Central to Searle's Chinese room argument is his claim that the room has the semantic properties of an electronic digital computer. In a post a few days ago in a discussion about the semantics, I argued that there's a fundamental semantic difference. But does this matter to the conclusion of the Chinese room argument?

The difference is about extrinsic semantics. Chinese symbols fall into the room through a slot in the door. Unknown to Searle, the CPU, their shapes have been given meanings, or interpretations (a “semantics”) by observers outside the room. Searle, who knows no Chinese, gets only the shape. Nothing in the shape indicates its meaning. There is no intrinsic semantics, but there is an extrinsic, or external, semantics.

The Chinese room is supposed to be a computer trying to be a mind. But what happens in the actual electronic device? The input to a computer is a sequence of clocked voltage levels, not Chinese ideograms. Clocked voltage levels enter the computer from outside. Unknown to Searle, the CPU, their voltage levels have been assigned meanings by observers outside the room.

That's the difference. This is impossible. An observer can perceive a shape and give it a meaning, but no one can perceive a clocked voltage level and give it a meaning because no human has the sensory apparatus to perceive a clocked voltage level. In other words, the things the Chinese room processes have and extrinsic semantics, but the things computers process do not. It's a fundamental semantic difference between the Chinese room and computers. But does this make any difference to the conclusion of the Chinese room argument?

The things computers process have no semantics at all (neither intrinsic nor extrinsic), so what about brains? They process mainly unary electrical pulses. Humans lack the sensory apparatus to perceive these, too. And what about pre-human times when no one was around to try to assign an extrinsic semantics to neural pulses? Neural pulses have no extrinsic or intrinsic semantics either. But brains understand the world. The biological brain proves that a system that processes semantically vacant objects can come to understand the world. So the semantics of the objects being processed is irrelevant to a system understanding the world? Is the CRA simply irrelevant to AI's quest for the thinking machine?

  • 3
    Central to Searle's Chinese room argument is his claim that the room has the semantic properties of an electronic digital computer. -- Stopped reading here. Computers are SYNTACTIC. They have no semantic properties. Semantics are provided by the people outside the computer. The room has no semantic properties, that's the entire freaking point of the argument.
    – user4894
    Dec 19, 2017 at 21:34
  • 5
    You seem to be confusing representations with what they represent. Voltages in computers represent digits, which have external semantics, just as scribbles on paper in the Chinese room represent Chinese characters, there is no difference of principle. And human sensory apparatus can be easily enhanced by a voltmeter to perceive voltage levels, so both parts of the argument do not work.
    – Conifold
    Dec 19, 2017 at 23:04
  • 1
    An observer gave them a meaning. Searle says computers also process objects that have an extrinsic semantics, but this is false. -- No, it's true. A computer flips bits according to pre-written rules. Humans interpret the bit flips as e-commerce, cat videos, electronic money, online discussion forums. All the computer is doing is flipping bits.
    – user4894
    Dec 20, 2017 at 21:41
  • 1
    0-1 patterns in magnetic storage also form distinct "shapes", and I do not see how the fact we can see cogwheels with a naked eye makes mechanical arithmometers different in principle from electronic computers. Would this mean that blind people have "meanings" different from the rest of us? What about people with bad eyesight, are glasses allowed? Microscopes, telescopes? Electronic microscopes, radiotelescopes? Printers convert voltages into scribbles, are they meaning-giving machines? I just do not see a reasonable way to draw the distinction you are trying to draw.
    – Conifold
    Dec 20, 2017 at 22:14
  • 1
    For the users to be notified of your comments the proper format is @username at the start of a comment. Unfortunately, rewriting your idea in steps does not make it any more plausible. Eyesight involves complex processing of electromagnetic waves inside the body, adding to that external devices makes no essential difference. So either nothing can be "perceived" at all or your distinction does not work. You need to rethink your background notions, especially meaning and its relation to perception, currently they are obscure and you are unhelpfully repeating yourself when asked to clarify.
    – Conifold
    Dec 21, 2017 at 0:58

3 Answers 3


The following statement and the rest of what you've stated considers a computer to only be electronic. But a computer can be a tape with symbols printed on it and a mechanism that reads each symbol, which together with a lookup table, causes the computer to move the tape and possibly read or write another symbol. Both your electronic computer, with electronic signalling, and the tape-machine (Turing machine) are equivalent. The latter is "directly" observable.

I can't answer the rest of your question, but your premise is wrong if I understand it correctly.

