7

The following reply to the Chinese Room argument came to my mind recently:

The whole activity of manipulating symbols described in CR is understanding in the usual sense of the word. My understanding of, say, English is nothing more than manipulating symbols in my mind. The only (extremely crucial) difference is that the manipulation process in my brain happens instantaneously, effortlessly, and at will, thus creating a very strong illusion of a "real" understanding (as opposed to a "fake" one the computer has).

I am 100% sure that this response has been made a thousand times in the past.

I am also almost certain it is deeply flawed. I am interested in knowing what its flaws are.

  • I want to say this is called the "ostrich defense" but I can't remember if that it's name in the literature. It also represents "structuralism." – virmaior Feb 12 '17 at 22:45
  • 2
    You are 100% right on the first, this is the oldest and the most popular response, and it is called the "systems reply" to the Chinese room:"The basic "systems reply" argues that it is the "whole system" that understands Chinese". As for "deeply flawed", Searle thought so, in fact he thought that people accepting it are "under the grip of an ideology" (of strong AI). But it seems that his basis for this are just some vague "intuitions", and Dennett charges him with "intuition pumping". – Conifold Feb 12 '17 at 23:56
  • @Conifold, that's interesting. I have come across the Systems Reply numerous times while reading about CR, but always thought this response was different from what I'm describing in this question. Is saying that the whole system has understanding, and saying that the process described in CR is understanding, essentially the same thing? – Constantine Feb 13 '17 at 5:45
  • 3
    Well, saying that the whole system has understanding is relevant in this context only if what goes on in CR amounts to "having" it. Of course, some opponents of CR might put it differently. Dennett says that what we colloquially call "understanding" is a composite of cognitive activities wrapped into a single concept (similar to Hume's treatment of causality), that Searle's reifies it into a figment and then chooses a degenerate example, far enough from prototypical examples, to create intuitive discomfort, which "disproves" that "understanding" is there. – Conifold Feb 13 '17 at 19:10
3

You've packed a lot into your refutation of CR. The first part:

The whole activity of manipulating symbols described in CR is understanding in the usual sense of the word. My understanding of, say, English is nothing more than manipulating symbols in my mind.

This can be interpreted in many ways:

  • As Conifold points out, it corresponds in a way to the systems reply. You commented that "Is saying that the whole system has understanding, and saying that the process described in CR is understanding, essentially the same thing?" - to which the answer is yes. The gist of the systems reply is that "the process described in CR" must include both the person inside the room as well the tools that the person is using to manipulate Chinese symbols. Searle's fallacy is that he misidentifies what the process is, when he focuses only on the person and not on the entire system. The person inside the room is only an incomplete part of the symbol manipulation process, which is why they can't be identified with "understanding" the way the entire system can.
  • As Virmaior points out, your idea of "English is nothing more than manipulating symbols in my mind." hints at Structuralism. It also sounds a lot like Later Wittgenstein's "meaning is use": Language is defined by the way it is used, and you can't separate the meaning of a language from its syntax.

The second part:

The only (extremely crucial) difference is that the manipulation process in my brain happens instantaneously, effortlessly, and at will, thus creating a very strong illusion of a "real" understanding (as opposed to a "fake" one the computer has).

Seems to echo Douglas Hofstadter in GEB and "I am Strange Loop": The difference between conscious systems and non-conscious symbol processing systems can be a matter of degree not just of kind. Consciousness arises when a symbol processing system is capable of self-reference, however the self-referential strange loop has to be very tight for it to be truly conscious. Your idea that human consciousness seems real compared to fake computer consciousness because of its speed and near instantaneous character strikes me as being in the same vain as Hofstadter's idea that the strange loop has to be "tight' to some extent for consciousness to arise.

