by Benny Mattis
This is a response to John R. Searle's "Chinese Room" argument that I originally wrote for Louise Antony's Philosophy of Mind class in Fall 2011. This paper also won the UMASS Amherst Philosophy Department's Jonathan Edwards Prize in 2012.
*****
Meaning Like We Mean It
The Chinese Room
Philosopher of mind John R. Searle, in an
article entitled “Can Computers Think?” argues that minds are not merely
programs that can be run on digital computers, but rather effects of some
“causal powers” of the brain.
Searle makes clear that, while the causal powers of the brain are
capable of producing something called “semantic meaning” or “intentionality” or
“mental states,” mere programs for digital computers have no way of producing
those phenomena, since they are only syntactical abstractions and devoid of any
semantic content.
He makes this point by performing a
thought experiment called the “Chinese Room.” The experiment places Searle in a room with a “large batch
of Chinese writing” (417), which is indistinguishable to him (unable to read
Chinese) from “ meaningless squiggles” (418). This assortment of squiggles, unbeknownst to Searle, is called
a “story” by the organizers of the experiment (418). Additionally, he is given a second “batch” of squiggles
along with rules in English on “How to correlate the second batch with the
first” batch” (this set is secretly called “questions), as well as English
instructions on how to reply to combinations of squiggles from the second batch
with members of a third batch of squiggles; this third batch is called
“answers” by the hypothetical organizers of this experiment (418). In short, Searle is given all the tools
he needs to answer questions in Chinese about a story written and read in
Chinese, all without actually “understanding” a single character of Chinese.
Searle compares this “Chinese subsystem”
with an “English subsystem” wherein he is simply given a story in English,
followed by questions about the story, which he can answer with the full
knowledge of what his words are describing. Searle maintains that no matter how complex the English
instructions (analogous to a computer program) are, the Chinese squiggles will
be no more “understood” in their semantic meaning in the same way that the
story, questions, and answers in English are understood (418). This notion is set in opposition to the
thesis of Strong A.I., which views the problem of creating artificial mental
states as simply a problem of designing the correct program; if Strong A.I. is
correct, the Chinese subsystem could be said to “understand” the semantic
meaning of the conversation if only the program were written properly. Searle suggests that the essence of
intentionality is not in the realization of a computer program, but rather in
some other “causal powers” of the brain (424). This conclusion evokes a plethora of responses, to which he
replies somewhat unsatisfactorily at times, especially regarding the specific
requirements for semantic meaning to obtain in any given biological or
mechanical machine.
Clarifying the Problem
Programs as syntactical structures could
very well exist as thinking minds in the same way that shapes “exist” as real
objects or groups of objects. Shapes
do not have any “causal powers” necessary to realize them, other than
functional organization; likewise, there is no reason to believe that there
must be some “causal powers” in addition to mere functional organization in
order to produce an actual mind.
The sun, for example, might be an example of a sphere. Does this mean that ‘anything which
causes a sphere to exist must have causal powers at least equal to those of the
sun?’ Of course not; the only
thing needed to be a sphere is for some matter or energy to fit a certain
functional pattern. Likewise, on
the view of Strong A.I., the only thing needed to realize intentionality is for
a machine to run a certain functional program.
If nothing realizes a mind-program, then
there are no minds instantiated, and quite obviously no meaning attached to any
symbols at all. The program is
just an abstract syntactical structure, which in fact is incapable of referring
to anything in reality. Even if
another step is taken, and this program is realized, it is incapable of
semantic meaning if it is not given anything real to attach meaning to—a string
of inputs, for example, via a keyboard, light sensor, or other object sensitive
to the outside world. Bruce
Bridgeman thoughtfully remarked in his reply to Searle that, “A program lying
on a tape spool in a corner is no more conscious than a brain preserved in a
glass jar” (427). A computer
program is not an existing mind until it has been instantiated and given
inputs. However, once it starts
getting tangled up in reality, semantics could become not only possible but
even inevitable.
The Secret Life of Thermostats
When a calculator makes a calculation
given some inputs, it is actually performing certain cognitive processes
(regardless of whether it is self-aware) and giving an output that really means
something about reality. It may
only manipulate symbols based on how they are represented in relation to each
other (much like the squiggles Searle manipulated in the thought experiment),
but that is all that mathematicians are doing when they manipulate quantities
(which are only known in relation to each other) according to memorized
rules. In this way, the warming
and cooling activity of a thermostat also means something about the belief- and
desire-states of the thermostat, which are directly related to the world around
it. In Searle’s Chinese room,
there is semantic meaning attached by
virtue of the system itself to input and output symbols, regardless of
whether anyone actually understands the meaning.
