[Peter Cariani, 960212, 1300]
I've been having some difficulty with the listserver,
so I may have missed some messages in the last few
days. If you reply to this post, could you also cc:
a copy to me directly: peter@epl.meei.harvard.edu ?
This is a response to Martin Taylor's message
[Martin Taylor 960209 13:40].
I agree with Martin,
that we're largely talking at cross-purposes.
It's true that the Turing Test specifies one
communications channel between tester, a typewriter
keyboard which is operationally provides
a string of symbols. The Turing machine
behind the curtain however, has no other
source of outside information available to it, save
what it gets through that keyboard or what has
already been programmed into the device. In this respect
the Turing machine is different from biological
organisms and robots, which have independent means
of gaining (new) information about the outside world.
Martin says:
If I can look out of
the window and have good reason to believe that I am
seeing what you are
seeing, then I have an extra kind of communication
channel with you other
than the typewriter. That's not allowed.
A shared material environment is not necessarily an extra kind
of communication channel. An event that both testor and testee
perceive does not communicate information unless one of the two
somehow caused the event to happen.
But even shared material environment were an extra channel,
why shouldn't such interactions be allowed in the test?
Martin:
The human is restricted to whatever
inputs and outputs are available to the tested entity. If that were
not so, the test would be inherently unfair.
Cariani:
I don't follow this logic at all. What does "fairness" have to do with it?
If you truncate the problem in this way, the resulting test can become
tautological. We could limit input and output to LISP statements, and it
would be hard if not impossible to distinguish the human from the Turing
machine, but the test itself would be rendered totally meaningless.
Martin:
* Why would it be hard? Because humans have a difficult time reading LISP
* statements? The feedback loop remains the same. One would look for evidence
* of control on the other side, and the higher the level of perception at
* which one probed, the more stringent the test. What does it matter what
* kind of symbol strings are allowed in the test, if their syntax allows
* an indefinite number of them?
···
*
* Fairness has to do with not requiring the machine to be a blue-eyed
* blond, with squishy pink-skinned flesh. The Turing test refers to its
* thought processes. If these attributes are _in fact_ required before an
* entity's thought processes can be humanlike, then _in fact_ no machine
* without them will pass the test. But the test does not, and should not,
* require them _a priori_.
My point about LISP statements was that if all there is
to the communication is syntactic structure (exchanging
'semantic-less' arithmetic expressions might have been a better example),
then a computer can carry on as well as a person.
In this case the task has been so limited
that there ceases to be a meaningful difference between how a computer
follows formal rules from how a human does it.
The point about "squishy pink-skinned flesh"
(as is carbon vs. silicon distinction cited elsewhere) is a red herring.
Being able to see the device/person behind the curtain is very different from
assessing whether it has independent access to the world (i.e. some kind
of sensor of its own in addition to the keyboard).
I have said nothing that would imply that such attributes
of appearance are needed to mimic human behavior
(I do believe that a robotic device could pass the kind
of expanded test that I have in mind).
* You are moving into the range of Stevan Harnad's "Total Turing Test" here.
* But think of what you are saying--that a human, placed in a soundproof
* room dark except for the screen with the tester's writing, would no longer
* be human and would be indistinguishable from a machine.
Harnad may have invented the term "Total Turing Test", but I think such
tests were contemplated long before him. It's a misleading name
for such a test because it emphasizes the "imitation game" parts of
the Turing test over the fundamental question posed by the test
(what are the limits of what Turing machines can do?).
I do agree that the human in sensory deprivation chamber
would be in the same position vis-a-vis perception
of the external world as the Turing machine. The
human could not then answer questions that depended
on sensing the world and reporting the result.
Here the functional capabilities of the human
in the tank are in fact qualitatively different
from those of the human who is informationally in contact
with the outside world. As in the case of perceiving human vs Turing
machine, I think there are analogous lines of questioning that
can distinguish between the perceiving human and the sensory-deprived human.
