The Summer They Invented the Future at Dartmouth

The mathematics building at Dartmouth College was quiet in the summer of 1956. No students rushed through its corridors; the academic year had ended. But on the second floor, in a cluster of offices borrowed for the occasion, a different kind of energy hummed. For eight weeks, a small group of men β€” mathematicians, engineers, a psychologist, and a computer scientist β€” met daily to pursue a conjecture so audacious it sounded like fantasy: that every aspect of learning and intelligence could be so precisely described that a machine could be made to simulate it.

The Proposition

The idea originated with John McCarthy, a young mathematics professor at Dartmouth. He had spent the previous year at IBM, where he encountered Nathaniel Rochester, a senior engineer who shared his fascination with the possibility of machine intelligence. Together with Marvin Minsky, then a junior fellow at Harvard, and Claude Shannon, the father of information theory at Bell Labs, McCarthy drafted a proposal to the Rockefeller Foundation. They asked for $7,500 β€” later increased to $13,500 β€” to fund a "Summer Research Project on Artificial Intelligence." The term "Artificial Intelligence" itself was McCarthy's coinage, chosen partly to distinguish the effort from the existing field of cybernetics and partly, he later admitted, to attract funding.

The proposal named a constellation of problems: teaching machines to use language, form abstractions and concepts, solve problems reserved for humans, and improve themselves. It listed potential attendees: Allen Newell and Herbert Simon from RAND, Arthur Samuel from IBM, Ray Solomonoff, Oliver Selfridge. The Rockefeller Foundation approved the grant in June. The workshop was set to begin mid-June and run through August.

The Gathering

Not everyone who was invited came. Some stayed for only a few days. But the core group β€” McCarthy, Minsky, Rochester, Shannon, Newell, Simon, Samuel, Solomonoff, Selfridge, and a few others β€” settled into a rhythm. They worked in offices, walked the campus green, ate together at the Hanover Inn. The atmosphere was less formal conference than extended brainstorming session. No proceedings were published; no formal papers were required. The goal was to think out loud, to test ideas against each other, to see how far the conjecture could stretch.

Newell and Simon arrived with a program already running on the JOHNNIAC computer at RAND: the Logic Theorist. It could prove theorems from Whitehead and Russell's Principia Mathematica. One of its proofs was more elegant than the authors' original. Samuel brought his checkers program, which learned by playing thousands of games against itself. Minsky talked about neural nets and the problem of representing knowledge. Shannon discussed the mathematical theory of communication and its relevance to machine learning. McCarthy pushed for a language β€” what would become LISP β€” built for symbolic manipulation.

They disagreed on fundamentals. Some favored neural networks, modeled loosely on the brain. Others, like Newell and Simon, believed intelligence resided in the manipulation of symbols according to logical rules. McCarthy sat somewhere between, convinced that the right formalism could capture common sense reasoning. But they shared a conviction: the barrier was not theoretical. It was engineering. Given enough memory, enough speed, the right programs β€” intelligence would emerge.

The Prediction

That conviction hardened into a timeline. In the proposal and in conversations that summer, several participants suggested that significant progress β€” perhaps even a machine with human-level general intelligence β€” could be achieved within a generation. Simon famously predicted in 1957 that a computer would be world chess champion within ten years. (It took forty.) Minsky told Life magazine in 1970 that within three to eight years they would have a machine with the general intelligence of an average human being. The optimism was genuine, grounded in the real breakthroughs they were already witnessing: programs that proved theorems, learned checkers, understood simple English in restricted domains.

They were not cranks. These were some of the sharpest minds of their generation. Shannon had laid the mathematical foundation of the digital age. Newell and Simon would win the Turing Award and, for Simon, the Nobel Prize in Economics. Minsky would build the MIT AI Lab into a powerhouse. McCarthy would create LISP and the concept of time-sharing. The tools were primitive β€” vacuum tubes, punch cards, kilobytes of memory β€” but the theoretical foundations of computation, laid by Turing, Church, and von Neumann, were solid. The only question, they thought, was scale.

The Long Winter

Scale, it turned out, was not the only question. The programs that worked in the "micro-worlds" of the late 1950s and 1960s β€” the Logic Theorist, the General Problem Solver, SHRDLU, which could converse about a simulated blocks world β€” proved brittle when faced with the complexity of the real world. The combinatorial explosion of possibilities overwhelmed the hardware. Common sense, the vast web of implicit knowledge humans navigate effortlessly, resisted formalization. The "frame problem" β€” how a system knows what doesn't change when an action occurs β€” became a philosophical and technical quagmire.

By the mid-1970s, the optimism had curdled. The Lighthill Report in the UK declared AI had failed to deliver on its promises. DARPA, the primary US funder, slashed its grants. The first "AI winter" had arrived. Researchers who had once spoken of machines thinking within a decade retreated to narrower problems. The term "Artificial Intelligence" itself became somewhat toxic; some rebranded their work as "advanced computing" or "knowledge-based systems."

The Legacy

Yet the Dartmouth workshop did not fail. It succeeded in a way its organizers could not have foreseen. It gave the field a name, a community, and a research agenda that persists six decades later. The problems they listed β€” language, vision, reasoning, learning β€” remain the pillars of AI. The split between the connectionist approach (neural networks) and the symbolic approach (logic and rules) that surfaced that summer still shapes the field, though modern deep learning has given the connectionists a dominant position.

The participants scattered. McCarthy went to Stanford and founded the SAIL lab. Minsky stayed at MIT. Newell and Simon built the computer science department at Carnegie Mellon. Samuel's checkers program became a template for reinforcement learning. The Logic Theorist proved that machines could do more than calculate β€” they could reason.

Today, the mathematics building at Dartmouth bears no plaque commemorating that summer. The offices have been renovated. But the conjecture they tested β€” that intelligence is a formal process, reproducible in silicon β€” has moved from hypothesis to infrastructure. Every time a phone recognizes a face, a car brakes for a pedestrian, a translation appears instantly, a protein structure is predicted, the echo of that summer sounds. They were wrong about the timeline. They were right about the possibility.

This is one episode in a much longer story. For the full account of the history of artificial intelligence, read “The AI Revolution for Everyone” by Deborah Sullivan on MixCache.com.

← Back to all posts
Comments (0)

No comments yet. Be the first to say something.

Leave a Comment

Please log in or create an account to leave a comment.