Updated 2026-10-11
from the news desk
Does passing the Turing test mean AI is conscious? Why it is not proof of experience
The Stanford Encyclopedia of Philosophy presents the Turing test as a behavioural benchmark about conversation and intelligence. Andrzej Porębski and Jakub Figura argue that even a pass shows imitation, not subjective experience.
No. The Stanford Encyclopedia of Philosophy says the Turing test is a behavioural test about whether a machine can sustain the kind of conversation that could fool an interrogator, while Andrzej Porębski and Jakub Figura argue that passing it shows successful imitation, not subjective experience. One reading of the test is that it can offer evidence for intelligence or mind-like performance, but the sources do not support treating it as decisive proof that an AI is conscious.
What the Turing test actually measures
The Stanford Encyclopedia of Philosophy entry says Turing proposed the test as a way of dealing with the question whether machines can think by replacing that broad question with a more precise game about conversation. In the setup the entry describes, an interrogator exchanges text with a human and a machine and tries to tell which is which.
The same entry says Turing predicted that a machine might one day do well enough that an average interrogator would have no more than a 70 percent chance of identifying it correctly after five minutes. The SEP also stresses a distinction that matters here: later philosophers have treated the test in different ways, including as a possible criterion, as probabilistic support, or, more controversially, as a logically sufficient test for intelligence.
One reading of that framing is that it already narrows what a pass can show. The SEP discusses the test mainly in terms of thought, intelligence and mind attribution, not as a direct meter of felt experience. In its section on the argument from consciousness, the entry notes that one objection was precisely that behaviour might fall short of proving that a machine feels anything at all.
Why passing it can still count as evidence
The strongest case for taking a Turing-test pass seriously also appears in the SEP entry. It says that if something can carry out a conversation with us under demanding conditions, we have good grounds to suppose that it has intelligence of a kind we possess. The entry adds that, if our attributions of mind to other people rest on behaviour, then a machine that performs equally well in the imitation game may deserve similar prima facie treatment. Prima facie here means initial evidence, not final proof.
Porębski and Figura describe why this line of thought appeals to many people in the case of large language models. Their paper says LLMs produce plausible, grammatically strong language, and it reports public claims that current systems may be conscious. Porębski and Figura note Ilya Sutskever’s remark: “It may be that today’s large neural networks are slightly conscious.”
One reading of these positions is that human-like conversation matters because conversation is one of the main outward signs through which humans judge minds at all. On that reading, dismissing a system that repeatedly survives informed questioning could look too quick. The SEP entry supports that limited point when it says the test may provide probabilistic support for attributing intelligence.
Why passing it is not proof of consciousness
The same sources draw a line between strong evidence of conversation and proof of experience. Porębski and Figura state the point directly when discussing the Turing test: a machine’s success shows that it performs the task of imitating a human being well, not that it has inner experience. Their paper says passing the test indicates successful pretence, and it compares that point to an actor who performs Hamlet without thereby becoming Hamlet.
The SEP entry supplies one philosophical reason this gap does not close automatically. It says many philosophers reject the idea that any purely behavioural test can be a logically sufficient condition for intelligence, let alone consciousness. Its summary of Ned Block’s Blockhead thought experiment says a system could, in principle, produce human-like responses by brute-force methods and still seem unintelligent. The SEP also notes that whether Blockhead is ultimately persuasive is debated, but the objection shows why behaviour alone may leave the inner question open.
Porębski and Figura make a similar point about LLM self-reports. They argue: “The recognition of the consciousness of LLMs on the basis of their assertions is flawed because the language usage of LLMs is strictly probabilistic.” In their account, a chatbot’s claim to be conscious is not neutral evidence of inner life, because the system was built to generate likely wording in context. The same paper says ordinary language patterns help explain why models may talk about consciousness more readily than they talk about non-consciousness.
Porębski and Figura also press a consistency objection: “If LLMs are conscious, why not autonomous vacuum cleaners?” Their point is not that vacuum cleaners and chatbots are the same in every practical respect. Their point is that anyone who moves from conversational fluency to consciousness needs a principled reason for stopping with one class of system and not extending the claim more widely.
What follows for AI consciousness debates
A careful reading of the two documents points to a narrower conclusion than either blanket skepticism or blanket acceptance. The SEP entry leaves room for a Turing-test pass to count as serious evidence about intelligence, especially if the performance is robust and repeated under informed questioning. But the same entry also records long-running objections to treating outward behaviour as conclusive.
Porębski and Figura go further than the SEP entry and reject conscious AI altogether for present and foreseeable systems. Their paper states its thesis bluntly: “The thesis of this conceptual study is simple: There is no such thing as conscious AI.” Even readers who do not accept that full conclusion can still accept the narrower point relevant here: nothing in either source says that passing the Turing test settles the question of subjective experience.
That is why the Turing test remains a benchmark in these debates without becoming a proof of consciousness. The SEP says it can tell readers something about whether a system can sustain human-like conversation under pressure. It does not, on the reasoning the SEP surveys and the objections Porębski and Figura develop, let readers skip the harder question of whether there is anything it is like to be that system. For readers asking whether a chatbot is sentient, this is also why sentience, tests for consciousness, and what would count as evidence stay separate questions from a famous conversation benchmark.
Frequently asked questions
Did Alan Turing design the test to prove consciousness?
The Stanford Encyclopedia of Philosophy says Turing proposed the imitation game as a way of dealing with whether machines can think, not as a direct test of felt experience. The same entry discusses a separate argument from consciousness, which shows that philosophers did not treat the two questions as identical.
If an AI says it is conscious, should that settle the issue?
No source here says that it should. Porębski and Figura argue that LLMs generate probable wording, and they say self-ascriptions of consciousness are flawed evidence for that reason.
Can passing the Turing test still matter if it is not proof of consciousness?
Yes. The Stanford Encyclopedia entry says a strong performance can provide prima facie, or initial, reason to attribute intelligence, and it also says the test may offer probabilistic support for mind attribution. That is weaker than proof of consciousness.
Does failing the Turing test prove there is no mind?
The Stanford Encyclopedia entry says some critics call the test too narrow or too chauvinistic, because an intelligent entity might not converse with humans in the required way. In the SEP’s account, the test is disputed both as too hard and as too easy, which is one reason philosophers do not treat it as the last word.
Caveats: Several contemporary examples and the direct claim that passing the Turing test shows imitation rather than consciousness come from Porębski and Figura’s 2025 paper. The broader historical and philosophical account of the test comes from the Stanford Encyclopedia of Philosophy entry.