A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence
Authors: John McCarthy, Marvin Minsky, Nathaniel Rochester, Claude Shannon
Year: 1955 | Unpublished proposal (workshop held 1956)
Source: mccarthy-1955-a-proposal-for-the-dartmouth
What it is
The 13-page proposal that created the field of Artificial Intelligence. McCarthy, Minsky, Shannon, and Rochester proposed a 2-month, 10-man summer workshop at Dartmouth College in 1956. The proposal contained one founding conjecture that defined symbolic AI: every aspect of intelligence can in principle be precisely described such that a machine can simulate it.
The founding conjecture
From the document:
“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.”
This conjecture is the philosophical core of symbolic AI. It says:
- Intelligence is not mysterious or irreducible
- Every cognitive process has a precise, finite description
- Therefore a machine — a symbol-manipulating process — can implement it
The conjecture is empirical, not provable. It remained the working hypothesis of AI for 30 years.
Why “Artificial Intelligence”
McCarthy deliberately chose the term “artificial intelligence” over alternatives:
- “Complex information processing” (Newell & Simon’s preference)
- “Automata theory” (Shannon’s domain)
- “Cybernetics” (Wiener’s term)
He wanted a term that clearly named the goal — machines that think — rather than a description of the mechanism. The name stuck, defining the field.
The seven research areas
The proposal outlined the problems the field should attack:
| Area | What it asks |
|---|---|
| Automatic computers | Can programs simulate any computable process? |
| Language | Can a computer manipulate words to reason? |
| Neuron nets | How can neurons form concepts? |
| Theory of computation size | How do we measure the efficiency of a calculation? |
| Self-improvement | Can a machine improve its own programs? |
| Abstractions | Can a machine form generalizations from examples? |
| Randomness and creativity | Can pseudo-randomness expand the search for solutions? |
The seven areas span what became AI’s core subareas: NLP, machine learning, computational complexity, heuristic search, knowledge representation.
The workshop
The Dartmouth Conference, summer 1956, brought together the people who would build AI:
- McCarthy: Lisp, knowledge representation, the frame problem
- Minsky: perceptrons, frames, society of mind
- Shannon: information theory, already in the collection
- Newell & Simon: Logic Theorist, GPS, Physical Symbol Systems
It was not a breakthrough event in itself — but it named the field and established the community.
What it assumed and what it didn’t
The proposal assumed intelligence = symbol manipulation. It did not assume:
- A specific architecture (neural vs. logical)
- A specific method (search vs. representation vs. learning)
- That the problem was easy (the authors expected it would take decades)
The seven areas show they anticipated both the symbolic approach (language, abstractions, self-improvement) and the neural approach (neuron nets). The field largely pursued the symbolic approach until the connectionist revival of the 1980s.
Historical position
Turing 1936 (computation is possible)
Turing 1950 (machines can think)
↓
Dartmouth 1955 (let's build thinking machines, and call it "AI")
↓
Newell & Simon 1956 (Logic Theorist: first AI program)
McCarthy 1958 (Advice Taker: knowledge representation)
Robinson 1965 (Resolution: mechanical theorem proving)
See also
- turing-1950-computing-machinery-and-intelligence — The “can machines think?” question the proposal answers with “let’s find out”
- newell-1976-computer-science-as-empirical-inquiry — The theoretical foundation of what the workshop program assumed
- mccarthy-1958-programs-with-common-sense — McCarthy’s first concrete system proposal, three years later