Thinkorithm

From Dartmouth to the AI Winters

The field got its name in a 1955 proposal and met its first serious setbacks less than twenty years later. A short account of early optimism and what it taught.

Published · 3 min read

Artificial intelligence became a named research field around a summer workshop in 1956. The decades that followed brought real progress, bold expectations and periods when confidence in the field fell sharply.

A summer project

On 31 August 1955 John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon signed a proposal for a two-month study at Dartmouth College the following summer. It stated that the study would 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”. John McCarthy is generally credited with coining the term “artificial intelligence”, which appears in the proposal’s title.

The workshop itself, held in 1956, is less important for any single result than for bringing together people who would lead the field for the next generation.

Early programs

One of the first programs to attract attention was the Logic Theorist, developed by Allen Newell, Herbert Simon and Cliff Shaw at the RAND Corporation. It searched for proofs of theorems in propositional logic taken from Whitehead and Russell’s Principia Mathematica, using rules of thumb rather than trying every possibility.

A different line of work took inspiration from the brain. In 1958 Frank Rosenblatt published his description of the perceptron, a model that learns to classify patterns by adjusting numerical weights on its inputs.

Limits become visible

Programs that worked on small, tidy problems did not carry over easily to the size and messiness of real language, vision and everyday knowledge. In 1969 Minsky and Seymour Papert published Perceptrons, a mathematical analysis that proved what certain simple perceptron-style machines could not compute. The book challenged the assumption that such devices would grow naturally into general learning machines, and research interest moved away from them for years.

In the United Kingdom, the Science Research Council asked the mathematician Sir James Lighthill for an independent assessment of the field. His report, published in 1973, supported work that modelled psychology and neurophysiology but was sharply critical of the broader ambitions of AI research. The report had been commissioned to help the Council decide on requests for research support.

Expert systems

Interest revived with expert systems: programs that encode the knowledge of specialists as large collections of if–then rules. A well-documented example is R1, later called XCON, which John McDermott developed to configure VAX computer systems for Digital Equipment Corporation, where it was used in regular production.

Expert systems were useful within narrow limits, but every rule had to be written and checked by hand, and the systems failed abruptly on cases their authors had not foreseen. By the end of the 1980s enthusiasm had cooled again, and the period that followed is often described as a second AI winter.

Lessons that lasted

The early decades taught the field several durable lessons. Common sense turned out to be far harder to capture than formal reasoning. Hand-written rules did not scale. And expectations, once raised, were costly to disappoint.

When AI recovered, it did so largely through methods that learn from data rather than from rules written by hand — the subject of How Deep Learning Came Back.

Sources

  1. McCarthy, J., Minsky, M. L., Rochester, N., Shannon, C. E. A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, 31 August 1955. Archived by Stanford University.
  2. Carnegie Mellon University. History — The Simon Initiative.
  3. Rosenblatt, F. “The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain”. Psychological Review, 65(6), 386–408, 1958. doi:10.1037/h0042519
  4. Minsky, M., Papert, S. A. Perceptrons: An Introduction to Computational Geometry. MIT Press, 1969.
  5. Lighthill, J. Artificial Intelligence: A General Survey. In Artificial Intelligence: a paper symposium, Science Research Council, 1973. Text hosted by the University of Edinburgh.
  6. McDermott, J. R1: The Formative Years. AI Magazine, 1981.