HPAVC/HPAVC CD-ROM.iso/HOMEWORK.ZIP/PHILSICT.TXT

Text File | 1998-07-25 | 28KB | 313 lines

Alan Mathison Turing, in his paper "Computing Machinery and Intelligence"1, proposes the
question 'Can machines think?'.  In considering this question Turing concludes that it lacks
a concrete methodological goal and decides to replace the original question with the
'Imitation Game'.  The Turing test, as the Turing's 'Imitation Game' is generally called, is
a procedure to test whether a machine is capable of humanlike thought.  Simply described, an
interrogator sits with a teletype machine isolated from two correspondents: one is another
person, one is a computer. By asking questions through the teletype machine and studying the
responses, the isolated person tries to determine which correspondent, A or B, is human and
which the computer. If that proves impossible, the computer is credited with having passed
the test.2  Turing's purpose in designing the test was to standardize a method for
determining if machines can act in a human way.
 Instead of asking a rhetorical question that can be answered in a myriad of ways, Turing
 focused on an empirical criterion for judging the "humanness" of a machine.  This new
 criterion, the Turing test, compares the capabilities of humans and machines.  If a machine
 is indistinguishable from the human in all possible ways, Turing suggested we must accord
 thought to this machine.  As Turing put it, "If a computer could not be distinguished from
 a human, then fair play would oblige one to say that it is thinking."3  To this day, the
 Turing test is heralded as a criterion that must be surpassed before a machine can claim
 the attribute of human thought.
Keith Gunderson in his article, "The Imitation Game"4, considers the soundness and
implications of Turing's 'Imitation Game'.  Gunderson takes Turing's statements to mean that
a machine that can play the 'Imitation Game' can be ascribed with thought.5  I do not
believe that this is a statement that Turing would agree with completely.  Turing thought
that the question "Can machines think?" was too imprecise.  His goal was to replace the
question with a different question that was closely related.  The 'Imitation Game', was a
first step in deciding if machines could exhibit humanlike behavior, not necessarily
thought.  For Gunderson, the Turing test becomes method of determining thought, which goes
beyond the scope that Turing assessed of his test.6  I do not believe Turing would grant a
machine thought if it merely passed his test.  He believed that once a computer is
indistinguishable from a human it must be thought of as 'thinking', yet his test is not a
sufficient criterion for this claim, merely a necessary one.  I will put aside this
objection to Gunderson's reformulation of Turing's statements and discuss the reservations
that he has about the Turing test. To illustrate the absurdity of the Turing test, Gunderson
reformulates Turing's original question from "Can machines think?" to "Can rocks imitate?"
He then follows the same steps as Turing did in the configuration of the 'Imitation Game',
except that instead of computers there are rocks and instead of asking questions the
interrogator's foot is stepped on.  The interrogator then must decide which entity, A or B,
is a rock and which is a human.  This new 'Rock test' is similar in structure to the Turing
test but it concludes with an absurd statement, "Rocks can think."  Gunderson suggests that
his facetious parody of the Turing test illustrates two points.  First, we cannot conclude
that rocks can imitate, but merely "that they are able to be rigged in such a way that they
could be substituted for a human being in a toe-stepping game"7  Second, is the idea that
thinking can not be measured by net results, such as those exhibited in the Turing test or
the 'Rock test'.  The Turing test is fundamentally incomplete in its design, since if the
principles are applied to rocks they yield the same results as humans. Gunderson continues
with the example of a door attendant and an electric door.  Both help other people exit a
building by opening the doors, yet we do not equate both with the attribute of thought.  The
net results are the same: I am assisted in exiting a building.  However, I have a
drastically different impression of the door attendant's capabilities from the electric
door's.  Similarly, the Turing test may produce the same net results, namely that both play
the game similarly, yet we do not attribute thought to all participants who are able to
produce the same results.  Gunderson suggests that the Turing test fails to account for this
difference and is, therefore, incomplete in judging whether an object being tested exhibits
thought. At first glance, I found that the analogy that Gunderson employs to illustrate his
point effective.  Nonetheless, upon a more scrupulous analysis substantial defects become
apparent in this seemingly parallel example.  Gunderson takes the liberty to ask "Can rocks
imitate?" but when designing his test he uses the concept of a 'rock-box'.  So although the
question deals with rocks, the game uses a more ambiguous 'rock-box' which, according to the
functions that it performs, consists of a mechanical device with electric eye sensors, and
some sort of self-propelling automata.  Even the label of 'rock-box' is misleading because
the mechanism that Gunderson is describing has to be significantly more complex than a mere
box full of rocks;  it has to contain a combination of fine-tuned mechanisms and series of
instructions that even today would prove hard to devise.  A similar inconsistency is not
apparent in Turing's 'Imitation Game'.  The question that Gunderson should be asking is "Can
a machine(rock-box) imitate?"  If this were the question, then I would be obliged to say
yes.  The 'rock-box', if it was designed in a sophisticated manner, could replicate the
motions of a human foot. Gunderson also interchanges, even equates, the words 'imitation'
and 'thought'.  There is a subtle difference between what Gunderson is asking in his
question and what Turing is asking in his.  Gunderson's question revolves around 'imitation'
while Turing's around 'thought'.  "The Imitation Game', by virtue of its name, can be
misleading.  Turing's final goal in deciding whether a machine can exhibit humanlike
behavior is whether it is indistinguishable from a human, not merely that it imitates one.
