an essay on john searle's chinese room argument and artificial intelligence
 
Understanding in the Chinese Room
by Anthony Peter Iannini | last edited: 06.11.2011 | originally written: 2000		  


Essay Overview:
_______________________________

This is an essay about John Searle's Chinese Room thought-experiment and whether or not an artificial intelligence can ever, principally speaking, have the same degree or type of intelligence, understanding, or consciousness that a human does (R1). The most important question, wherever one falls in terms of a position on this issue is what reason(s) do we have for making assumptions about beings that we may be capable of creating artificially?

Strong artificial intelligence is the positive view that artificial computer systems can, in principle, have understanding, intelligence, or even consciousness equivalent to that of normal adult humans. Weak artificial intelligence is the negative view that artificial computer systems can not, in principle, ever have understanding, intelligence, or consciousness like that of a normal adult human for some reason.

For clarity and emphasis, Searle's quotations have been colored in light green and quotations by others have been colored in light blue. Section headings are written in bold salmon.

Numerous and diverse replies and rebuttals have been made for and against Searle's argument since it was written during the infancy of artificial intelligence research in 1980 (R3, R10). I begin the essay by explaining and summarizing the Chinese Room scenario and then move towards explaining Searle's refutation of strong artificial intelligence.

I conclude with my own rebuttal of Searle's claims and conclude with my own interpretation of the issue that takes no strong side on the issue of strong or weak artificial intelligence but rather appeals to an agnostic and skeptical position that denies our ability to be certain concerning the first-person phenomenal reality possibly experienced by such systems.

 



0. Introduction: The Chinese Room thought-experiment, and subsequently derived argument, was presented by John Searle as a principled refutation of theoretical strong artificial intelligence (R1). Searle uses the scenario in order to argue that any potential artificial computer system could not ever understand the way a normal adult human does as demonstrated by his example.

In this essay, I will (1) describe the Chinese Room and explain how Searle sets up the experiment. Then, I will (2) describe how Searle interprets his thought-experiment and how he uses it to argue that it proves something fundamental about the nature of human understanding versus what he sees as the non-understanding of the Chinese Room occupant.

1. The Chinese Room and It's Occupant: Searle presents a situation in which a monolinguistic native English-reading man, who does not understand any Chinese symbols, is placed in an isolated room. In this room there are stacks of papers with written Chinese symbols on them.

While isolated in the room, the man is slid sheets of written Chinese under the door, to which he is supposed to respond by sliding the correct written responses, in Chinese, back under the door. As he does not understand any Chinese symbols, the man is incapable of responding.

In order to overcome his lack of understanding of Chinese symbols, the man is provided with rules in written English that tell him how to respond to one Chinese symbol with another so that it appears as if he actually understands Chinese to someone outside the room. We could imagine that these instruction sheets would appear something like what is shown in the following fictional examples:

  • If you see chinese symbol then send back chinese symbol .
  • If you see chinese symbol chinese symbol chinese symbol then send back chinese symbolchinese symbol chinese symbol .
  • If you see chinese symbol chinese symbol chinese symbol chinese symbol chinese symbol chinese symbol chinese symbol then send back chinese symbol chinese symbol chinese symbol chinese symbol (R11).

We could imagine that the man could learn or even memorize these rules and move towards more and more complex versions of the rules. In fact, the man would only be limited by his ability to memorize the characters and rules that govern them which, for a computer system rather than a human, could allow for incredible speed and complexity of symbol exchanges.

Searle suggests that the man can become so proficient at manipulating Chinese symbols based on rules given in English that he can appear to be fluent in Chinese to someone who reads Chinese and who is outside of the room. This entire example mirrors what has been called the Turing Test, which is passed when a computer is sufficiently indistinguishable from a human in terms of its input and output responses (R12).

2. Searle's Argument Against Artificial Understanding: Searle’s main point is that, no matter how well a program mimics or simulates understanding, it will never really understand the way that a human does in the same way that the man in the room will never really understand Chinese, though it may seem that he does.

Searle is not making the claim that machines are incapable of any thought. He is making a more difficult to capture claim that whatever it is that goes on inside an artificial "mind" it will never be fully aware of the meaning of the symbols it is working on in the same way that a normal adult human can understand his or her native language or learned language.