(Sensory deprivation doesn't make the person less 'human', but it does
change what s/he is capable of doing.)
I, Cariani, had said:
I think that there are lines of questioning (e.g. empirical questions that
could be verifiable by a (cooperating) human but not a Turing machine)
that would invariably expose the difference between a machine without
inputs (save those from the tester) and a human being.
Martin replied:
* Now again you are asserting an answer to the question Turing posed. You
* say that such questions exist, and I think that you should archive them
* so that if and when a plausible candidate machine passes other attempts
* to test it, your lines of questioning should then be used to unmask the
* pretender.
Maybe I'll do that, but you've categorically disallowed any questions
that require empirical inquiry. That's what I call a rigged test.
Every time I've seen mention of a Turing test being actually performed,
there are heavy restrictions on what kinds of questions can be asked.
I will, for now, resist parallels with public demonstrations by psychics
and parapsychological 'wizards.'
I, Cariani had said:
As you (Martin) say,
"'Pure syntax' has very little to do with language in use." but
this is what is going on on the Turing machine side of the dialog.
Martin:
* The difference between the Turing machine and the human in this respect is
* that you know how the Turing machine works and can make that assertion (using
* a particular notion of "syntax" to mean "logical operations"). You don't
* know how the human works, and therefore you assert that the human doesn't
* use logical operations (syntax). Since the point of the test is to determine
* whether such a distinction is generic, it is improper of you to assert
* that it is before running the test. Your comment is like the following
* "syllogism": A is an X. I don't know whether B is an X. Therefore B is not
* of the same kind as A.
I'm not making any assumptions here about whether or how a human uses logic
operations, and I think such assumptions are irrelevant
to this particular point. No such syllogisms is being made, even implicitly.
(the query was based on what you said re: 'syntax' and 'language').
We could argue about whether the Turing machine's actions
are purely syntactical/logic-based or not. This would get
into issues of what "semantic" operations mean (as opposed to
"syntactic" or logical operations) in the context of a computer program.
But it's a longer discussion/debate that we could take up some other time.
Finally, regarding issues of feedback/communication in the Turing test:
Cariani:
How does the Turing machine, "the other partner", "have a reference
state that the originator should come to be satisfied with the recipient's
state"?
Martin:
* In most computers, that's built in by the designer. If one were to design
* a program with the intention of passing the Turing test, one would surely
* have to build in the possibility that it had such a reference perception
* and an analysis of the state of the tester to compare it with.
Cariani:
I don't see any perception of the state of the testor happening
I still don't see how the machine (successfully)
determines how well it is doing just from
the line of questioning being asked by the testor. Sure, you can put in
a module that tries to guess from what the testor is asking what responses
might fool him/her, but there doesn't necessarily have to be any usable
information in the questions asked. The Turing test is
a very artificial situation, very much unlike a normal conversation in which
both parties are freely conveying (through various means) information of
pleasure/displeasure, agreement/disagreement, that allows the error fn's
to be built up and to operate properly. As constructed, the Turing Test
encourages the testor (interrogator) to minimize feedback to the testee,
and encourages the testee to evade wherever possible.
In general, I think the framework
of the Turing Test is really badly suited for dealing with interactivity
of the kind that you want. And behavioristic, imitation criteria lead
in the wrong directions if one is interested in underlying
functional organization rather than superficial appearances.
Martin:
To relate computability to intelligence is an enormous mistake, far more
damaging to cognitive science than the damage you impute Turing as
having done to AI.
Cariani:
I'm not sure if I understand this.
I definitely agree that dragging 'computability' into cogsci and AI was
a big, big mistake, but I wasn't sure what Martin's reasons were
for saying this. I imagine they are very different from mine.
We should probably let those differences rest, for now at least.
I certainly agree with Martin that
time is of the essence, but computability is not.
Peter Cariani
peter@epl.meei.harvard.edu