Gunderson's question deals with the imitation of a human foot, a deceptively weaker claim
than that of Turing.  Gunderson takes imitation and thought to be the same thing in the
Turing test.  This is not so for Turing who does not claim that imitation is a suitable
replacement for thought.8 Although Gunderson's example is partly designed to exaggerate the
absurdity of Turing's arguments, the basic tenets of what Gunderson's and Turing's tests are
supposed to be testing are drastically different.  The replication of a human foot motion is
almost peripheral to human thought, while the replication of human conversation and
interaction seems much more central to it.  The thought and level of complexity involved in
using the foot to stomp on someone is minuscule compared to what it takes to interact with a
human in an intelligible manner.  Thus, whether a machine can stomp in proper way is
irrelevant, or hardly relevant, to human thought, while the skills needed to pass the Turing
test are much more so. Gunderson continues his criticism of the Turing test by employing
Scriven's discussion of "the performatory problem".  Gunderson takes a certain portion of
Scriven's account in making his point.  He makes a distinction between "achieving a similar
result" and "doing the same thing".9  For Gunderson, and Scriven, these two are not the
same.  The Turing test is designed to judge the similarity of a certain result, namely
humanlike behavior.  This does not automatically imply that if a machine and human achieve
the same level of results, they are engaging in the same process in achieving this results.
Humans, in order to converse intelligently, have to think.  Computers, if they are able to
converse intelligently, might also be thinking but there is no way to verify this with the
Turing test.  Gunderson adapts Scriven's "performatory problem" to attack the necessity in
imputing that a machine has thought.  People and machines are clearly different in their
composition, yet it is possible for them to achieve same results in certain functions.  It
is not clear whether passing the Turing test is a clear indication that a machine is
engaging in the same processes that a human is.  While imitating results, a machine may not
be imitating method during the course of the Turing test.  Unfortunately, we do not know
what method is quintessential to human thought.  In not being able to take 'doing the same
thing' into account, the Turing test is inadequate to guarantee that a machine is exhibiting
thought. Gunderson continues his examination of the 'Imitation Game' by elucidating another
shortcoming of this test.  He aptly points out that human beings are judged in a greater
context than is outside the Turing test.  Gunderson claims that "we would only regard
it(Turing test) as one of the many examples we might give of Peterson's mental
capacities."10  The suggestion is that the Turing test is not representative of all that is
required for thought.  We see human beings in a greater context than just participating in a
certain test.  It is difficult, if not impossible, to disassociate all the other examples of
human thought when judging a human's performance in the Turing test.  Machines, on the other
hand, do not have the privilege that humans have in taking the test, namely being associated
with numerous other activities and examples that are deemed to be a result of thought.  For
Gunderson, the Turing test by itself creates an incomplete picture of human behavior.  The
'Imitation Game' has to be included within a greater context in order for us to associate
thought with a machine that has passed the test.  "Thinking, whatever positive
characterization or account is correct, is not something which any one example will explain
or decide."  It is Gunderson's argument that Turing's 'Imitation Game' is just once example
of thought.  By being one example, the test is an inadequate procedure for testing the
presence of thought. Due to the layout of the Turing test, I believe that Gunderson's claim
is premature.  The interaction between the interrogator and the correspondent is only
limited by their interface, a teletype machine.  Turing's goal was not to design questions
or techniques within the framework of the game.  Essentially, the interrogator may demand
anything from the correspondents, except anything that would require them to perform some
physical feat or demonstration.  Intellectually, the Turing test has no theoretical limits.
