QuestionsCan a machine think?The imitation game: replacing the question with a test
Can a machine think?Thinking is computing

The imitation game: replacing the question with a test

The Turing test

Alan Turing 1912 · 1954A mathematician who proposes replacing the question 'Can a machine think?' with an imitation game.
See profile
What it claims

If a machine cannot be told apart from a human by its answers, we have no more reason to deny it thought than to deny it to other humans.

Why

Of other humans we know only their answers; if a machine answers the same way, we have no reason to deny it thought.

In short

"The imitation game: replacing the question with a test" is an argument by Alan Turing ('Computing Machinery and Intelligence', Mind, 1950) on the question "Can a machine think?" (answer: thinking is computing). Its claim: If a machine cannot be told apart from a human by its answers, we have no more reason to deny it thought than to deny it to other humans. According to AskPhilosophy's logic check (Z3 solver), the argument is valid only if one grants a premise the author leaves unstated (P4). AskPhilosophy presents 5 objections, 4 of them with a documented reply, and the original text behind each premise, checked word for word.

'Computing Machinery and Intelligence', Mind, 1950
Valid with an implicit premise

In 1950, Turing finds the question 'Can machines think?' too vague to discuss, since everyone gives the words 'machine' and 'think' their own meaning. He replaces it with a game: an interrogator puts written questions to a human and a machine, without seeing them. If the machine is taken for the human, we have no more reason to deny it thought than to deny it to other humans, whose thinking we judge only from their answers. In the same article he rebuts nine objections.

P1The question 'Can machines think?' depends on the meaning given to the words and cannot be settled by a poll. It must be replaced by a closely related question expressed in sufficiently precise terms.
P2The replacement question is the game's: does an interrogator who cannot see his interlocutors decide wrongly as often when a machine takes the part of one of them as when a man and a woman play?
P3For other humans, we have nothing but their answers: requiring to be the person oneself to know that she thinks is solipsism. We grant by polite convention that everyone thinks.
P4 · implicit premiseIf a machine is indistinguishable from a human by its answers, the evidence we have is the same for it as for the human. Denying thought to one and granting it to the other thus requires an extra reason, and there is none.
So if a machine cannot be told apart from a human by its answers, we have no more reason to deny it thought than to deny it to other humans.

click a premise to reject it

If you accept every premise, logic forces you to accept the conclusion. So the debate is about the premises.
Checked by a logic solver: the premises cannot all be true while the conclusion is false.
Without the implicit premise: Without P4, the computer finds a case where the machine gives the same answers as the human and yet there is a reason to deny it thought (for example, we know it merely recites ready-made answers).
Valid with P4. P1 and P2 turn the question into a behaviour test; P3 says that for humans behaviour is our only criterion. P4 links the two: same evidence, same conclusion. Turing does not write it as a syllogism, but it is what his reply to Jefferson assumes. Without P4, nothing forbids treating the machine differently from the human despite identical answers.
Analogy. You cannot enter your neighbour's head, and yet you are sure he thinks, because he answers sensibly. Turing asks why we would change the rule if a machine answers.

Objections

5

Each objection is a conversation: click to see who replies and where the debate stands.

Geoffrey Jefferson 1949 · attacks premise 3

Until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, we cannot say the machine equals the brain. It would have to know it had written it and to feel it.

Turing replies
Alan Turing 1950 · Turing's reply
Turing replies that this objection leads to solipsism: one would know only that one thinks oneself, never that another does. Instead of arguing endlessly, we grant by convention that everyone thinks. He proposes an oral examination, in which the interrogator checks whether the witness really understands what he says.
Where the debate stands (our summary)Our summary: Turing shows that the consciousness objection, pushed to the end, applies to humans too. Later critics (Searle, below) maintain that answers do not suffice to prove understanding: a core disagreement about what counts as evidence.
Ada Lovelace 1843 · attacks the conclusion

The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform. A machine can therefore never do anything really new, which thinking would require.

Turing replies
see the argument
Alan Turing 1950 · Turing's reply
Turing replies first that Babbage's engine, of which Lovelace spoke, was a universal computer and could in principle mimic any machine. Then, that machines take him by surprise very often, because he does not calculate in advance what they will do.
Where the debate stands (our summary)Our summary: Turing foresees that his critic will say the surprise comes from our mind and not from the machine. The disagreement is about what 'creating' means: being unpredictable to us, or producing what the programmer had not fixed. A disputed point.
Kurt Gödel 1931 · attacks the conclusion

In any sufficiently powerful logical system, statements can be formulated that the system can neither prove nor disprove. A machine, being such a system, therefore has questions it cannot answer, whereas the human mind would have no such limit.

Turing replies
Alan Turing 1950 · Turing's reply
Turing grants that any given machine has limits, but it has only been stated, without proof, that the human intellect has none. We too give wrong answers, and for any machine there may be a cleverer man, but also a cleverer machine.
John Lucas 1961
Lucas argues that for any given machine, a mind can see the truth of that machine's Gödel statement, which the machine itself cannot prove: so the mind is not a machine.
see the argument
Where the debate stands (our summary)Our summary: the limit of machines is established by mathematics; what is disputed is whether the human mind escapes it. An open point, see the Lucas and Penrose argument.
John Searle 1980 · attacks premise 4

A man locked in a room who follows a program to manipulate Chinese signs can give answers indistinguishable from a Chinese speaker's without understanding a word of Chinese. The same answers therefore do not prove the same understanding.

a defender replies
see the argument
No reply possible: objection raised in 1980, after the author's death (1954).
Paul and Patricia Churchland defender of the argument · 1990
The Churchlands answer with a parallel experiment: waving a magnet produces no visible light, but this does not show that light is not electromagnetic motion. The intuition 'this is not meaning' therefore does not prove that symbols can never produce it.
Where the debate stands (our summary)Our summary: Searle denies that manipulating signs suffices for meaning; his critics dispute the force of the room intuition. Core disagreement about what understanding is. Turing died in 1954 and could not reply.
N
Ned Block 1981 · attacks premise 4

One can imagine a machine that stores in advance a sensible reply to every possible conversation of limited length: a 'Blockhead'. It would pass the test with the intelligence of a jukebox. Passing the test is therefore not enough for thinking: how the answers are produced matters.

no reply
No reply possible: objection raised in 1981, after the author's death (1954).
Where the debate stands (our summary)Our summary: Block shows that the same answers can come from very different mechanisms. Defenders of the test say such a machine is physically impossible and that the test concerns realistic machines. An open point: no widely accepted reply.