Re: What's the common criticism on Turing Test?

Peter Chan <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> I'm new to A.I. and currently reading on the Turing Test.
> After some reading, I knew only the Chinese Room criticism.
>
> Can I know what's the others common criticism on the Turing Test?



Peter Chan <[EMAIL PROTECTED]> wrote in message
news:<[EMAIL PROTECTED]>...
> I'm new to A.I. and currently reading on the Turing Test.
> After some reading, I knew only the Chinese Room criticism.
>
> Can I know what's the others common criticism on the Turing Test?

I don't know about common criticism, but I can tell you some of my
criticisms,
for what they are worth.

First of all, for a really intelligent machine to pass the test
it has to either a) lie or b) be deluded about its true nature.
Now ability to harbour a delusion is probably not a good measure
of intelligence; and the use of ability to lie as a measure of
intelligence is a slight to all those people for whom lying is
anathema. Would you say that habitual truth-tellers are all
unintelligent?

Second, it is very easy for an unintelligent machine to seem
intelligent if there are strict limitations on the communications
channel. For instance, a gramophone playing a record (unintelligent,
though the contents of the record are a product of intelligence)
is indistinguishable from live speech of you're only allowed to
listen, not interject. The Turing test seems to overcome that
problem by allowing free two-way communication, but in reality
all communications channels are limited, and some are more
limited than others. A babbling "stream of consciousness" program
could overwhelm a low-bandwidth channel such as an old teletype,
putting the human judge on the defensive, and forcing him to answer
lots of questions instead of asking them. It makes a difference if
you are allowed to interject in mid sentence or if you have to
wait for a response before you are allowed to type. The test can
be skewed in favour of the machine by using the smallest allowable
bandwidth, imposing strict rules about who is allowed to type when,
and making the program exploit the rules as much as possible.
In these circumstances a program will occasionally fool people,
which would have no chance if there really was free communication.

I can think of other objections off the top of my head, but
that's probably enough for now. Most of my objections centre
around the notion that the machine in the test is trying to fool
people, and the test doesn't distinguish between real intelligence
and a successful pretence. Which is probably entirely deliberate.
But which would you prefer to be operated on by, a real surgeon,
or somebody really good at pretending to be one?

The philosophical basis behind the test is a form of pragmatism,
the notion that if a pretence works well enough then it is as
good as the real thing, or (in the case of intelligence, especially
if we have an operational definition of intelligence) it is the
real thing. But is it?

Thoughts?

Martin