That's the difference. This is impossible. An observer can perceive a shape and give it a meaning, but no one can perceive a clocked voltage level and give it a meaning because no human has the sensory apparatus to perceive a clocked voltage level. In other words, the things the Chinese room processes have and extrinsic semantics, but the things computers process do not.

  • If I understand your answer, you say that since electronic computers are equivalent to Turing machines, anything true of a Turing machine is also true of an electronic computer. Can I reply that this equivalence as I understand it is a computational - not semantic - concept of equivalence, but my arguments are about semantic properties. Also, it's fascinating that neither of Turing's 1936 Turing machines, once running, allow input. But say input arrives and a symbol magically appears in a square. What happens then? Computers can handle this, but can Turing machines?
    – Roddus
    Dec 21, 2017 at 0:55
  • @Roddus, As you might know, a TM that disallows inputs after starting is equivalent to a formal system. This was important for Hilbert's program, in vogue at that time, and in particular, it was needed for solving the Entscheidungsproblem. Adding an input to a TM, changes its formal system (it's tantamount to adding a new axiom to the FS). This makes typical computers into input-driven sequences of formal systems, which is why the Godelian attacks on computationalism miss the mark.
    – Phil_132
    Dec 21, 2017 at 8:14
  • @Phil_132. Thanks. I was looking at TMs from the angle of input as perception. In his 1950 paper on machine intelligence Turing doesn't explain perception (he does - intriguingly - promote ESP). Why? Is there some fundamental reason why computation can't explain perception? (I thought the answer was Yes.) Say a symbol magically appears in a square. What if the Scan operation can't identify it? That's the sort of thing I was looking at.
    – Roddus
    Dec 22, 2017 at 2:11

So the semantics of the objects being processed is irrelevant to a system understanding the world?

When I read this, I understood it. How?

For me, the objects being processed were photons from my computer screen impinging on cells in my retinas. Those photons contained zero "semantics".

However, their physical properties and configuration were far from accidental. Primarily, the semantics in your head determined them (based on a human & technical foundation).

That physical signal was sufficient to cause in my head (1) a symbolic representation followed by (2) a syntactic & semantic interpretation (all based on a similar foundation), which hopefully is reasonably close to the one that your head intended (although certainly not identical).

Semantics does not lie in transmitted tokens (signals). It lies in the (more or less shared) structure, function, and content (memory) of the machines doing the formulating and interpreting.

Is the CRA simply irrelevant to AI's quest for the thinking machine?

Largely, yes. It is scientifically vacuous. It is primarily a social phenomenon (duh!), unfortunately one that generates a lot more heat than light. However, true to its original motivation as a critique of 1970s symbolic-AI approaches to language processing, it might have had some value in focusing attention on the need for a better understanding of semantics and its relationship to computation.

EDIT (elaboration, based on comments)

@Roddus, Your comment...

@Conifold. It's pretty hard explaining this. By "meaning" (in the present context) I mean a neural structure inside a brain. By "perception" I mean the mental process of activating an itnernal representation (also a neural structure) of an external object or object quality (eg of a certain shape). The biology of seeing doesn't matter. I'll set out premisses and conlusion as a formal argument.

...is pretty much my answer: Meaning is internally constructed from (or formulated into) external physical constructs or events, which are themselves meaningless but decidedly non-arbitrary.

However, what counts as "meaningful stimuli" is not objectively determinable. Rather, it's in the eye of the beholder. If we agree that removing one's glasses means buy and scratching one's chin means sell, then those are meaningful signals to us, but no one else.

Thus, whether patterns of ink on paper, or photons from a display, or transistor voltages constitute "legitimate semantic representations" is neither true nor false. It depends on the potential interpreters, their capabilities, and their contexts.

  • You say "Semantics does not lie in transmitted tokens (signals) [but] structure, function, and content (memory) of the machines...". This seems right. A semantics is a structure including content plus algorithms that race round inside it. Interestingly, Searle calls baskets of Chinese characters "databases", and symbols entering the room, "bunches" (1990 Scientific American and elsewhere). But databases are symbols plus relations; the input is a stream of temporal relations. Searle never talks of relations. Incredibly, he never addresses structure. (Maybe structure = semantics.)
    – Roddus
    Dec 22, 2017 at 1:57

Externalist views of semantics mean that meaning is what it is independent of our internal perceptions of things. We can't 'shape' reality with our language, rather our language must conform to the evidence of what reality is. I think it's plausible to say that our language creates a model of reality. Whether it perfectly captures its externalist and intrinsic nature is not knowable. Whether our senses fail us and we can't access the ultimate nature of reality is not knowable either. We should use the most rational arguements to determine whether this is the case and use the best evidence and tools at hand to justify these beliefs about reality.

