Computing Machinery and Intelligence
"Computing Machinery and Intelligence" is Alan Turing's 1950 paper in the philosophy journal Mind (volume 59, number 236, October 1950, pages 433–460), the founding text of the philosophy of artificial intelligence. It opens with the sentence "I propose to consider the question, Can machines think?", declines to answer it directly, and substitutes a game, then spends most of its length answering objections and sketching how a machine might actually be brought to play. It is the document, more than any single experiment, that made machine intelligence a respectable subject of argument before the field had a name.
The paper is organized as a staged argument, not a single thesis. Its seven sections run: the imitation game; a critique of the new problem; the machines concerned in the game; digital computers; the universality of digital computers; contrary views on the main question; and learning machines. Sections three to five set up the technical premise that matters: a digital computer with enough storage and speed can imitate any other discrete-state machine, so the question of whether some machine can do something reduces to whether a suitable program exists. Turing was writing for philosophers, so he explained the stored-program computer from scratch, down to the idea of a store, an executive unit and a control.
He also made a dated, hedged prediction, which people routinely misquote. Turing estimated that in about fifty years' time it would be possible to program computers with a storage capacity of about 10^9 to play the imitation game so well that an average interrogator would have no more than a seventy percent chance of the right identification after five minutes of questioning. That is a claim about a modest, time-limited game, stated as a belief, and he separately said he expected the use of words and general educated opinion to change enough by the century's end that one could speak of machines thinking without expecting to be contradicted. It is not a claim that machines would be conscious or that the game defines intelligence.
The nine objections are the paper's most reusable part, because the same arguments return in every later debate. Turing's replies are short and unequal in quality. He treats the theological and "heads in the sand" objections lightly, concedes that the mathematical objection shows limits that apply to machines but notes that human minds have no proven immunity from error or limits. Against the argument from consciousness he observes that taken seriously it leads to solipsism, and that a viva-voce style of questioning gives the interrogator a workable test. Against the "never do X" list he points out that such claims rest on a narrow stock of machines seen so far, and he treats the continuity of the nervous system as irrelevant because a digital machine can approximate a continuous one closely enough.
His reply to Lady Lovelace's objection is the bridge to the final section. The objection, drawn from Ada Lovelace's 1843 notes on Babbage's Analytical Engine, holds that a machine can only do what it is told. Turing reframed it as the claim that machines never take their makers by surprise, and answered from experience that they do so often, because a programmer cannot foresee all the consequences of a program, and surprise requires creative mental effort on the viewer's side, not on the machine's. The point is that predicting what a rule-following system will do is not the same as writing down its rules, which anyone who has debugged a large program knows.
The final section, on learning machines, is the part that reads most as prophecy. Instead of writing a program simulating the adult mind, Turing asked why not write one simulating the child's mind and then subject it to an appropriate course of education. He split the problem into the initial child machine, the education process, and everything else, compared the structure of the child machine to hereditary material and changes to mutations, and imagined teaching by reward and punishment with a human supplying the judgement. He saw the process as slow and experimental, with the teacher unable to inspect the machine's internals fully, which is a fair description of training opaque models today. The modern parallels to training from data, reinforcement learning and fine-tuning by human feedback are obvious, though no one in 1950 had the data or the compute.
Common pitfalls follow from reading only the first section. Quoting the paper as a definition of intelligence misses that Turing treated the game as a replacement for a question he considered too meaningless to discuss. Treating it as a failed forecast because no machine passed a strict test by 2000 overlooks how limited the claim was. And taking the paper as a purely philosophical essay ignores its extended technical groundwork about digital computers and its engineering-minded last section.
Read "Computing Machinery and Intelligence" as a 1950 research program disguised as a philosophy essay: define the question operationally, answer the standing objections, then get on with building machines that learn.