We humans know, Searle would argue, what each word means or refers to and we can understand how these symbols in our language work together without having to fake it the way the man in the Chinese Room has to.

Searle then goes from his example of the Chinese Room and makes the claim that not only do we understand in a way that artificial systems can not but that this is primarily because we humans possess the right "causal powers" to do so. Brains and biological matter, Searle argues, are composed of the right kind of substantial material to have intentional states or states that are about the world such as beliefs and desires. His view is that the actual physical properties of the brain are intimately tied to the mind and that the two can not be separated.

According to Searle, the human mind can not be realized on a physical system other than the biochemical medium in which it actually resides, and, if it were able to be realized in another medium, such a medium would require the same "causal powers" as the brain.

Searle’s formal argument, one that is made clear in later articles(R2), is that while computers have syntax, they have no semantics, and syntax, by itself, is not a sufficient condition for semantics. In other words, computers have the ability to process and manipulate symbols but they can not reference the meaning or content of such symbols.

Searle has made a number of claims at this point and throughout his thought-experiments, claims of knowledge concerning substance causality, and his formulation of a linguistic theory of why artificial minds must be forever dumb, he has left all of his myriad scattered claims vulnerable to a number of foundational objections.

3. Three Objections to Searle's Claims: At least three possible salient objections to Searle's argument are available at this point in the debate. For clarity, I brliefly list each of the objections before delving more deeply into the first two and spending the remainder of the essay on the third objection, which I find to be the most damaging to Searle's claims against artificial minds.

(O1) The objection that perhaps the whole system but not any particular part of the system does understand.
(O2) The objection that Searle's theory of "causal powers" is nothing short of magical.
(O3) The objection that Searle doesn't understand understanding.

3.1 The Objection to Searle's Wholly Systematic Lack of Imagination: (O1) The first objection is that Searle seems to ignore the question of whether or not the whole system of the Chinese Room, including the Chinese illiterate man inside, the instructions, the stacks of Chinese symbols, the walls, the door, the slit under the door, etc. all may be capable of understanding in the way humans do as a combined conglomerate.

This line of objection to Searle's argument has been termed the systems reply. We can assume that it is possible for the whole system to understand. Why? Bbecause we can not know, principally, either if the whole system of the Chinese Room does or does not understand in the way Searle is referring to, but we can know that, behaviorally through observation of the whole system, it will act indistiinguishably from a system that understands.

Whether or not this behavior can be attained from the speed and complexity of handling the Chinese symbols has already been answered by Searle who already allows in his initial scenario that the system can become as quick and proficient as possible in terms of delivering answers when given a question in Chinese.

3.2 The Objection to Searle's "Causal Powers" as Both Powerless and Without Cause: (O2) The second possible objection that can be brought forth against Searle's argument is that his theory that only certain substances possess the correct "causal powers" for human-like understanding is an assumption that has no basis but rather rests only on prejudicial assumption about the first-person experience of non-biological physical matter.

Another reply to Searle along the same lines is that of William G. Lycan, who presents the functionalist reply from Ohio State. In his article, Searle writes, "Stones, toilet paper, wind, and water pipes are the wrong kind of stuff to have intentionality in the first place- something that has the same causal powers as brains can have intentionality." (R4).

All of these items, of course, have no intentionality by themselves. It seems as though Searle has missed the point however. As Lycan points out, if you were very tiny observer in a brain, all you would observe would be "...neurons stupidly, mechanically transmitting electrical charge..." (R5). and that there is no understanding in such a cellular system.

To say that there is absolutely no intentionality in the cell by itself, however, seems to fall victim to the Sorites paradox (R6).

This problem is overcome if we adopt a notion much like the one pointed out by Marvin Minsky in his reply to Searle, which is taken from a work of science fiction. The author writes, "The ancient concept of belief proved inadequate until replaced by a continuum in which, it turned out, stones placed near zero, and thermostats scored 0.52. The highest human score measured so far is 67.9..." (R7).

While this example is arbitrary, it points to the notion that we may find it useful to consider things (such as thermometers and brains) as having degrees of intentionality rather than a binary concept of either having or not having any cognitive functions.

3.3 The Objection that Searle Does Not Actually Understand Understanding: (O3) The third objection available is that, even more deeply than (O1) or (O2), Searle does not present a formulation or clear idea of exactly how we humans understand our language, or anything else for that matter, in the deeper way he says we do. There appears to be, at least on the surface of human cognition, a lot of room for automated and behind-the-scenes processes related to our own human phenomenal consicousness that act very much like the mechanism of the ignorant man in the room.