The interrogator may argue, question, observe and interact in all possible way with the
correspondents.  The imitation game might last an unlimited amount of time and the
interrogator may even demand that the correspondents demonstrate a sense of opinion,
judgment, reflection, deliberation, wonder, calculation and even emotion.  Granted, it is a
different situation interacting with a live person, but the limitations that the Turing test
imposes on the interaction between correspondent and interrogator are exaggerated by
Gunderson.  Turing designed his question and answer method in a versatile way: his test
"seems to be suitable for introducing almost any one of the fields of human endeavor that we
wish to include."11  Either Gunderson failed to realize this remark or he failed to properly
substantiate his claim.  He gives no examples, that are not physical demonstrations, that
can not be incorporated within the Turing test. Gunderson tries to clarify his criticisms of
Turing, through a Vacuum cleaner salesperson and homemaker example.  The scenario involves
the salesperson trying to persuade the homemaker into purchasing the 'all-purpose Swiss 600
vacuum' cleaner.  A misunderstanding occurs when the homemaker demands to see the features
of this vacuum cleaner.  The salesperson can only demonstrate one example (picking up dust)
and nothing else.  The homemaker seems disappointed, since she assumed that by 'all-purpose'
the vacuum cleaner could do a multitude of things, not just one.  Thus, how can the
salesperson call the Swiss 600 'all purpose', yet only give one example of it's 'all
purpose' capabilities?  The analogy that Gunderson tries to propound is that 'all-purpose'
and 'thinking' are terms that share certain features.12  Thinking, for Gunderson and most
other people, is not an example of just one thing.  I, for instance, would not be convinced
if something was able to play cards that it was thinking.  Furthermore, I still would not be
convinced if that something was able to utter sounds in a coherent way.  Gunderson suggests
that thinking has to be viewed in a holistic context, not merely by one activity.  It is not
enough to site one example and the immediately decide whether a thing 'thinks', or for that
matter is 'all-purpose'. The crucial criticism of Turing comes when Gunderson suggests that
the Turing test is just one example of thought.  This is parallel to the sucking of dust
being one example of the 'all-purpose' Swiss 600.  The 'Imitation Game', being one example
of a 'thinking' activity, fails to convince Gunderson that a machine can think simply
because it is one illustration of 'thinking'.  One example can not represent or justify a
word such as 'all-purpose' or 'thinking'.  Therefore, both the homemaker and Gunderson are
unconvinced of the Swiss 600's and computer's capabilities.  Trusting one example(the Turing
test) is not enough to convince Gunderson that machines can be capable of thought: It is
because thinking cannot be identified with what can be shown by any one example or type of
example; thus Turing's approach to the question "Can a machine think?" via the imitation
game is less than convincing... Turing, like the vacuum salesman, has trouble making his
sale.13 Yet, Gunderson does not clearly state how many examples would be needed to satisfy
his apprehension about admitting that a machine can think.  He does not provide criteria for
deciding if a machine thinks.  Does the object have to be composed of cells, like a human
being or does it merely need to exhibit all the same behavior as one?  Gunderson claims that
'thinking' is inherent in our conception of a human, but he does not state what is inherent
in our labeling something as 'thinking'.14 We may question Gunderson's analogy by asking "Is
the Turing test solely one example of 'thinking'?"  I believe Turing would argue that the
his test is not a strictly defined example such as swimming, playing cards or an activity
with clearly particular and stringent boundaries.  Gunderson fails to consider the
'Imitation Game' as a dynamic framework that can not be categorized within the strict
boundaries of one example.  Instead of being an example of 'thinking', the Turing test may
be envisioned as a set or collection of examples of 'thinking'.  Whether this set of
examples is sufficient to justify thought remains to be resolved.  Gunderson, if he agreed
with this view, would probably disagree that the test is enough.  What is more important, is
that Gunderson hastily classifies the Turing test as a mere example of 'thinking' rather
than a framework that, to a large degree, exemplifies 'thinking'. I may be a thousand miles
away from my friend, but I am still able to communicate with her in a very similar fashion
than if she was right next to me.  I am able to sense her emotion and judge her intellectual
capabilities through a medium that does not involve physical or immediate contact.  In a
similar sense, the Turing test's restrictions are only limited in the way I correspond with
my distant friend.  Whether this is enough of a limitation to prevent me from judging my
friend or computer as thinking is the essential question.  Gunderson takes the 'Imitation
Game' as one example of a multitude of examples that would identify a human as thinking.