It may not be surprising that Searle does
not agree with this ascription of semantic ability to an object as rudimentary
as a calculator. In his rebuttals,
Searle differentiates between “intrinsic” and “observer-relative” ascriptions
of intentionality (452). Human
beings have intrinsic intentionality, because they know what their actions and
words will mean to an observer, but the thermostat does not understand its own
actions at all; an intelligent observer is necessary for its actions to be
understood in relation to the temperature and settings in objective
reality. Thus, Searle would say
that any intentionality a calculator has is strictly observer-relative. Searle remarks that we speak as if
thermostats have beliefs “Not because we suppose they have a mental life much
like our own; on the contrary, we know they have no mental life at all”
(452). It seems quite obvious to
Searle that the beliefs of simple mechanisms are not only different from those
of people, but entirely nonexistent.
The dichotomy between “a mental life much
like our own” and “no mental life at all” is questionable at best. No functionalist or believer in Strong
A.I. would say that the experience of a thermostat is similar to that of a
human; under those views, the experiences would only be similar to the extent
that the thermostat’s functional organization resembles that of the human
brain. The thermostat is not aware
of how its actions appear to an outside observer, and therefore cannot intend
to express an idea the same way one person can intend to communicate with
another; it does, however, by its very nature compute thermal inputs and
respond appropriately, and this could very well be considered a proper type of
mentality.
At this point, it makes sense to question
whether full consciousness is necessary for intention or semantics. An intoxicated, barely-conscious
individual presents a problem for this assumption: if “Bob” becomes too acquainted with his favorite substance,
for example, he may begin to say things without actually intending to express
those ideas beforehand. It is
clear that a certain type of intention is missing in Bob’s case; he is
incapable of assessing himself through an approximated observer’s perspective
and acting appropriately to express his feelings. I do not think, however, that Searle would suggest that Bob
lacks the ability to express semantically meaningful statements.
Bob’s words semantically mean the same
thing that they would mean if he uttered them sober. Bob could not change this fact even if he wanted to; thus,
the inevitability of semantic meaning.
If we picture Bob doing math, his mathematics would not be meaningless
or unintentional merely because he is not in a state of perpetual reflection on
the concept of quantity and the axioms on which his math is based. Likewise, a computer’s calculations or
the calculated heat adjustments of a thermostat are not rendered semantically
meaningless by the mere fact that they are not conscious of their own cognitive
acts, or what those acts are ‘about.’
It would be absurd to claim that a calculator has the same mind as a
mathematician, but it is plausible that the aspect of a mathematician’s mind
involved in adding and subtracting numbers could be instantiated in a
calculator with sufficient operational similarity to the their mental
problem-solving method.
The fact that there is a fixed causal
relation between the inputs and the outputs in the Chinese room, akin to the
relation between the problems given to a drunken mathematician and his or her
reflexively scribbled solutions, is the source of the semantic content of what
it says; it does not merely make it possible for people with brains to ascribe
meaning to them, as Searle suggests, but rather it is the source of the meaning
itself.
From Calculators to Terminators
Searle would likely respond to this idea
in the way he responded to a number of other replies to his article in Behavioral and Brain Sciences (452):
Even if formal tokens in the program have
some causal connection to their alleged referents in the real world, as long as
the agent has no way of knowing that, it adds no intentionality whatever to the
formal states.
There
are two possibilities I find likely that Searle refers to when he says, “The
agent has no way of knowing that.”
The first is that the symbol does not inspire a mental representation of
the referent when it reaches the person in the room, or the central processing
unit in a computer. The second
interpretation is that the person in the room has no idea that the symbols
given to him are even connected to the outside world at all.
The
former interpretation suggests that Searle’s problem, and indeed the problem
with common intuition, is that “The English subsystem knows that ‘hamburger’
means hamburger. The Chinese
subsystem knows only that squiggle squiggle is followed by squoggle squoggle”
(453). I do not believe that there
is as big a difference as Searle suspects between these two semantic relations. The Chinese system gets a squiggle
squiggle, relates it to a series of rules on the grammar of “squiggle squiggle”
and the squiggle’s relation to others recorded in the system’s memory (for
example, “squiggle squiggle has relation X to snuggle duggle) and outputs
“squoggle squoggle” in an appropriate way. The human sees “hamburger,” decodes the visual stimuli and
matches them with an associated cluster of stimuli (the “hamburger-concept”),
and outputs a response based on its memory or “belief-state.” Thus, it appears that concepts of concrete
objects, like formal mathematical quantities and Chinese “squiggles,” are known
only in relation to each other.