For instance, we humans are not immediately aware of how our mind connects from the image of an ape to the image of the jungle but the relation exists and moves from one representation to another nonetheless. Even our grammatical ability has mechanistic hard-wired, man-in-the-Chinese-Room-like, mechanistic modes of operation that are subconscious in their execution while we fluidly speak and write using such rules.

Bruce Bridgeman, in a reply to Searle, examines the notion of intentionality by considering the cellular level of biological organisms (R3). There seems to be no intentionality (in the way Searle uses the term) in single-celled animals, but only reflex responses to outer stimuli and chemical changes. However, the brain, at some point does have intentionality, although its individual components do not.

The fact that intentionality is an ability of non-intentional parts lends itself to the notion that syntax can, in fact, constitute semantics at a certain level of complexity and organization. Bridgeman refers to the "illusion" of consciousness and other higher cognitive functions. Mental characteristics, in my opinion, need not be reduced and thereby referred to as illusory. Rather, there must be an acknowledgement that abilities or functions not present in the parts of a system may arise from the interaction of such parts.

Zenon Pylyshyn, in his reply to Searle, imagines a case in which each of the cells in an individual’s brain is replaced by an integrated circuit chip that maintains an identical input-output relationship as did the original neuron (R8). By Searle’s criteria, if this brain were eventually replaced entirely by circuits that operated just like the neurons, the person to whom this new circuit-based brain belongs would act just as he did before but would stop meaning or understanding anything.

Supposing that this replacement yields an identically inputting and outputting person, it would be absurd to assume that he no longer means or understands what he is saying. Again, Searle’s Chinese Room illustration, which distinguishes between understanding and non-understanding of a language says nothing about how humans understand. Obviously, the man in the Chinese room does understand English and does not understand Chinese, but why? Answering this crucial question will be the focus of the following Section 4.

4. Objection Three (O3) Continued, Searle's Misunderstanding of Syntax and Semantics: This section focuses on the grand leap Searle makes from the stupid nature of simple syntax that machines have to the deeply aware and understanding nature of semantics that he thinks we humans alone possess. For any system to understand a language, whether it be spoken or visual, it must have three necessary components:

(1) a database of symbols.
(2) a set of rules for manipulating such symbols.
(3) a method of referring to the meaning or semantic content of such symbols.

By symbols, I mean anything that stands for or represents something else. A symbol could be a written word, a spoken sound, or a scene from the environment. The set of rules for manipulating symbols include grammar, constructing a mental image, or any other way of arranging symbols in a way that serves some purpose- which, in the case of humans, seems overwhelmingly to be communication with other humans.

Finally, to understand at the level of human languages, a system must be able to refer to the semantic content or meaning of a symbol. For example the word red, when thought of by a functional human, illicits a mental image which contains a visual representation of red, which is not a part of the word red itself, but rather, what the word refers to or means.

Similarly, if I ask someone to think of everything that they know of as being "red" they can imagine a number of things (such as apples, fire engines, and balloons) that are associated to the word red.

The monolinguistic English speaking man in Searle's Chinese Room has (1) and (2) but not (3). Without the ability to refer to Chinese symbols, it is rather irrelevant if the man can speak fluently based on memorized rules for manipulating such symbols. But, if the man were able to know that a symbol such as chinese symbol for red in green referred to the color red in Chinese (which this symbol does refer to) then he would have some, albeit minimal, understanding of Chinese.

What underlies the ability to refer? Reference is an unconscious, automatic process- formal or syntactic in nature. My basic argument against Searle is that symbols, in a system that "understands", refer to other symbols.

When we say that we understand some thing x, it is because we can describe x in terms of properties or features of x such as y and z. If someone is asked to describe what the word "red" means or refers to, they can not linguistically describe the color red in itself. One could say of red that it is a color, that it is has a wavelength of about 700 nanometers, but it can never be, in itself, described to a person that has always been blind.

The underlying composition of symbols themselves, as parts of memory, must be syntactic in nature. When we form a mental image of an apple, we do not have conscious access to the neuronal structure from which the information was retrieved (if that is, in fact, something like how the representational formation works)- though it must have been derived from some pattern in the mind.

