Philosophy of AI · the Turing question

Can machines think?

Alan Turing thought the question was too vague to answer, so he replaced it. Instead of asking whether machines think, he proposed a test: if a person conversing with a machine cannot reliably tell it from a human, on what grounds would we deny it thinks? The move was brilliant, and the argument about whether it works has run ever since.

Turing proposed replacing the question 'can machines think?' with a behavioural test: if a machine's conversational responses are indistinguishable from a human's, we have no clear grounds for denying it thinks. Critics argue that behaviour alone cannot establish understanding or consciousness, and that a system might imitate thinking perfectly without doing it.

The positions · at a glance
The debate at a glance
Position 01

Replace the question

Turing's move

'Can machines think?' is too ill-defined to answer. Substitute a test we can actually run: can it converse indistinguishably from a person?

Position 02

Behaviour is the evidence

The case for

This is the only evidence we ever have for other minds. We grant thinking to other people on exactly these grounds, so refusing it to machines needs a reason.

Position 03

Imitation isn't thinking

The case against

A system could produce all the right outputs with no understanding whatever — which is precisely what the Chinese Room argument is built to show.

Position 04

It depends what we mean

The dissolving view

Perhaps there's no fact of the matter beyond our concepts. 'Thinking' may need revision rather than discovery, as 'flying' did when we built machines that fly differently from birds.

Turing opened his 1950 paper by proposing to consider whether machines can think, and then almost immediately abandoned the question as too ill-defined to be worth pursuing. Deciding it would require agreeing on what "machine" and "think" mean, and he judged that agreement unreachable. So he replaced it with something answerable: an imitation game in which an interrogator converses with a human and a machine and tries to identify which is which.

The philosophical force of this move is easy to miss. Turing was not claiming that passing the test proves an inner life. He was pressing a question back at the sceptic: we attribute thought to other people on the basis of how they behave, since we have no access to their inner experience either. If a machine matches that behaviour and we still deny it thinks, what exactly is the extra requirement — and how would we ever check whether other humans meet it?

We grant other people minds on behavioural evidence alone.The awkward parallel Turing pressed

He also anticipated most of the objections still made today. To the theological objection he replied that it assumes what it needs to prove. To the claim that machines can only do what we tell them, he noted that learning machines can surprise their designers, which they routinely now do. To the "machines lack consciousness" objection he gave the sharpest reply: taken seriously, that argument leads to solipsism, since the same doubt applies to everyone else.

The strongest objections are the later ones. Searle's Chinese Room argues that symbol manipulation, however sophisticated, never constitutes understanding — that syntax does not yield semantics. Others point out that the test measures the ability to imitate a human specifically, which is neither necessary nor sufficient for intelligence: a system could be vastly capable and fail by being obviously non-human, or pass by exploiting conversational tricks that have nothing to do with thought. And modern language systems have made the practical version of this concrete, producing fluent text that many people find convincing while the question of what is happening inside them remains entirely unsettled.

Perhaps the most interesting position is that the question dissolves. When we built machines that move through air, we did not discover whether they "really fly" — we extended the concept, while retaining that they fly differently from birds. Thinking may go the same way: not a discovery about hidden machine essences, but a decision about how to extend a concept built for one kind of case to another. That would not make the question meaningless. It would make it a question about us.

People also askQuick answers

What is the Turing test?

A test proposed by Alan Turing in 1950: if an interrogator conversing with a machine and a human cannot reliably tell which is which, Turing argued we have no clear grounds for denying the machine thinks.

Does passing the Turing test prove a machine is conscious?

No, and Turing didn't claim it did. The test is behavioural. Critics argue a system could produce perfect responses with no understanding or experience at all — the point of Searle's Chinese Room argument.

What were Turing's replies to objections?

He argued the theological objection assumes its conclusion, that learning machines can surprise their designers, and — most sharply — that demanding proof of consciousness leads to solipsism, since we can't verify it in other people either.

Can machines think today?

It depends what thinking means. Modern systems produce fluent, contextually apt language, which is much of what the test asked for. Whether that constitutes understanding or is sophisticated pattern-matching remains genuinely disputed.

Sources & further reading

A summary following standard scholarly accounts. Whether behavioural tests can establish thinking remains contested.