His notion of the test is so restrictive that it prevents the interrogator from judging the
correspondents' 'thinking' capabilities.  I would contend that such a conception of the
Turing test is inaccurate.  The test, instead of being an example, is a system that helps us
determine if the object being tested exhibits thought.  Therefore, many different examples
of human actions that exhibit thought can be incorporated within the test instead simply
one.  The restrictions of the test are very similar to the restrictions that humans have
when interacting with each other. The two views of the Turing test, Gunderson's one example
of 'thinking' and Turing's comprehensive method for determining if a machine can exhibit
humanlike behavior, stem from different perceptions of thought.  Turing sought to isolate
the factors that he considered sufficient to exhibit thought and incorporated them in his
test.  Gunderson's view of the thought is that it is inseparable from the other activities
that humans perform.  For Gunderson, when these activities are conglomerated they constitute
'thinking'.  Even if a Turing machine was more than just an example of 'thinking', maybe
even several examples, Gunderson still believes that it still fails to satisfy our intuitive
concept of 'thinking'.15  Gunderson poses the question of 'how many examples would satisfy
our criteria of thought?"  He does not answer this question.  Nonetheless, Gunderson,
without defining the range of examples that would satisfy this question claims that Turing's
machine is not within these bounds.  I find this hypothesis incomplete and contradictory,
since a limit has to be established before one judges an object to be within or beyond that
limit.  Gunderson is judging the Turing test without providing definite criterion for an
object to be 'thinking'. Gunderson's last critical remark of the Turing test has not much to
do with the 'Imitation Game' as it does with the replication of human behavior.  Primarily,
Gunderson claims that much of what humans do requires a limited amount of thought.  For
example, a person adding and subtracting numbers can think about something else while he or
she is engaged in this activity.  So if a machine can replicate behavior, such a counting
and subtracting, it does not necessarily mean that it thinks.  Tasks, such as inventing
original solutions or comprehending a situation from different perspectives and levels, are
much better criteria for thought.  Still, we do not have a definite qualitative scale for
thoughts so we can not be completely positive what tasks are indicative of human thought.
At what level of complexity is an action derived from a thought?  Gunderson does not attempt
to answer this question but neither does Turing. There exist two general views of the Turing
test.  Gunderson's very exclusive view of the test's capabilities: The test is a mere
example of 'thinking' and has to substantiated with other examples to illustrate the wide
spectrum that we consider to be 'thinking'.  On the other hand, there is my assessment, and
I believe Turing's assessment also, of the test as a more inclusive tool that can be adapted
to include every constitutive element of thought.   The parameters of the test are defined
as such:  "The object of the game for the interrogator is to determine which of the other
two is the man and which is machine"16  I believe that these parameters are essential and
sufficient in judging an object as having thought.  Our only means of judging whether
something is 'thinking' is by judging whether it imitates us, 'thinking' beings, in its
intellectual behavior. A more general question arises in how I am able to determine if other
people are thinking?  A large part in attributing though to an object is done automatically
and without a series of tests.  When I meet another person I do not ask myself, "Is that
person thinking?"  My mind is preconditioned so that when I meet another human I expect the
same behavior as I exhibit.  Thus, it is enough for me to even glance at another human being
and I immediately assume that human is thinking.  I would not interact with them, observe
them and  then according to my judgment deem them as thinking or not.  By being repeatedly
exposed to humans that think, I automatically induce that all humans also think.  Machines,
especially computers, do not have the luxury of having this opinion thought about them by
humans. An even more general question arises of whether behavior is enough to prove
'thought'.  Or, is imitation, even if this imitation was perfect in all means, enough to
prove thought?  Although Gunderson's concern is whether the Turing test is able to fully
incorporate all the essential behavior of though, we might not consider this enough to
guarantee thought.  Turing's view on the matter is this: A reason for believing in the
possibility of making thinking machinery is that it is possible to make machinery to imitate
any small part of a man.  The microphone does this for the ear, and the camera for the eye.
One way of setting about our task of building a thinking machine would be to take person as
a whole and to try to replace the parts of him by machinery.17 Turing, in this statement,
takes a materialist view of the situation.  If an android was to be constructed identical,
part for part,  to a human, then it would have to exhibit thought.  This is in contrast to a
dualist approach that would not ensure that physically identical objects would both have
minds.  It is possible that thoughts, derivatives of the mind, can only be granted by a
supreme being or by an act that is beyond the material.  So, even if a computer was
identical in all functions with a human, this would not be sufficient to satisfy the dualist
that this machine is 'thinking'.