Could not the networked relations of
squiggles in the Chinese room (or, indeed, those of binary symbols in a
digitally programmed computer) be made in a way similar to those that the English-speaking
human draws upon intuitively upon hearing “hamburger?” The Chinese room is disposed to relate
inputs to outputs with the use of past inputs; the human is disposed to relate
sense-data to actions based on past sense-data (for example, the memory of
eating or hearing about a hamburger).
There is no reason not to believe that an isomorphic similarity between
the two would be sufficient to produce intentionality, whether such
similarity be realized in a death-dealing android or a Turing machine made of
toilet paper.
With that counterintuitive notion, we can
move on to the second interpretation of Searle’s response: how, in fact, does
the person in the Chinese room know that a Chinese hamburger squiggle even
refers to anything concrete at all? Putting oneself in the shoes of the man in the Chinese room,
it seems quite obvious that a certain sense of what is “real” is
lost—everything in the Chinese subsystem is mediated through strange symbols.
That
sense of what is “real” may be just that—a sense. Rene Descartes noted the possibility that all of one’s
senses are an illusion, and Searle himself even remarks how intentional states
are not a matter of the outside world, but rather private mental phenomena: “It
is the operation of the brain and not the impart of the outside world that
matters for the content of our intentional states, at least in one important
sense of the word ‘content’” (452).
As many modern philosophers are all too aware, we do not know for a fact
that our words really refer to anything outside of arrangements of related
concepts in our minds; the fact that Searle acknowledges this suggests that
this is not what he meant by “The agent has no way of knowing that,” but it is
a topic worth bringing up anyway, and sheds light on why the Chinese subsystem
seems to be different from the English one.
We act according to the inputs we are given, and produce outputs based
on our memory and tendency to pursue certain goals, just like the man in the
Chinese room.
Why,
then, does the English subsystem seem so different from the Chinese subsystem
in the Chinese Room experiment?
The subject actually does not know
as much in the English subsystem as he does in the Chinese subsystem. In the Chinese subsystem, the
alienation of Searle’s responses from their referents is made explicit; he is
only acting by relating formal symbols to another, such as the squiggle for
“hamburger” with the squoggle for “ketchup,” and as a result the absurdity of
his formal operations are brought to the fore. In the English subsystem, Searle is faced with a feeling
that he ‘knows what he is talking about.’
He is still only manipulating formal packets of quantitative sense-data
(the probability of concurrence of taste-value K and taste-value H, perhaps),
but his intimate familiarity with those packets produces in him the illusion of
a special connection to the outside universe.
Programs Writing Programs
Now, it’s possible that one can never
know for sure whether what it’s like to
be a robot is similar to what it’s
like to be a person; we can never verify beyond doubt that a computer feels
“thoughts, feelings, and the rest of the forms of intentionality” (450) in the
same way we do, just as we can never verify beyond a doubt that even John R.
Searle has such feelings. The fact
that a thermostat probably lacks a mental life like our own, however, does not necessitate that it has nothing
resembling mentality at all. There is no reason to believe that
functional organization is not, in fact, the property that gives the brain its
“causal powers” to produce consciousness and intention. This leads to seemingly ridiculous
results, such as the possibility of a water-pump computer gaining
consciousness, but I would dare say that it is not harder to believe than the
suggestion that intentionality is “Likely to be as causally dependent on the
specific biochemistry of its origins as lactation, photosynthesis, or any other
biological phenomena” (424). This
is weird stuff.
Neither
human nor computer actually knows whether their internal tokens refer to
anything external. Both human and
computer ascribe semantic meaning to tokens gathered from reality by
associating them with each other according to rule-bound functional relations. The question of whether their mental
states are similar remains open.
Strong A.I. may still be wrong when it comes to consciousness, but it
also may be right—when it comes to semantics, however, a denial of such power
to artificial intelligence would also put our own abilities in question, as
Bob’s example demonstrates. By
virtue of being instantiated in the real world and being given input, a
computer ascribes meaning to that input (by a fixed relation between
interconnected symbols with their own hidden causal connections to an external
reality) and produces an output, which also means something relative to the
input and the program’s internal states.
This is, for our purposes, indistinguishable from a brain taking
sense-data from the real world, relating it to other packs of sense-data, and
producing an output which means something relative to the input and the
person’s internal states. A
computer, like a person, ascribes meaning to its inputs and outputs by virtue
of the fact that it exists and is consistently causally connected with the
real world—it may be challenging to learn the language they are processing
symbols in, but there is a language in both cases nonetheless, semantic
reference included. Most
importantly, all of these similarities apply regardless of what material the
machine is made of; whether gray matter or water pipes, a program instantiated
produces equally semantically meaningful output, whether it “means to” or not.
Works Cited
Searle,
John R. “Minds, Brains, and
Programs.” Behavioral and Brain Sciences.
1980: 3, 417-457. Print.
