Origins: cybernetics and computation (1943–1956)

The first point on this site is McCulloch and Pitts, "A logical calculus of the ideas immanent in nervous activity", in the December 1943 issue of the Bulletin of Mathematical Biophysics. Starting from the all-or-none character of nervous activity, they treated neural events and their relations by propositional logic. The paper concludes in both directions: the behavior of any net can be described in these terms, and for any logical expression satisfying certain conditions a net can be found that behaves as it describes. Before the phrase "thinking machine" was current, the claim that thinking can be written as computation was already in print.

Wiener's Cybernetics (1948, Technology Press/Wiley) declared: "We have decided to call the entire field of control and communication theory, whether in the machine or in the animal, by the name Cybernetics", and placed feedback, time series, computing machines and the nervous system in one frame. The 1948 printing is lending-only, so this version read MIT Press's open-access 2019 reissue of the 1961 second edition, and confirmed in the notes to Part I that it cites McCulloch–Pitts and Rosenblueth, Wiener and Bigelow 1943.

Five years later Shannon's "A Mathematical Theory of Communication" (July 1948, Bell System Technical Journal) measured information on a logarithmic scale and named the base-2 unit the bit. It is a theory of communication, not of intelligence, but the later chapters of this site cannot be written without it: the perplexity of language models, learning as compression, entropy loss functions all come from here. In a footnote Shannon pointed to Wiener's "forthcoming book Cybernetics", one of the four citation lines confirmed in this chapter.

Shannon immediately set another problem. "Programming a computer for playing chess" was read at the IRE convention on 9 March 1949 and printed in the Philosophical Magazine in March 1950. Chess was chosen as a test of whether a general-purpose computer could handle non-numerical problems; the paper proposes an evaluation function of positions (material, pawn structure, mobility) combined with look-ahead over move variations. The lineage of measuring intelligence by games — Samuel in chapter 2, Deep Blue in chapter 4, AlphaGo in chapter 5 — starts here.

1949 also brought Hebb's The Organization of Behavior (1949, Wiley). On page 62 of a book meant to connect neurophysiology and psychology is one sentence: when cell A repeatedly takes part in firing cell B, a change takes place in one or both cells such that A's efficiency is increased. That sentence is the original text of what later literature calls the Hebb rule; Rosenblatt's perceptron in chapter 2 and Hopfield's network in chapter 3 cite the book. Its bibliography does not list McCulloch–Pitts: the two traditions at first ran separately.

The chapter's last point is neither a network nor a game. Chomsky's "Three models for the description of language" (September 1956, IRE Transactions on Information Theory) states that no finite-state Markov process producing symbols state by state can serve as an English grammar. That it appeared in an information-theory journal, and cites Shannon and Weaver's book, shows the air of the period: language, logic, information and neurons were discussed in one room for a few years. That room was the Dartmouth meeting of summer 1956. Its proposal (McCarthy et al., August 1955) did not pass this version's selection rule and has no point, but the textbooks agree that the name "artificial intelligence" was coined there.

Turing's "Computing Machinery and Intelligence", in the October 1950 issue of Mind, does not try to define the question "Can machines think?" but replaces it: in an "imitation game" in which an interrogator questions a machine and a human by teleprinter, can a suitably programmed digital computer play satisfactorily? The rest of the paper answers nine objections to machine thinking and proposes learning machines taught like a child. Minsky 1961 and McCarthy–Hayes 1969 in chapter 2 and Newell–Simon 1976 in chapter 3 cite it in their reference lists. This version read it in an Internet Archive scan of the Mind pages.

What was and was not checked

Of the seven papers, the original printings of Shannon 1948, Hebb 1949, Turing 1950 and Chomsky 1956 were read; Wiener was read in the 2019 reissue. McCulloch–Pitts was read in the 1990 reprint (Bulletin of Mathematical Biology vol. 52), Shannon 1950 in a retyped copy without page numbers, so no page is given for its quotation.