Re: what is the Self

On Jul 6, 9:33 am, Wolf K <weki...@xxxxxxxxxxxx> wrote:
casey wrote:
casey wrote:
[...]

To understand how I am using various terms such as
"machine" or "behavior" you might like to down load
and read the book that first made me think about
these things many decades ago.

http://pespmc1.vub.ac.be/ASHBBOOK.html

On Jul 5, 5:29 pm, Wolf K <weki...@xxxxxxxxxxxx> wrote:
But what is a "goal" functionally speaking?

For my use of the word see book above.

A program is a linear machine (hence Turing's
idealisation of it as a tape reading machine).
I believe that a machine capable of learning
will be non-linear.

Do you know how to program computers?

-
JC

Yes. Mind you, I haven't written a program in years. I've designed
programs in meta-language a couple of time, for my own amusement, but
haven't coded them. I've considered the properties of programs as
abstract machines, too, but that's not at all easy, mostly because I
lack the requisite math. Anyhow, I find it easier to think of a program
visually, as a network or graph. This makes program execution equivalent
to traversing a path through the graph. Writing a program amounts to
specifying a path through the graph. A badly written program will have
edges that are never traversed, BTW. A "useful" program will have
several possible paths, but on von Neumann machines each execution will
traverse one and only one of these. Etc. But I'm sure you've come across
these ideas many times.

And now you may want to tell me that because a program is a network, it
isn't linear. Well, it's useful to map the program as a network, but
that network is always traversed in a linear sequence. Even when two or
more threads are running at the same time, the machine itself is merely
switching between them, and logically, the threads form a sequence. And
while two or more CPUs may each be running a thread, at some point one
of the CPUs has to send its output to the other. Multiple CPUs merely
save time, since certain subtasks can be run simultaneously instead of
sequentially. This method IFF there is no fixed sequence of some subset
of tasks. But logically, these subtasks still form a linear sequence.
There is ultimately only one path through the program, even if the logic
allows for some variations in subtask sequencing, i.e., if these
variations don't affect the output. That's why I claim the program
(machine) is linear.

OTOH, it's fairly obvious to me that animal brains are non-linear. That
is, it makes a difference in which order tasks are performed, and/or
which tasks are performed synchronously, and/or how often the tasks have
been performed, etc. For example, if you train an animal to perform A
you may not be able to train it to perform B, unless you first
extinguish A. Or the  training sequence A, B may not permit training C,
but the sequence B, A will so permit. Or if A has been performed often
enough, it can't be extinguished, so the animal can't be trained to do
B. Etc. Example: typing. Self-taught typists must unlearn their
non-standard typing skills if they wish to learn the standard ones. As
Asimov wrote in one of his seminal essays, forgetting a task may be
essential to learning a new one.

AFAICT, conditioning is a non-linear process. Sequence of discriminants
matters. So does timing, which suggests timing variations cause
variations in signal processing sequences within the system. It's also
fairly obvious I think that conditioning requires the simultaneous, not
sequential, processing of at least two inputs in some unknown number of
sub-tasks.

I come back to a point you've avoided commenting on: that an AI machine
will IMO have to be a network of many processors (certainly hundreds,
probably thousands, possibly millions), each more or less autonomous,
the whole network's behaviour contingent on what each processor does at
any given time. There will be no linear path through the program, such
as there is with von Neumann machines.

Or so it seems to me.

cheers,
wolf k.

In theory, there will always be a contiguous "linear" path through all
the programs - imagine a super-fast computer timesharing the millions
of individual processors. There will always be some sufficiently high
execution speed that can split the hairs between interleaved
processes, yielding equivalent results to independent processes.
Impractical, yes. You've identified one of the weaknesses of
computationalism - abstract computation ignores the importance of
timing, but for real-world computation, timing is (almost) everything
and the particulars of the implementation can never be ignored. You
might call it the new vitalism.

--
Joe
.

Relevant Pages

  • Re: What did that thread indicate?
    ... > pyramid-type hierarchical network. ... lots and lots of feedback for image recognition), ... > with the problem of action timing. ... Learning to create the correct sequence is natural for this type of network ...
    (comp.ai.philosophy)
  • Re: what is the Self
    ... And now you may want to tell me that because a program is a network, ... Well, it's useful to map the program as a network, but that network is always traversed in a linear sequence. ... Even when two or more threads are running at the same time, the machine itself is merely switching between them, and logically, the threads form a sequence. ... There is ultimately only one path through the program, even if the logic allows for some variations in subtask sequencing, i.e., if these variations don't affect the output. ...
    (comp.ai.philosophy)
  • Re: divergent series
    ... (One way to obtain summation procedures is the following ... S -> S be the linear map which drops ... Let C be the subset of S of convergent ... a convergent sequence to its limit. ...
    (sci.math)
  • Re: Some simple questions about Banachs Space
    ... X->Y to be a linear function. ... F is linear - we want to prove that it is continuous in the norm ... So there exists a sequence ... we show now that it is impossible that "<" inequality is true. ...
    (sci.math)
  • Some simple questions about Banachs Space
    ... X->Y to be a linear function. ... F is linear - we want to prove that it is continuous in the norm ... So there exists a sequence ... we show now that it is impossible that "<" inequality is true. ...
    (sci.math)
Loading