 Even if we accept a materialist view, we might question whether identical behavior by two
 objects, with two different mechanisms for attaining that behavior, both can be thought of
 as thinking.  If we built an android that would be indistinguishable from a human in terms
 of its behavior yet would not contain a brain but a different mechanism for realizing this
 behavior, would we deem it as 'thinking'?  Gunderson addresses this discrepancy between
 'behavior' and 'method of materializing this behavior'.  Turing's statements about
 identical behavior among a machine and a human justifying thought in the machine are
 vulnerable to attack here.  He takes it for granted that this identity is not automatically
 granted and that it could be the case than only a certain procedure produces thought.
 Similar behavior may not be the result of thought in all cases, as Gunderson tries to
 establish.  It seems that Turing is taking a neo-physicalist account of thought.  The
 phenomena of thought can only be described in spatiotemporal terms and consequently any
 notion of thought can in principle be reduced to an empirically verifiable physical
 statement.  If such statements are identical, though not necessarily manifested in the same
 fashion, then thoughts occurs.  As long as behavior is synonymous with a certain reality,
 namely thought, then the mechanism or technique that is required to produce this behavior
 (or thought) can be varied.
The parameters that the Turing test provides are only limited by the limits of human
interaction.18  If the limits of human interaction are sufficient to determine if an object
has thought, the I believe that the Turing test is a fair method of determining if machines
think.  However, if our knowledge of what constitutes 'thinking' is incomplete or cannot be
determined merely through human interaction with a machine or human then the Turing test
comes up short in its assessment of a machine's capability of thought.  The contentious
issue for Gunderson is whether the parameters of the Turing test allow a demonstration
sufficient enough to impute thought.  Gunderson believes that this is not so.
 I, on the other hand, take a more liberal view of the Turing test.  The parameters that
 Turing defined in the 'Imitation Game' can allow for games, learning of languages,
 translation of languages, cryptography, mathematics, deduction, demonstration of original
 thought, emotion and all the facets of what we term 'thinking'.  The interrogator has no
 boundaries in the questions he or she might ask.  Computers have had very limited success
 in this test and after a rigorous examination few have fared well even after a short time.
 I believe that this either shows that the Turing test is much more sophisticated than
 Gunderson claims or that artificial intelligence is in a very primitive state.  Only the
 development and further testing of computers through the Turing test can provide answers to
 whether a machine can be termed as 'thinking'. It is also important to realize Turing's
 original purpose for creating the 'Imitation game'.  It was not an answer to his original
 question of "Can machines think?".  He assessed that this question was beyond his
 capabilities as a scientist to answer, since it required an ontological definition of
 'thought'.  The goal of the Turing machine was to test how well a machine was compared to a
 human in terms of its behavior.  This is a different question from the original, albeit
 very similar to it.  Turing's later claim, that if machines were indistinguishable from
 humans, not just by means of the 'Imitation Game', we would have to impute that they think,
 is much more contentious.  Gunderson picks up on this ambiguity and challenges the Turing
 test as valid way of imputing thought in a machine.  Although I do not believe that this
 was part of Turing scheme, I have delineated Gunderson's objection to the Turing test and
 further commented on them.  Finally, I went beyond Gunderson's essay further examined the
 relevance of thought to Turing's 'Imitation Game'.
1  This essay was published in the MIND: A Quarterly Review of Psychology and Philosophy,
Vol. LIX, No. 236, October 1950. 2  The original method involved the distinction of  man and
a woman, but the goal of determining whether a machine can imitate a human being  is the
same. 3 Hodges, Andrew. The Enigma of Intelligence: Alan Turing.  Unwin;1985. pg. 415. 4
This essay was published in Minds and Machines edited by Alan Ross Anderson. 5 Anderson,
Alan Ross. Mind and Machines, pg. 62 6 Ince, D.C. Collected Works of A.M. Turing.
North-Holland;1992. Pg. 133 7 Anderson, pg. 63 8 Ince, pg.136 9 Anderson, pg. 65 10 Ibid.
pg. 66 11 Ince, pg. 135 12 Anderson, pg.69 13 Ibid. pp. 68-69 14 Ibid. pg.68 15 Anderson,
pg. 70 16 Ince.  pg.133 17 Hodges, pg.116 18 The Turing test provides unlimited mental
interaction.  As mentioned earlier, it does not allow for physical demonstrations.  I do not
believe that physical demonstrations are needed for thought to be demonstrated.  Gunderson
does not cite any examples where this is necessary either.  If thought requires a
manifestation through a human body, rather than an android one, then the Turing test is
inadequate.  I have taken note of this objection but it is not substantiated by Gunderson in
his article. 2