Re: The Demise of Computationalism?

Don Geddis wrote:
Stephen Harris <cyberguard-1048@xxxxxxxxx> wrote on Fri, 16 Mar 2007:

You, on the other hand, seem to be familiar (or at least can search for)
a great deal of literature. Yet somehow you seem to understand very little
of it.


Well, we will see. You have said you know quite a bit about Turing Machines, and I think you have made some mistakes about them. I hope
somebody like Rickert comments, because although he is a terse grouch
he knows quite a bit about this stuff.


Computationalism asserts that the right program will instantiate
a mind. There is no mention of effectors or sensors. Just the right
program. What else can that program be but some type of Turing
machine?

OK, sure.

Although one quibble is that, in the real world, computational complexity
matters, and that depends on architecture. If some "right program"
instantiates a mind, you are correct that there is some Turing Machine that
would, in principle, perform the same computation (assuming that nothing is
time critical, and you had infinite storage). But that doesn't mean that
anyone would actually implement the thing as a Turing Machine. In the real
world, time does matter, and resources are limited. Turing Machines are NOT
equivalent to other architectures under those conditions.


Yes, time and memory are missing from a physical realization of a TM.
But are those important issues? We are talking about a program which
passes the Turing Test (in the future). It must have sufficient time
and memory capability or it can't do that. Also the program could be
written in any formal language, C++ or Lisp for instance and it must
have a Turing Machine equivalence. I'm not talking about implementing
some program as a Turing Machine, I'm speaking of the logical purpose and functionality similar to this Computationalism slogan.

Cognition is the computation of Turing-computable functions.

Comp is asserting that the mind is computing Turing computable
functions. So is there something wrong with saying that a TM
computes turing-computable functions? Well each TM can be preset
to compute some particular functions for some purpose. According
to Comp, each human thought is a realization of some TM.

Is this right so far? Now the architecture is said not to matter.
The human mind and a computer program can both function (alism)
so as to produce very many of the same turing-computable functions
which can also be called Turing machines. The computer could do
more in this scenario as some TMs would require more memory than
a human can retain. And the mind thinks it thoughts in real time
with a limited physical memory, so it is not an architecture
which exceeds the potential of a future TT passing program. Thus
my quibble is that your quibble doesn't impact Computationalism,
due to functionalism handling architecture and the ideal resources
of a TM aren't required, just the principle of computable logical calculi. This could probably be a bit more precise but it seems
like it should be understandable-- I can do quotes if necessary.
Maybe I should have said a series of computable functions = TM.


There's also Neil's concern that a formal Turing Machine has all its input
set before it begins computation. There's no place for real-world sensors in
the formal Turing Machine model. Whereas we expect a consciousness to be
interacting with some external world.

Nor in Computationalism with Turing-computable functions without
bringing in the formal TM model; I can't see an unbounded tape
as necessary to pass a TT. I was using TM as a name for a logical
structure, and its common to do that, use TMs instead of describing
something in more detail in discussing Comp. Some of the slogans
are in terms of TMs and mind, not actually about implementing TMs.



My focus is the Computationalism description that 'the right program instantiates a mind'. Now I also happen to think that the right program
will need sensors (at least) but Comp didn't provide for that. Which
is why I think the reasoning to support Comp is both incomplete and
not particularly coherent.

But now the focus is on Comp and mind. Because of a reason much like
Neil's or maybe it is the same, I don't think a Turing machine executes
a program. So the 'right program' can't be an ordinary TM because they
don't execute programs.

But leaving those two problems aside, sure it's some kind of Turing Machine.

I asked you whether that program would be a Turing machine or a universal
turing machine. You responded 'that you knew a lot about turing machines'
but that my question didn't make any sense. Why not?

A Universal Turing Machine is just one kind of Turing Machine out of many.
There's nothing particularly special about it, in terms of how the basic
operations work. They're the same operations that any other Turing Machine
uses.


I don't think so. There is a major difference, a UTM can simulate all
other TMs. But a TM does not simulate others TMs. And this is relevant
to what, if any, kind of TM, in this case a UTM can execute code and
thereby meet the Comp requirement of 'right program'. The explanation
of this is technical enough to require a quote IMO.

Corey: "Gualtiero, ...you claim that Universal Turing Machines execute programs, and I am not so sure."

gualtiero piccinini replies: "Corey, you raise a very interesting question. Two preliminary points: (1) the question of what does and
does not count as program execution is, to some extent, a matter of stipulation (although we should strive to make good stipulations; e.g., stipulations that respect relevant scientific practices); (2) I agree that UTMs simulate ordinary (non-universal) TMs. It remains to be seen whether (it’s a good idea to say that) UTMs simulate ordinary TMs by executing programs.

I think you make an important conceptual point: there are significant differences in the way UTMs produce their output vis a vis the way ordinary digital computers do. But there are also important similarities: both compute by responding to explicit instructions, and they compute different things depending on what instructions they have. I see this as good enough reason to subsume both under the rubric of program execution. Nevertheless, a detailed theory of computing mechanisms should identify the differences between the two mechanisms and account for them. It would be interesting to read a detailed comparison and contrast of the two mechanisms (regardless of how we
call them).

