbolism
Through the 1970s, the MIT AI Lab was the heart of the "Symbolic AI" golden age. It held the conviction that intelligence could be reduced to search and manipulation over symbolic structures—a conviction that, paired with Stanford's SAIL on the other coast, dominated the entire AI agenda for two decades.
SHRDLU (1971) was the most dazzling demonstration of this line. Winograd's doctoral thesis—a program that could understand English commands inside a "blocks world," plan its actions, answer questions, and even discuss why it had done what it did. "Put the blue block on top of the red block." SHRDLU answered, "OK." A whole generation of AI researchers, having watched a recording of that dialogue, became convinced that general AI was just around the corner.
ELIZA (1966) was another myth. With only a few hundred lines of code, Weizenbaum wrote a program that played the role of a Rogerian psychotherapist, conversing through pattern matching and template substitution. To his shock, secretaries treated ELIZA as a real counselor and confided in her late at night. Weizenbaum became one of AI's earliest critics; in 1976 he published Computer Power and Human Reason, asking what technology could do versus what it ought to do.
Logo (1967, led by Papert) carried AI thinking into education: a small turtle followed program commands and traced shapes on the screen, letting children grasp geometry and recursion. The Constructionist philosophy behind Logo went on to shape Scratch and the maker movement decades later.
Frame theory (Minsky, 1974), early sketches of Cyc (late 1970s), the first systematic attempts at commonsense reasoning—every one of these efforts pointed toward the same goal: giving machines a human-like understanding.
By the late 1970s, however, the boundaries of the blocks world began to show. SHRDLU could not step outside its blocks; ELIZA had no idea what it was saying; the scale of commonsense knowledge was bottomless. The Lighthill Report of 1973 ignited the first AI winter in Britain. MIT did not suffer the funding crash as severely, but the halo of "the symbolic golden age" had begun to fade.
The Cradle of Hacker Culture and Free Software
The MIT AI Lab had another identity: it was the birthplace of computing's hacker culture.
Steven Levy's 1984 book Hackers: Heroes of the Computer Revolution opens with the hacker ethic of the MIT AI Lab: information should be free, authority should be questioned, the world should be changed through code. Richard Greenblatt and Bill Gosper—mathematician and master programmer—were canonized as "first-generation hackers."
Richard Stallman joined the AI Lab in 1971. He watched, through the late 1970s and early 1980s, as Symbolics and LMI split off from the lab to commercialize LISP machines, taking most of the core engineers with them and shattering the lab's culture of sharing. Stallman resolved that this should not happen again. In 1983 he announced the GNU Project; in 1984 he resigned from MIT to work on GNU full-time; in 1985 he founded the Free Software Foundation. The entire free software movement—and the open source movement that branched from it—can be traced back to a single programmer on the ninth floor of Tech Square who refused to accept the commercial bargain.
SICP (Structure and Interpretation of Computer Programs)—the textbook by Abelson and Sussman built on the Scheme language and published in 1985—was also born here. It became the canon of MIT's 6.001 course and shaped how three decades of computer science undergraduates learned to think about programming.
The CSAIL Era
In July 2003, the AI Lab merged with the Laboratory for Computer Science (LCS, founded in 1963) to form CSAIL (the Computer Science and Artificial Intelligence Laboratory). The first director was the Australian-born roboticist Rodney Brooks.
The merger itself marked the close of an era: AI was no longer a separate kingdom but a constituent part of computer science. The combined CSAIL was vast—more than 100 faculty, over 900 researchers, spanning networking, algorithms, computational biology, robotics, and HCI alike.
CSAIL's modern AI roster remains formidable: Regina Barzilay (NLP and medical AI; after a breast cancer diagnosis in 2017 she focused on AI for early cancer detection), Josh Tenenbaum (computational cognitive science, modeling human learning with Bayesian methods), Tomaso Poggio (co-director of the McGovern Institute for Brain Research, brain-inspired computation and learning theory), Daniela Rus (CSAIL director since 2012, soft robotics and distributed robotic systems), Dina Katabi (wireless sensing and health monitoring), Antonio Torralba (computer vision), and Max Tegmark (the intersection of AI safety and physics).
Sir Tim Berners-Lee has long worked at CSAIL's DIG group on Web architecture and decentralized data (Solid); though more often associated with the W3C, he too is on CSAIL's roster.