Note the above graphical chart that shows the process of interpretation, from syntax-type automatic processes on the left through higher level, concpetual semantic processes on the right. The entire chart describes understanding, as this is a word that refers to awareness of the final product of perceiving a symbol like the hearing the word "tiger" in the chart above.

If someone can be said to understand what the word "tiger" means, then this process of accessing knowledge and concept building must occur behind the scenes rather unconsciously because it is very fast (from our temporal perspective as humans) and, unless it occurs from magic, then it must happen through algorithms and computer-like search routines in the mind or brain or both (on a side philosophical note, there is no reason to assume mind-brain identity as a philosopher though such assumptions can be useful in theorizing about scientific-realist formulations such as in the case of how neural cellular pathways and structures may be running programs to find information and store it in the brain).

Perhaps around or between steps (D) and (E) in the chart above, we begin to become consciously aware of (A) through mechanical and hard-wired processes in (B) and (C) that ultimately result in the representational formation of a thought about the tiger that may be an image, a sound, a scary thought about a past feeling of a tiger's claws ripping into our flesh, or maybe even a silly thought about our house cat growing ten times larger. What representations a word produces in an interpreter depends on the word in question and the history of the interpreter relative to that word as well as a great number of possibilities about the present state of the interpreter.

If the interpreter is a native expert Spanish-speaker, then to hear the word "Casa" will, under normal circumstances, quickly result in the meaningful representation of a house being formed in the Spanish-speaking interpreter unless any number of things has happened. When I mentioned this person's history relatived to a word, this history could include being hit on the head by an anvil falling from a plane that happened to be dumping anvils over Barcelona that day and, in that case, the anvil may very well interrupt or remove the vocabulary from the interpreter or shut down the vital systems of the interpreter completely.

If the word is lost, through blunt force trauma, alzheimers disease, a virus, cancer, or anything else, then this will be a complete hinderance to understanding, meaning, conceptualization, or any semantic or even syntactic results from being achieved. One can ask, "¿Donde es tu casa?" all dia long, but if the person you are asking has an anvil where their cabeza used to be, the process of interpretation is not going to get started without some major surgical procedures.

If someone shouts "Achtung!" to another person who does not have any history of knowledge of this particular German word, then there will be little way to understand what is being shouted except, perhaps, for the intense body language of the Lederhosen-wearing, large Swiss man who is warning you not to continue skiing on that mountainside in the Swiss Alps because he knows (and you obviously do not know) that the Swiss Ski Patrol is about to cause the morning avalanche with a Howitzer cannon, as they do each morning to clear the slopes.

Because the human mind so fluidly refers to symbols and is able to communicate through structured manipulations of such symbols, it seems as though there is fundamental understanding beyond that of other systems. Compared to current artificial systems, human minds are able to understand concepts in a much richer context, but the same type of semantic properties are emerging from syntactic aspects of the system.

By richer I mean to a greater degree. All humans understand one thing or another, such as a language, to various degrees. A three-year old, a ten-year old, and a forty-year old English professor all (normally) understand the English language to greater degrees. Humans are programmed by genetic information and by the environment, which results in complex, goal-directed, dynamic programming.

Computers are becoming more complex, more able to store data, and are being given various sensory peripherals to perceive their environment. Systems, such Schank and Abelson’s (the one attacked in Searle’s article), have been designed that can refer and understand within specific domains.

Though debated, empirical analysis yields evidence that basic semantics have already been reached with the advent of programs, such as Terry Winograd’s SHRDLU (R9) program, that can refer in highly restricted environments.

It is difficult to imagine the common Von Neumann architecture computer understanding anything in large part because of its serial processing and relative simplicity in comparison to the human mind and it’s programs. However, it is also difficult for us, as humans, to image having the inputs and outputs of a desktop computer- namely, a standard keyboard, mouse, and a few other peripherals.

In the words of Roger Schank, "Certainly we humans understand, but does that lump of matter we call our brain understand? All that is going on there is so many chemical reactions and electrical impulses, just so many Chinese symbols." (R10). When we look at the brain at one level, there seems to be no understanding. Yet, these non-understanding parts together have the ability to understand- which I have defined as the ability to reference symbols with other symbols.

