QuestionsCan a machine think?A machine might utter words, but without knowing what it says
Can a machine think?Understanding is missing

A machine might utter words, but without knowing what it says

The two tests for telling a machine from a man

René Descartes 1596 · 1650Doubt everything to find what survives doubt.
See profile
What it claims

A machine with the shape of a man that imitated his actions would not be a real man: it would act without knowledge, and would not understand what it says.

Why

A machine can utter words, but reason works in every situation while organs need a specific arrangement for each case.

In short

"A machine might utter words, but without knowing what it says" is an argument by René Descartes (Discourse on Method, V, 1637) on the question "Can a machine think?" (answer: understanding is missing). Its claim: A machine with the shape of a man that imitated his actions would not be a real man: it would act without knowledge, and would not understand what it says. According to AskPhilosophy's logic check (Z3 solver), the argument is valid only if one grants a premise the author leaves unstated (P3). AskPhilosophy presents 3 objections, 3 of them with a documented reply, and the original text behind each premise, checked word for word.

Discourse on Method, V, 1637
Valid with an implicit premise

Descartes holds that the body of animals, and even the human body, works like a machine. He then asks a question we take up again today with robots and software: if an automaton had the shape of a man and imitated his actions, how could we know it is not a real man? He answers with two tests. The first is about speech: a machine might utter words, but not arrange them in endless ways to answer what is said to it. The second is about reason: it serves in every circumstance, whereas a mechanism is only ready for what it was set up to do. In both cases the machine would act by the arrangement of its organs, not by knowledge.

P1One can conceive of a machine that utters words, even about what is done to its organs (touched, it cries out that it is hurt). But one cannot conceive that it arranges them in different ways to answer the meaning of everything said in its presence, as even the dullest men do.
P2Reason is a universal instrument, which serves on every occasion. The organs of a machine, on the contrary, need a particular arrangement for each particular action. So it is morally impossible for a machine to have enough different organs to act in all the situations of life as our reason makes us act.
P3 · implicit premiseIf a machine fails either of these two tests, then it does not act from knowledge but from the arrangement of its organs: it does not understand what it says or does.
So a machine with the shape of a man that imitated his actions would not be a real man: it would act without knowledge, and would not understand what it says.

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 P3, the computer finds a case where the machine fails both tests and yet nothing says it lacks knowledge: a man shut in a room, unable to answer everything (he does not speak the language, he is exhausted), is not therefore an automaton. Failing alone does not prove that something is missing inside.
The argument is valid once the hidden premise P3 is added. P1 and P2 describe what a machine could not do. P3 takes the decisive step: from that failure it concludes that the machine does not act from knowledge. Descartes almost writes it ("they did not act from knowledge, but solely from the disposition of their organs"), without proving it separately. Whoever rejects the conclusion can dispute P1 or P2 (the machine would succeed), or P3 (success would not be a sign of knowledge, or failure would not prove its absence).
Analogy. A phone answering system says "press 1, press 2": it has sentences ready for foreseen cases. Ask it an unforeseen question and it can only repeat its menu. Descartes says the same of any machine: it is set up for cases, whereas reason adapts to any case, even the most unexpected.

Objections

3

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

La Mettrie 1747 · attacks premise 1

The difference between man and animal is only one of degree. He defies anyone to name an experiment that makes it impossible to teach a well-trained ape to speak, as the deaf-born are taught: given how like ours its structure is, he has very little doubt it would succeed.

Descartes replies
see the argument
Descartes 1637 · Descartes (1637 text, written before the objection)
Very little reason is needed to be able to speak. Even if some animals are easier to train than others, it is not credible that the most perfect ape or parrot of its species should fail to equal in this the most stupid child, or at least a child with a troubled brain, unless their soul were of a nature wholly different from ours.
Where the debate stands (our summary)Open point on animals. La Mettrie bets that a well-trained ape would speak; Descartes holds that even the most gifted does not equal a child, which would show a difference of kind. Twentieth-century attempts to teach apes a language did not settle it: people still dispute whether the learned signs are real sentences (our summary).
Alan Turing 1950 · attacks premise 1

He replaces the question "can a machine think?" with a blind game of questions and answers. This game resembles the oral examination used to find out whether someone really understands or has only learnt "parrot fashion". If a machine answered as well as a candidate, Descartes's first test would be passed, and it would have to be judged on its answers.

a defender replies
see the argument
No reply possible: objection raised in 1950, after the author's death (1650).
E
Emily Bender and Alexander Koller defender of the argument · 2020
On grounds other than Descartes's, they argue that a system trained only on form (strings of words) has, a priori, no way to learn meaning. Answering well therefore does not prove that one understands: this is the idea of premise P3.
Cameron Jones and Benjamin Bergen 2025
Pre-registered three-party Turing tests were run with language models. One model, prompted to adopt a human persona, was judged to be the human 73% of the time, more often than the real human. Descartes's first test (answering aptly what is said) is thus passed in short conversations.
Where the debate stands (our summary)Open point. Today's language models answer what is said to them, sometimes better than hurried humans: the speech test as Descartes describes it seems to be passed in short conversations. The basic disagreement is about what that proves: for some, answering like a human is enough; for others, answering well is not understanding. There is no widely accepted answer (our summary).
Alan Turing 1950 · attacks premise 2

It is not necessary to build a different machine for each task. A single digital computer, programmed differently for each case, can do them all. Descartes supposed a particular organ would be needed for each action; a computer does not need this multiplicity of organs.

a defender replies
see the argument
No reply possible: objection raised in 1950, after the author's death (1650).
Hubert Dreyfus defender of the argument · 1972
On grounds other than Descartes's, he argues that human intelligence in a situation cannot be reduced to explicit rules that a program could follow: knowing what matters in a situation depends on a background that cannot be written out in full. A universal computer can compute anything, but it still needs the right program for each case.
Where the debate stands (our summary)Disputed. Turing shows that a single machine can imitate any discrete-state machine: the "not enough organs" argument falls. What remains is Descartes's deeper question, moved: can a program be written (or learnt) for all the circumstances of life? Views differ (our summary).