Definition by inessentials
Greg Swann
thank goodness, but Turing has left us an enduring
legacy notwithstanding. It consists of contriving
a specious and inanely limited definition for
something, then claiming that that thing has
replicated when the litmus-test of the retarded
definition is achieved.
* If intelligence is defined as fooling people,
then software that can fool people into believing
that it is intelligent is therefore intelligent.
* If life is defined as anti-entropic chemisty,
then chemicals artificially sustained in a
temporarily anti-entropic state are therefore alive.
I've had a lot to say here over the years about
flawed or chimerical definitions. A day or two ago
someone equated free-as-in-freedom-from-coercion
with free-as-in-moral-free-agency, a classic
Randian/Marxian definition swap. But the Turing
method is quite a bit worse, since it creates
entire new academic disciplines of error.
Here is a quote from a paper I read about
Artificial Life (a branch of software
engineering):
I argue that the ability of an artificial
organism to interact seamlessly with existing
biological organisms is an important criterion
of its success as an A-Life model.
It's the Turing Test of the Terrarium! If the
other creatures don't gape and point, then my
little Baby Frankenstein is ALIVE!
Mind what goes into you mind...
Greg Swann
David Tomlin
> No one talks much about the Turing Test anymore,
> thank goodness, but Turing has left us an enduring
> legacy notwithstanding. It consists of contriving
> a specious and inanely limited definition for
> something, then claiming that that thing has
> replicated when the litmus-test of the retarded
> definition is achieved.
I don't think this is fair to Turing, or to the many people
(including myself) who find the Turing Test interesting.
Turing wasn't laying down a dogmatic "definition." He was
taking a first stab at a very difficult problem.
Dave O'Hearn
> No one talks much about the Turing Test anymore,
> thank goodness, but Turing has left us an enduring
> legacy notwithstanding. It consists of contriving
> a specious and inanely limited definition for
> something, then claiming that that thing has
> replicated when the litmus-test of the retarded
> definition is achieved.
>
> * If intelligence is defined as fooling people,
> then software that can fool people into believing
> that it is intelligent is therefore intelligent.
This is not very charitable. The nicest way you can describe the
Turing Test is that it is based on "input/output equivalance". That
is, if a machine is capable of generating the same outputs (sending
you text) when given the same inputs (you send it text), as a human
would, then we could guess that it is intelligent.
This is ok as far as it goes, but if you take it fundamentally, you
have to use a form of indirect realism to defend it. Some AI theories
do just this; they call it the "systems theory" of consciousness. I
don't think there are any good arguments for the systems theory, other
than the argument from authority that it makes a fundamental defense
of the Turing Test, and there is so much literature on the Turing Test
that we need a good defense of it.
Turing wouldn't've liked the systems theory either. He did not believe
a simple "lookup table" would be conscious or intelligent... but the
systems theory says it would be.
> * If life is defined as anti-entropic chemisty,
> then chemicals artificially sustained in a
> temporarily anti-entropic state are therefore alive.
It was an attempt at a more abstract definition of life. The one you
will find in most biology textbooks contains as one of its
"essentials" that all living things are composed of cells. That is
certainly a good-enough definition for all known living things, but it
would be nice to have a more abstract one, no?
> I've had a lot to say here over the years about
> flawed or chimerical definitions. A day or two ago
> someone equated free-as-in-freedom-from-coercion
> with free-as-in-moral-free-agency, a classic
> Randian/Marxian definition swap. But the Turing
> method is quite a bit worse, since it creates
> entire new academic disciplines of error.
>
> Here is a quote from a paper I read about
> Artificial Life (a branch of software
> engineering):
>
> I argue that the ability of an artificial
> organism to interact seamlessly with existing
> biological organisms is an important criterion
> of its success as an A-Life model.
>
> It's the Turing Test of the Terrarium! If the
> other creatures don't gape and point, then my
> little Baby Frankenstein is ALIVE!
This is not very original stuff; it is ripped right out of sociology
and phenomenology. They define truth as "consensus". So sure, some AI
researchers might import bad ideas from other places into AI. But it
is nothing wrong with AI per se, they are just ripping off bad
theories from elsewhere. Indirect realism is very popular stuff, and
any area of study is going to have some people dragging in indirect
realism to make their arguments or defend their definitions.
--
Dave O'Hearn