These symbols are, at one level, syntactic, formal, and impenetrable to conscious cognition. As to Searle’s assertion about the right kind of substance and the correct "causal powers", there seems to be no reason why electro-chemical potentials across organic neurons are better than or more sufficient for understanding than silicon, toilet paper rolls, or anything else.

5. Conclusions About the Chinese Room and Searle's Claims: I would summarize my position as one that remains skeptical of proponents and detractors of aritificial minds alike though I tend towards believing that anything that behaves as if it is intelligent and as if it understands, no matter its substantial make-up, should ethically be treated as an agent that is aware of itself.

We can never know what it is like to be you, me, or anyone or anything else with certainty though we all desire to be treated as if we are capable of understanding and feeling.

The difference between biochemical and silicon-metallic may be no different when it comes to the functional capacities of systems. If there is some substance-related causal powers that allow some forms of matter to experience and others to not do so, then we will probably never know this truth because, behaviorally, the systems we create may even come to form beliefs about their own experience of the world and understanding that, while possibiy false in some noumenal sense, are never possibly capable of being verified.

And, these systems may come to believe that they understand, feel, experience, etc. whether or not they do and all we can do is err on the ethical side of caution and a deep awareness that we are forever ignorant not only to artificial minds and the worlds they experience but to each other, animals, plants, and to the stranger possibilities that consciousness may be a part of even seemingly innanimate reality- a last point that we can not take for granted as we begin to turn the innanimate into the animate.

For, what part of ourselves, if removed or chopped off, would not also be so much lifeless protein without electricity or chemical signalling? Think of frontal lobotomy patients of psychiatric treatments done in the last century.

While much of their personality was once thriving inside the pre-frontal neural cortex, after such procedures, it is no more than a piece of dead tissue- no more alive, sentient, or part of their now deficient personalities than anything else detached from them, be it a loaf of bread or a brick of gold.





References:
____________________________

(1) Searle, J. (1980), "Minds, brains, and programs", The Behavioral and Brain Sciences 3, 417-457.

(2) Searle, J. (1990a), "Is the brain's mind a computer program?", Scientific American 262(1):26-31.

(3) Bridgeman, B. (1980), Reply to: "Minds, brains, and programs", The B.B.S. 3, p. 427.

(4) Searle, J. (1980), "Minds, brains, and programs", The Behavioral and Brain Sciences 3, p. 423.

(5) Lycan, W. G. (1980) Reply to: "Minds, brains, and programs", The B.B.S. 3, p. 431.

(6) The name ‘sorites’ derives from the Greek word soros (meaning `heap') and originally referred, not to a paradox, but rather to a puzzle known as The Heap: Would you describe a single grain of wheat as a heap? No. Would you describe two grains of wheat as a heap? No. ... You must admit the presence of a heap sooner or later, so where do you draw the line? (Taken from The Stanford Encyclopedia of Philosophy Copyright © 1999 by Stanford University and Edward N. Zalta).

(7) Oliva, R.D. (2063) Robot reflections, Phenomenological Science 67:60. (referred to by Minsky, Marvin in (1980), Reply to: "Minds, brains, and programs", The B.B.S. 3, p. 439.).

(8) Pylyshyn, Z. (1980), Reply to: "Minds, brains, and programs", The B.B.S. 3, p. 442.

(9) SHRDLU operates in a ‘block world’ in which it must reference the semantic content of words in conversation with human keyboard based input. This example was referenced from: Gardner, H. The Mind’s New Science: A History of the Cognitive Revolution, 1985. pp. 158-59.

(10) Schank, Roger C. (1980), Reply to: "Minds, brains, and programs", The B.B.S. 3, p. 447.

(11) It should be noted that the Chinese symbols in this example are not arranged in any meaningful way, but just to show possibilities.

(12) The Turing Test, conceived by theoretician Alan Turing (1912-1954), could be passed when the human participants involved would be unable to determine which of a random sampling of systems were human and which were computer only. As of mid 2011, no system has yet passed the test.

 

All content on hiartx.com is by Anthony Peter Iannini © Copyright | All Rights Reserved. If properly attributed and referenced, all images and excerpts of written content from this site may be used for non-profit and/or educational purposes freely. Please provide a hyper link back to the website page where the images or text was found. E-mail contact regarding all uses of content on this site is appreciated. For all other uses of content on this site, please e-mail me at: apiannini@yahoo.com.