Through the Winters and the Renaissances
The MIT AI Lab lived through both AI winters—and was a key spark in each renaissance.
The 1973 Lighthill Report attacked AI's "combinatorial explosion" problem, and DARPA tightened its purse strings—but the Mansfield Amendment exempted parts of MIT's portfolio, allowing it to keep a sizable AI program. When the LISP-machine market collapsed in 1987 and Symbolics slid into bankruptcy, the MIT AI Lab was hit directly, but its basic research did not stop.
By 2012, when the deep-learning revolution arrived, the dominant paradigm at the MIT AI Lab had already shifted from symbolic AI toward statistical learning, Bayesian methods, and deep learning. The work of Tenenbaum, Poggio, Torralba, and others rode the new wave; after 2016, CSAIL stood up several centers focused on deep learning and reinforcement learning.
Yet MIT did not chase the dream of "growing its own OpenAI." Its self-positioning has always been as a wellspring of academic research. The most recent landmark breakthroughs—Transformer, RLHF, Chain-of-Thought, AlphaFold—were not MIT's work, but nearly every author of those works had read SICP, used MIT textbooks, cited MIT papers, or carried some MIT academic lineage.
The Lab as an Institution
Why does the MIT AI Lab matter?
It is a namer. When McCarthy drafted the Dartmouth proposal in 1955, he coined "Artificial Intelligence." In 1959 he turned that phrase into a laboratory at MIT. In 1970 the lab became independent, and "artificial intelligence" finally had its own institutional body.
It is a methodological womb. Almost every core idea in symbolic AI—knowledge representation, search, planning, learning, natural language understanding—was first attempted here. Every later generation of AI researchers had to converse with these MIT works, whether to inherit them or to refute them.
It is a conveyor belt of talent. From Winston to Stallman, from Brooks to Rus, from Tenenbaum's students to Barzilay's PhD students, the MIT AI Lab has, over six decades, sent out generations of engineers, professors, and founders who reshaped the world.
And it is also a cautionary tale. When Minsky and Papert in 1969 quietly pronounced neural networks a dead end through their book Perceptrons, they single-handedly froze connectionism for twenty years. That misjudgment came from MIT—a historical burden for the institution and a lesson for those who came after.
By 2026, the MIT AI Lab as an independent body no longer exists, but CSAIL continues to operate inside Cambridge's Stata Center. Every young researcher who walks through that building still feels something of the obsession passed down from Tech Square's ninth floor: to make machines think, and to make that effort worth doing.
Historian's Note
The MIT AI Lab is to artificial intelligence what Mount Qi was to the Zhou or Feng-Pei to the Han. It was not the largest lab, nor the wealthiest, but it was the first—the first lab to bear the name, the first to study it systematically, the first to inscribe it in textbooks. That "first" brought immense glory and an equally heavy burden: the symbolic golden age bloomed here and here too grew overconfident; the critique of perceptrons sounded here and here, too, froze a road that might have flowered far earlier. By the deep-learning era, MIT was no longer the principal wellspring of new breakthroughs, but it remained the source. Every new generation of AI researcher still makes a pilgrimage here—not for the most cutting-edge algorithm, but to see how the original questions were first asked. Questions outlast answers; this is the greatest legacy MIT AI has left for our day.
Eyewitness Accounts
Call for contributions
If you have worked at or studied in the MIT AI Lab or CSAIL, please contribute on GitHub.
References
- McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (1955). "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence."
- Winograd, T. (1972). "Understanding Natural Language." Cognitive Psychology, 3(1), 1–191.
- Weizenbaum, J. (1966). "ELIZA—A Computer Program for the Study of Natural Language Communication." Communications of the ACM, 9(1).
- Minsky, M. L., & Papert, S. A. (1969). Perceptrons. MIT Press.
- Levy, S. (1984). Hackers: Heroes of the Computer Revolution. Anchor Press/Doubleday.
- Abelson, H., Sussman, G. J., & Sussman, J. (1985). Structure and Interpretation of Computer Programs. MIT Press.
- Stallman, R. (2002). Free Software, Free Society: Selected Essays of Richard M. Stallman. FSF.
- Crevier, D. (1993). AI: The Tumultuous History of the Search for Artificial Intelligence. Basic Books.
- MIT CSAIL Annual Reports (2003–2025).
- Rus, D. (2015). "The Future of Robotics." Scientific American.