My position statement is: If in a sense a laptop computer is a Turing
machine, then in the same sense a robot is also a Turing machine.
I think that most people missing the point here . . . . Analog can
always be converted to digital of a specified granularity, simultaneous can
always be sequenced, and *anything* can be recorded on a large enough and
detailed enough paper tape (and *any* behavior that is systematic/non-random
can be reproduced by a state machine with enough states). The world may not
be predictable (particularly because of quantum effects) but all of our
current computing artifacts are.
A robot's input paper tape is simply generated at a *much* faster clip
than a laptop's. They are both Turing machines. However, once a Turing
machine is complicated enough (and laptops and robots are *way* past this
point), it is no more useful to treat them as Turing machines as it is to
treat ordinary objects as a collection of quantum waves or the sea or the
weather as a collection of molecules.
I would also argue that the concepts of "disembodiedness" and
"groundedness" are most frequently mis-applied and inaccurately used when
applied correctly. Something is "embodied" and "grounded" to the extent
that it has the I/O to sense and influence *it's* world. An AI that is
intended to live solely in the digital world can be just as grounded in it's
world as a robot or a human is in the physical world. The difference in
environments will cause differences in mind-set, analogies, and expectations
which will lead to difficulties in communication (or attempting to deal in
another entity's world without practice and training) but it is entirely an
anthropic conceit to say that a laptop is necessarily disembodied
(particularly as you can control more and more physical things from an
Internet connection).
----- Original Message -----
From: "Pei Wang" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Monday, October 02, 2006 2:22 PM
Subject: Re: [agi] Is a robot a Turing Machine?
Richard,
The discussion is becoming more complicated than I expected. ;-)
There are three related, but different issues:
(1) For AGI research, whether it is always a good idea to treat a
complicated AI system as a Turing machine.
My answer is "No", and the argument is given in
http://nars.wang.googlepages.com/wang.computation.pdf . However, it is
not what I want to discuss here.
(2) For AGI research, whether it is always a good idea to treat a
physical entity or event, like your cappuccino or a hurricane, as a
Turing machine.
My answer is also "No", but it is a different issue, which I would
rather not discuss in the near future.
(3) For AGI research, whether there is a qualitative difference
between a conventional computer and a robot.
This is the issue I plan to address in a new paper I'm working on.
Some people (such as Brooks and Harnad) argue that the sensorimotor
mechanism of a robot is necessary for AGI, and there is a fundamental
difference between a robot and a laptop, since the latter, as a Turing
Machine, is disembodied and ungrounded. I want to argue that this
opinion is based on a misconception.
My position statement is: If in a sense a laptop computer is a Turing
machine, then in the same sense a robot is also a Turing machine.
Of course a robot and a laptop have many differences, which are
relevant to AGI, but irrelevant to this discussion.
Pei
On 10/2/06, Richard Loosemore <[EMAIL PROTECTED]> wrote:
All right, Pei: I'm going to turn the question back on you and ask more
questions, by way of an answer.
Suppose I had a computational system that consisted of large numbers of
small units, each of which was approximately like a conventional
processor, but with some of its activities governed by real-world noise
sources, to make the overall behavior not-quite-turing. (In other
words, a neuron! Or in other words again, a turing machine plus noisy
inputs that were ultimately governed by quantum-scale noise).
Now, suppose that the thing was best modelled and understood by us as an
almost-turing-machine, because it was so close.
Suppose further that this was part of an intelligent system, in which
these processing units were playing a role down at a very low level
(roughly equivalent to the neural level in our brain), and that there
was a higher level (roughly equivalent to the conceptual or logical
reasoning level in our brain) where there was once again an
almost-turing level of description.
Now, to an outside observer, there is a choice. They could go right
down to the quantum-turing level (as David Deutsch did) and declare that
the system is functionally equivalent to a Turing machine. Or they
could go up to the unit level and say, well, it is almost a turing
machine there. Or they could go up further and say that it is almsot
turing there too.
My point is: what conclusions does one want to draw from the turingness
of something? If I can (in Ben's words) take my cappuccino here and say
that it can be "approximated arbitrarily closely by a quantum Turing
machine", the currency has been debased so much that I don't see it
being any use.
Richard Loosemore
Pei Wang wrote:
> Sergio,
>
> I'm asking about your (b).
>
> Put it in different words, my question is whether there is fundamental
> difference between a robot and a conventional computer, such as my
> laptop. I see that both you and Charles think there is a difference
> (the sensorimotor capability), while Ben thinks there is no
> theoretical difference, as far as the current discussion goes.
>
> My own answer is the same as Ben's, though for a different reason ---
> I'd rather not involve concepts like quantum Turing machine. Before
> arguing for my position, I'd like to see what arguments are out there.
> ;-)
>
> Pei
>
>
> On 10/2/06, Sergio Navega <[EMAIL PROTECTED]> wrote:
>> From: "Pei Wang" <[EMAIL PROTECTED]>
>> >
>> > We all know that, in a sense, every computer system (hardware plus
>> > software) can be abstractly described as a Turing machine.
>> >
>> > Can we say the same for every robot? Why?
>> >
>> > Reference to previous publications are also welcome.
>> >
>>
>> That's a quite interesting question. Unfortunately, I don't
>> remember of any particular references (but I guess they might
>> exist). I classify this question in three steps:
>>
>> a) The computational component of the robot
>> This is as close as one may get to a turing machine, of course
>> without enforcing the physically impossible infinite tape.
>>
>> b) The computational plus the bodily component
>> (or just the bodily). Here we are not talking about turing-like
>> devices. Physical idiosyncrasies of the mechanical components
>> cannot be computed (it's a too complex problem).
>>
>> c) The system robot plus its environment
>> This is as far of turing formalisms as one can get. And this is
>> the situation most AI systems will have to face.
>>
>> Sergio Navega.
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/[EMAIL PROTECTED]
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/[EMAIL PROTECTED]
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/[EMAIL PROTECTED]