There is also a historical reason for labeling both UTMs and ordinary computers as program executors. Turing and von Neumann were among the designers of the first modern (electronic) digital computers (i.e., computers responding to programs that are executed so as to make the machine universal up to its limitations of time and memory). Both Turing and von Neumann were trying to design physical machines that would approximate the computational universality of UTMs. And UTMs are universal because they respond to explicit instructions written on their tape. So using the concept of program execution for both the processes of UTMs and those of digital computers is a way to mark the impact that UTMs had on the origin of modern digital computers. (As I point out elsewhere, though, this impact should not be overestimated either.) It would be interesting to know at what point the term ‘program execution’ became commonly used for digital computers, and whether and when computer scientists started using it to describe UTMs."

They were discussing this quote from a Piccinini,

"Some Turing machines can compute only one function. Other Turing machines, called universal, can compute any computable functions.
The difference between universal and non-universal machines has a mechanistic explanation. Non-universal Turing machines manipulate the digits on their tape in accordance with their machine table, without executing any program. Universal Turing machines, by contrast, have such a special machine table of their own that they treat some of the digits on their tape as programs and others as data, so as to manipulate their data by appropriately responding to the programs. Because of this, universal Turing machines—unlike non-universal ones—may be said to execute the programs written on their tape. The behavior of all Turing machines is explained by the computations they perform on their data, but only the behavior of universal Turing machines is explained by invoking the execution of programs."


A UTM is more an application area, the way you might carve off a subset of
all Turing Machines that do sorting (including quicksort, or heapsort, or
bubblesort), and think of them as a class of TMs with some properties.

The class of UTMs are just regular old Turing Machines that happen to have
very interesting behavior. As input, they take a _description_ of some other
_random_ Turing Machine, and then emulate its behavior, as though the
subsequent input were processed by the Turing Machine that they were given
a description of. Kind of a "meta-" thing, a Turing Machine that processes
Turing Machines.


I left this in to ask a question: Why _random_? Why not selected?

So. You're asking whether a program that passes the Turing Test would be a
"regular" Turing Machine, or a "universal" Turing Machine (which is actually
a subset of the "regular" kind).

The first answer is that, a program that passes the Turing Test probably
wouldn't be a Turing Machine at all. Of any kind.


OK, but I'm not a computationalist. Computationalism says
Cognition is the computation of Turing-computable functions. Then,
"Other Turing machines, called universal, can compute any computable functions."

I'm using Turing Machine as a designation for computing a certain type
of computable function or a UTM for computing any computable functions.
I'm not sure what you mean by Turing Machine, maybe something physical.
My usage and the usage of Comp is for each function, a logical entity,
and that the thoughts of the mind have an equivalence to this entity/TM.

But if you laboriously constructed a Turing Machine which had the same
computations (if not with the same speed) as the TT-passing program, then
it would surely be a "regular" Turing Machine. If for no other reason than
its behavior is to take in English questions and produce English answers.
While a "universal" Turing Machine has the behavior of taking in a description
of another Turing Machine and emulating that TM's behavior.

Right, and simulating that TM's behavior produces the same output
as the regular Turing Machine doing English question and answer.
The question is, if there are countably infinite valid Turing Tests,
whether a regular TM has the flexibility to answer them all correctly,
or would it require integrating the TMs at the disposal of a UTM?



Yes, it appears you mean some type of physical realization.
Computationalism and I are talking in terms of an abstract
logical entity which has a function on either a human mind
(according to Comp) or a computer. For some C++ program
there is a TM equivalence because they both are computable.
This is not a physical or even tape moving idea required.

Since what a UTM does is different than what a TTPP does, the answer to your
question is no, a TTPP would not be a UTM, it would be a regular Turing
Machine.


Can you explain why a regular TM which you state would be passing
the TT, produces a different output, given the same input, than
a UTM which simulates the regular TM mentioned above? I think given
the same input, the regular TM and the UTM simulating that same
regular TM, both produce the same output. It seems to me there is
a leak in your bucket of understanding Turing machine theory.

It is true there is a difference in how the regular TM and the
UTM realize their output, but the output is still the same, and
your reasoning for the selection basis appears to be based on a difference in the output.

But it isn't a very interesting question. The fact that you thought it was
interesting suggests that you don't understand something.


Actually, that remarks seems a bit condescending. I find the question
interesting because it helps determine the plausibility of Comp.

It does appear that one of us has a gap in their understanding, it
seems to me you don't understand that given the same input, both a regular TM and a UTM simulating that regular TM, produce the same output. I get that impression from your answer above.

I think you might reply because a program that passes the turing test will
have a lot of other stuff. But that isn't part of the Computationalism
thesis. You say you believe in Computationalism but you have made
statements that contradict the tenets of Comp.

What have I said that contradicts the tenets of Computationalism?


After rereading this post, your thinking seems to hinge around this
idea, at least it explains why we arrived at different conclusions:

> The class of UTMs are just regular old Turing Machines that happen to > have very interesting behavior. As input, they take a _description_
> of some other
> _random_ Turing Machine, and then emulate its behavior, as though the
> subsequent input were processed by the Turing Machine that they were
> given a description of.

I don't think it is "random". I think this is programmable for a UTM.
Hmmm, I notice you emphasized _random_ like you are certain of it.
I'll be looking that up shortly, as it is quite significant. Otherwise
I agree with your explanation about this.
.

Relevant Pages
Loading