Alan Turing

The mathematician who defined what a computer is before one existed, helped break Enigma, and in 1950 asked whether machines can think. By the end you will be able to explain the Turing machine, the imitation game, and the strongest objections to it.

“I propose to consider the question, ‘Can machines think?’” So begins one of the most cited papers in the history of science, published in the philosophy journal Mind in October 1950. Its author had already, at 24, worked out what it means for a machine to compute anything at all, and had spent the war helping to read Germany’s secret messages. He would be dead within four years, after being prosecuted for his sexuality. Alan Turing (1912–1954) did not build the first AI program, but he set out the questions the field still argues about.

1936: a machine on paper

In 1928 the German mathematician David Hilbert posed a challenge called the Entscheidungsproblem, the “decision problem”: is there a mechanical method that can decide, for any statement of mathematical logic, whether it can be proved? To answer it, someone first had to say exactly what “a mechanical method” means. Turing, a young fellow of King’s College, Cambridge, did so in his paper “On Computable Numbers, with an Application to the Entscheidungsproblem” (1936).

He imagined a human clerk working with pencil and paper, and stripped the job down to its bare essentials. The result is what we now call a Turing machine. Climb the ladder to see how it works:

  1. A tapeAn endless strip of paper divided into squares, each holding one symbol (or blank). This is the machine’s memory.
  2. A headIt looks at one square at a time. It can read the symbol, write a new one, and move one square left or right.
  3. A stateThe machine is always in one of a finite number of “states of mind”, like a clerk remembering which stage of a sum they are at.
  4. A rule tableA fixed list says: “in this state, reading this symbol, write that, move this way, and switch to that state.” Nothing else is allowed. This is an algorithm.
  5. The universal machineTuring showed that one special machine can read a description of any other machine from its tape and imitate it exactly. One machine, endless programs: the idea behind every stored-program computer.
  6. The limitHe also proved that no machine can decide, for every machine, whether it will go on producing output for ever — a result usually restated today as the halting problem. So Hilbert’s question had the answer “no”.

The American logician Alonzo Church reached the same answer about Hilbert’s problem a few months earlier by a different route, and Turing went to Princeton to do his doctorate with him. Turing’s version won out as the intuitive picture of computation, because it was so obviously something a machine could do. Its lesson for AI is profound: if thinking is a kind of step-by-step symbol processing, then a single universal machine could in principle do it.

Bletchley Park and the Bombe

In September 1939 Turing joined the Government Code and Cypher School at Bletchley Park, the British codebreaking centre. Its great target was Enigma, the cipher machine used by the German armed forces, which scrambled each letter through a set of rotating wheels whose settings changed daily.

The breakthrough had come first in Poland. Consensus Polish mathematicians at the Cipher Bureau — Marian Rejewski, Jerzy Różycki and Henryk Zygalski — had been reading Enigma messages since the end of 1932, and in 1938 built an electromechanical device, the bomba, to search for settings. In July 1939, weeks before the German invasion, they shared their methods and replica Enigma machines with British and French codebreakers at a meeting near Warsaw. Their work is too often left out of the Bletchley story.

When the Germans made Enigma harder to break, Turing designed a new and more powerful machine, the British Bombe, which used guessed fragments of plain text (“cribs”) to rule out impossible settings at speed. Gordon Welchman added a key improvement, and the first Bombe began work in March 1940. Turing went on to lead Hut 8, which worked on the especially difficult German naval Enigma. The intelligence produced at Bletchley helped the Allies win the Battle of the Atlantic, though historians caution against precise claims about how many years it shortened the war.

Machines that learn: 1948 and 1950

After the war Turing designed an electronic computer, the ACE, at the National Physical Laboratory, then moved to Manchester, where one of the first stored-program computers was running. In 1948 he wrote a report for the laboratory called “Intelligent Machinery”. It attracted little notice at the time and was not published until 1968, but it is full of ideas that became central decades later: networks of simple, randomly connected units (“unorganised machines”) that could be trained into useful shapes, a kind of reward-and-punishment teaching, and searching through possible solutions.

His 1950 paper, “Computing Machinery and Intelligence”, is the famous one. Turing thought “Can machines think?” was too vague to answer, so he replaced it with a game. An interrogator types questions to two hidden players, a human and a machine, and must decide which is which. If the interrogator is fooled about as often as when both hidden players are human, Turing argued, we would have no good reason to deny the machine intelligence. This is the imitation game, now called the Turing test. He even made a prediction:

“… an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.”Alan Turing, “Computing Machinery and Intelligence”, Mind (1950), forecasting computers of about fifty years later

Much of the paper answers nine objections in turn. Explore some of them:

“Machines can’t be conscious.” (The argument from consciousness)

Turing replied that we cannot directly check that other people are conscious either; we accept it because of how they behave. Demanding more of machines than of people, he argued, leads to solipsism, the view that only you can be sure you have a mind.