  • Thank you for this wonderful answer. I have read quite a lot of the later Wittgenstein, so everything makes more sense now. Haven't read GEB (shame, it's a must-read) or "I am Strange Loop", though. – Constantine Feb 15 '17 at 16:10
0

As you and others here have said, your CRA reply is essentially the most common one, the Systems Reply. But your statement misses some important aspects, such as Searle's point about formal symbol manipulation, which is the one thing his CRA gets right. To understand how both things can be right: lack of semantics at the Searle-computer level and the potential for understanding at the CR-system level, one should understand the relationships among the Searle-computer and its two kinds of input.

The Chinese Room's heat source is the fact that it produces two levels of Turing computation, both of which are processing the Chinese symbols. Level-1 is Searle himself, acting as a UTM-computer (a universal Turing machine; like a CPU). Level-2 is "the system" whose computation of the Chinese is determined solely by the program and its execution configuration (memory, state, etc.). Note that the existence of the two computations is a fact, not as in "I'm stating my opinion as a fact", but as in "it is an objective mathematical fact, like the Pythagorean Theorem, which anyone can understand and verify for themself".

Searle's self-consciousness is a kind of 3rd level of "information processing" (which might or might not be a computation). It allows him to introspect on his own level-1 universal computation. But he cannot similarly introspect on the level-2 computation. For level-2, the Searle-UTM is merely its physical substrate, analogous to a person's neuron-level, not their consciousness-level. Thus, the Searle-UTM never knows what his program is doing, whether it is doing a Turing Test or taxes or tic-tac-toe.

Also note that the Searle-UTM is processing two classes of symbols: #1 the English symbols of the program, which people always ignore, and #2 the Chinese symbols, which captivate people like shiny objects. Thus, to correctly address "symbol manipulation in the CR", one must fully address the three symbol-processing relationships:

  1. the level-1 relationship: Searle-UTM to the English program symbols (#1),

  2. the level-1 relationship: Searle-UTM to Chinese symbols (#2), and

  3. the level-2 relationship: the program's computation to the Chinese symbols (#2).

In the CRA, Searle only addresses the 2nd relationship: his own UTM processing of the Chinese (#2), which he correctly characterizes as formal symbol manipulation, i.e,. syntactic and meaningless or non-intentional to him. Your reply above does not account for this 2nd relationship. For when you speak of manipulating English symbols in your mind, the key difference is that your mind can interpret them, whereas the Searle-UTM cannot itself interpret the Chinese symbols. That's Searle's main point, and he is correct about that much of it. (His mistake lies in missing the other aspects of the two computations.)

Virtually no one addresses the 1st relationship. However, clearly the Searle-UTM must process the program symbols with some level of semantic intentionality.* That is why they must be in English! This fact alone contradicts Searle's own CRA conclusions.

Most importantly, the 3rd relationship is approximately the famous Systems Reply, as mentioned. Searle tries to dismiss the Systems Reply, and apparently people are left to decide for themselves whether they agree, disagree, or prefer to focus on some other aspect of the CRA or science fiction or philosophy, etc.

But the key point about the Systems Reply is this: although Searle cannot discern it, the additional level-2 computation is not merely some philosophical assertion. It is a mathematical fact about how Turing computation works. A universal computation on its #1 program input and its #2 nominal input, instantiates the program's own distinct TM computation on the nominal input. Period. Trying to rebut this would be like trying to rebut the Pythagorean Theorem.

The slight difference here with the usual Systems Reply is that one cannot further assume that the Turing Test is valid. The CR's external behavior does not prove that the program's algorithm is equivalent to a human's internal understanding of the Chinese. But the CRA does not disprove it, so it remains an open possibility. Presumably, that is why over the years Searle has evolved his critique to more properly address computation in general, rather than just the CRA's focus on universal programmability.


* Obviously, full blown self-conscious intentionality is not required for computers, but translating a program's symbolic references into physical entities, actions, and events is required. That goes beyond formal symbol manipulation into enacting non-formal associations.

Ref (and see refs therein): Refuting Searle's "syntax is not semantics" argument.

Your Answer

By clicking "Post Your Answer", you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.