Symposium on the Mechanization of Thought Processes is held at the National Physical Laboratory

Background

Around 1958 the electronic computer was still a scarce, room-sized instrument, and what it could be made to do was an open question. A group of researchers had begun to ask whether machines could be built to carry out the tasks people find easy and cannot easily describe: recognising patterns, using language, drawing on ordinary experience to decide what to do next. The vocabulary for this work was not yet settled. The meeting itself was called a symposium on the mechanization of thought processes, while the papers inside it spoke of artificial intelligence, heuristic programming and automatic programming.

The meeting

The symposium was held at the National Physical Laboratory in Teddington, Surrey, on 24, 25, 26 and 27 November 1958 ^[1]. A later paper in the IEEE Annals of the History of Computing describes it as a four-day conference held on the outskirts of London, attended by over 200 people from around the world, with a two-volume proceedings published in 1959 ^[2]. The printed proceedings did appear in 1959, and the Science Museum Group holds volume one of the set ^[1].

A date that does not quite agree

One small discrepancy survives in the record. John McCarthy himself records that his paper was given at the Teddington conference on the mechanization of thought processes in December 1958, a month later than the dates in the printed proceedings ^[3]. The printed volumes and the author's own note therefore place the same talk a month apart, a reminder that conference records are compiled after the fact and can drift from what happened.

The programme

The proceedings were divided into sections, among them General Principles, Automatic Programming and Speech Recognition ^[1]. The headings alone show how the field was then imagined: questions of principle, questions of learning from data, and questions of tools.

The General Principles section carried Marvin Minsky's paper *Some methods of artificial intelligence and heuristic programming* and John McCarthy's *Programs with common sense* ^[1]. Placing the two side by side put a survey of heuristic methods next to a paper whose title announces the problem of giving a machine everyday knowledge.

The Speech Recognition section carried Oliver Selfridge's *Pandemonium: A paradigm for learning* and Frank Rosenblatt's *Two theorems of statistical separability in the Perceptron* ^[1]. Selfridge's title proposes a paradigm for learning, while Rosenblatt's points at the perceptron, a machine for classification, and at the statistical conditions under which its answers can be separated.

The Automatic Programming section included John Backus on the properties and performance of FORTRAN systems I and II, and Grace Hopper on the present status and future trends of automatic programming ^[1]. Here the subject was not how a machine might think but how a person might be spared the labour of writing machine instructions directly.

What the meeting left behind

Read as a list, the programme gathers strands that were later pursued separately. Symbolic reasoning and common sense sit in one section; learning and pattern recognition in another; languages and programming tools in a third. Minsky, McCarthy, Selfridge, Rosenblatt, Backus and Hopper all appear within the same set of volumes, which is a fair picture of how wide the interests of the period were and how little separated the people working on them.

The proceedings also fixed the moment in print. Because the papers were collected into two volumes published in 1959, and because copies entered institutional collections such as that of the Science Museum Group, later readers could return to what was actually said rather than rely on recollection ^[1][2]. For anyone asking how people in 1958 talked about machines and thought, the volumes are a primary record.