“Mathematics proves machines have limits.” (The mathematical objection)

Results such as Gödel’s and his own show that any given machine has questions it cannot answer. Turing agreed — but pointed out that humans make mistakes and have limits too, and nobody has shown that human minds escape them.

“A machine can only do what we tell it.” (Lady Lovelace’s objection)

Ada Lovelace wrote in 1843 that Babbage’s Engine could not “originate” anything. Turing answered that machines surprise us all the time, and that a machine that learns could do things its programmer never specified.

“Machines will never be kind, have a sense of humour, fall in love …” (Arguments from various disabilities)

Turing suggested these claims were mostly generalisations from the limited machines people had seen, rather than arguments.

The paper ends with a practical suggestion that reads like a description of modern AI. Rather than programming an adult mind directly, Turing wrote:

“… why not rather try to produce one which simulates the child’s?”Alan Turing, Mind (1950)

Build a “child machine”, he proposed, and educate it with rewards and punishments. Today’s systems are not raised like children, but the core idea — that intelligence might be learned rather than written down — is exactly the bet that machine learning made.

Prosecution, death and apology

In 1952 Turing reported a burglary at his home in Wilmslow, near Manchester, and in the course of the investigation told the police he had had a sexual relationship with a man. Homosexual acts were then a crime in Britain. He was convicted of “gross indecency” in March 1952 and, to avoid prison, accepted a year’s probation on condition that he undergo hormone (oestrogen) treatment intended to suppress his libido. As a convicted homosexual he lost his security clearance for government codebreaking work.

He kept working — on the mathematics of how patterns form in living things — until his death on 7 June 1954, at 41, from cyanide poisoning. Debated The inquest recorded suicide. Some biographers, notably Jack Copeland, have argued that accidental poisoning from chemistry experiments at home cannot be ruled out; others accept the inquest’s verdict. The evidence does not settle it.

In 2009, after a public petition, Prime Minister Gordon Brown issued a formal apology for the “appalling” way Turing had been treated. In December 2013 he received a posthumous royal pardon, and in 2017 a law popularly called the “Alan Turing law” extended pardons to thousands of other men convicted under the same laws. Since 2021 his portrait has appeared on the Bank of England’s £50 note.

Is the test any good?

For more than seventy years the Turing test has been both AI’s best-known goal and one of its most criticised ideas. The most famous attack came from the philosopher John Searle in 1980. Imagine, he said, a person who speaks no Chinese locked in a room with a rulebook. Chinese questions come in through a slot; by following the rules, the person sends back perfect Chinese answers. To people outside, the room “understands” Chinese — yet nobody inside does. If so, passing the test by manipulating symbols cannot be enough for understanding. This is the Chinese Room argument.

The test also turned out to be gameable. As early as 1966, Joseph Weizenbaum’s simple chatbot ELIZA convinced some users that it understood them. In 2025 a study by Cameron Jones and Benjamin Bergen reported that a large language model (GPT-4.5), prompted to adopt a humanlike persona, was judged to be the human 73% of the time in five-minute three-party tests As of 2026. By Turing’s own five-minute standard, the prediction has arguably come true — and few researchers think the question of machine thinking has been settled.

Debated What, if anything, does passing the imitation game show?

Other views
  • Behaviour is the evidence we have (Turing’s view): we judge other people’s minds by what they do; refusing to do the same for machines is prejudice. Many functionalist philosophers broadly agree.
  • Simulation is not understanding (Searle): symbol manipulation alone, however convincing, produces no meaning or understanding.
  • The “systems reply”: the person in the room does not understand Chinese, but the whole system — person, rules and paper — might. Searle rejected this; the argument continues.
  • Wrong target: many AI researchers now see the test as a measure of how easily people are fooled, and prefer to test specific abilities such as reasoning, reliability and factual accuracy.
Quick check

What did Turing’s universal machine show?

The universal machine reads another machine’s rule table from its tape and runs it — the principle of the stored-program computer.

Who first broke Enigma, years before Bletchley Park?

Rejewski, Różycki and Zygalski read Enigma from late 1932 and shared their methods with Britain and France in July 1939.

What does Searle’s Chinese Room argue?

Searle accepts the room could give perfect answers; his point is that nothing in it understands them.

// Pause and reflect

  1. Turing replaced “Can machines think?” with a test of behaviour. Was that a clever move or a way of dodging the real question?
  2. If a chat assistant can now fool people for five minutes, what would you want a machine to do before calling it intelligent?
  3. How might Turing’s treatment have changed the history of computing and AI, had he lived and worked into the 1960s?

Your notes are saved privately in this browser.

What happened next

Turing was not working alone. In the same decade, neuroscientists, engineers and mathematicians on both sides of the Atlantic were modelling neurons as logic, studying feedback in machines and animals, and measuring information itself. Continue with Neurons, Feedback and Information.

Key terms on this page

Sources & further reading