If a computer was programmed to simulate a theory of semantics just like humans, what's the difference? If it was programmed to give the right response to these questions and answer the way you'd expect a human to do so, then so what then? The Chinese Room is supposed to show how strong AI is impossible. Strong AI being human-like AI, thinking or responding like a real human and actually 'know' it is (no matter what the computer can talk about: philosophy, science, art, etc.).

Since the participant (who doesn't understand Chinese) in the Chinese Room doesn't actually understand Chinese can we say an AI actually 'understands' what it is saying? Or does the room 'know' Chinese akin to a complete computer system 'knowing' the meaning of what it is saying. As in, thinking about the room itself is just as important in determining who or what understands Chinese as it is to consider the participant who doesn't understand Chinese. Somebody had to have made those cards with all the Chinese characters and responses on them in the first place. So the creator of the room is just as much a part of the room as the person who doesn't understand Chinese. However, that's why it is argued that the person who is conversing in Chinese but doesn't understand a word of it doesn't really understand Chinese at all and therefore a computer could never really think like a human either and 'know' it.

An AI, a strong AI, that could show it really 'knows' what it is saying isn't important to passing a Turing Test. As long as it appears to be doing so and is indistinguishable from a human counterpart, then this doesn't really matter. However, what this gets at is thinking about whether a computer can become conscious. As we know this can only occur in living organisms. Consciousness (if it exists) is an emergent property of the laws of physics, chemistry, evolution, and biology. Whether it can be simulated atom for atom, neuron by neuron on a computer is another question.

I think we are many years from finding this out if we ever do. But I don't believe just software can think like a human even if it does an extremely convincing job that it can think like us. A human has to give a computer its initial axioms, rules, constraints, and programming. Even though a computer could rewrite these it is still limited by this. It can't 'recreate' itself in the way we can. Imagining possibilities even if the universe itself follows deterministic laws. Intentionality within an AI system is circumscribed by the original architecture that the software is operating. An AI could be 'creative' and evaluate aesthetics given prior data sets, and it could ask questions based on some type of algorithm but it would never do these things unprompted or unsolicited in the way a human brain thinks. We don't have to have any reason or logical impetus pushing us to think a certain thing (emotions, intuition, imagination, or instinct may push us to think a thing). These thoughts appear to happen spontaneously even though the laws of nature may have determined this to be the case.

Silicon machines just aren't made of the same 'stuff' that we are. We may be able to model the human brain on a computer one day but will this mean it will be conscious? Again, something similar to human or animal consciousness would need to emerge from this combination of hardware and software for this to happen. That seems like an empirical impossibility right now as we don't think this can occur. Computers aren't 'living' like biological sapient creatures (no DNA, cells, neurons) even if they can act like them. Perhaps a fusion of living synthetic cells and computer hardware that uses software for its logical processing will be the only way we achieve an AI singularity. But without a biological component, a thinking machine will never be able to think like a human does. Silicon chips, transistors, hardware, circuits, electricity, and magnetism just aren't functionally the same as the emergent biological properties of life such as cells, DNA, metabolism, evolution, emotions, and brain chemistry. AI will always be a difference in kind because of this. It will think in its own way. Probably better than humans in many if not most ways but not all.

  • 1
    Infoatu. I agree that solipsism is something AI has to explain. I mean the genuinely intelligence machine will also be unable to "prove" the existence of the external world. Maybe the right theory of intelligence will tell us why a mind can't prove reality exists. For the TT: sure, the classic AI view is so what, as long as the machine passes the TT. But what if the theory of how to make it pass (computationalism) is false? The computation needed is regarded as impracticably huge. Maybe there's another way. For biology, we shouldn't assume it is un-mechanisable.
    – Roddus
    Dec 21, 2017 at 1:25
  • @Roddus true. Biology may one day be mechanized (in fact synthetic cell creation is already in its early stages — genome synthesis and artificial cell-division). I also wonder what 'other' types of life this could one day constitute (cyborgs). But a cyborg would still have a biological component rather than a purely mechanical one. However, if atom for atom a human-like brain can be recreated, it should be conscious given the laws of this universe. Whether we could replicate other types of non-human or non-animal consciousness (one which is a total product of human engineering) is not known.
    – user28485
    Dec 21, 2017 at 4:32

Not the answer you're looking for? Browse other questions tagged .