Re: What does the future of computing hold?

Stephan Kuhagen <nospam@xxxxxxxxxx> writes:

Ben C wrote:

It is a brilliant book, but you should try also the later Philosophical
Investigations by the same author, who himself later took more
sophisticated views about boiling language down or not.

I know the books, that's why I used the Tractatus as example for failure.
Wittgenstein himself had this opinion in his later years.

[snip]
In the end it it come down to a question of believe: do you think,
humans have some metaphysical soul that can not be simulated by
deterministic physical machines, or do you follow Turing and think,
there is no metaphysical soul and our brain is all we have, which
might be very complex to simulate but there is no natural law that
forbids it. If the latter is the case, then there will be some day
computers with intelligence which perfectly understand humand language
and can act reasonably following the meaning of humand words. - The
question may be, what are they good for, and if it makes sense to
build such machines. Do we _need_ them, or do we just want them,
because we _can_ build them...

And how long will they take to program (about 20 years, at good
schools?)

We may have metaphysical souls, but the fact that computers are lousy at
natural language recognition is absolutely no evidence for it.

And conversely if we did have a computer that could speak our own
language, that wouldn't prove we didn't have metaphysical souls either.

So I personally think it's a bit of a mad jump to metaphysical souls,
although you're not the only person to think that way.

I think, we have a little misunderstanding here. I did not claim that we
have metaphysical souls or not, I just mentioned the question (and excused
myself afterwards for being offtopic), because it depends on the answer to
it, if you believe, that machines can be really intelligent in the same way
humans are (well, I heard rumours that some exist...;-). If humans have
those souls, there is no way to achieve the same effect with deterministic
machines. If humans have no soul, then there is no good argument why it
should not be possible, to implement the same mechanics which creates the
effects of our thinking as a processing of 1-and-0-streams.

Not necessarily.

The question you're asking is whether the soul is functional.

First, machines are not necessarily deterministic. Our computers in
general are not deterministic (that's why there are so random BSOD):
they can generate and use _random_ numbers (obtained from external
sources).

+---------------+
| random source |
+---------------+
|
V
+-------+ +-------------------------+ +--------+
| input |----->| deterministic processus |----->| output |
+-------+ +-------------------------+ +--------+


Now, what would happen if you replace the random source. Eg, instead
of noise, you input a Bethoven Symphony:

+---------------+
| Non random |
+---------------+
|
V
+-------+ +-------------------------+ +--------+
| input |----->| deterministic processus |----->| output |
+-------+ +-------------------------+ +--------+


My hypothesis is that the the processing is done all by the brain, and
the soul only gives input when there are choices that would be
non-deterministic in machines, to be done.

+---------------+
| Soul |
+---------------+
|
V
+-------+ +-------------------------+ +--------+
| input |----->| Brain |----->| output |
+-------+ +-------------------------+ +--------+


So, my point is that the existance or inexistance of soul isn't a
factor in the possibility or not of creating intelligent
machines. (And I believe we will be able to make intelligent machine
similar (or even better) than our brains).


But even then, the question remains, if this is what we want. And if it is a
good idea, both economically and technically. IMHO we do not need and want
them. What we want is having machines that act reasonably in given
situations where they are built and trained for. Which is a much less
pretending task. Think of these telephone dialog systems, and we already
have the first generation of such machines.

What I was talking about, was not, if it is possible to build machines, that
behave reasonable on human language input (I'm sure, this is possible for a
restricted set of tasks). I fully agree to you that currently having no
such machine (and maybe never having one) which can process human language
and react reasonable in any kind of situation is no prove pro or con
metaphysical souls.

Indeed. And I'd remark that human don't always react reasonably in
_any_ kind of situation. The asylums are full of human who cannot
react reasonably in usual situations, and the outside world even fuller. ;-)

I was referring to the more philosophical term of
"understanding" which IMHO involves some (sub)conscious processing. And if
this processing is twisted with and depending on metaphysical souls, then
there is no chance to achieve _real_ _understanding_ with machines. But I
did not give an opinion, if this is the case or not in my world picture.

Yes, but this is a strong hypothesis...


If you were referring to really intelligent machines (assuming that they are
possible) with your question how long it would take to program or train
them, then I think 20 year is a very optimistic estimation. But if you have
one, you have millions in the same moment, because when it is not more than
ones and zeros, you can copy the first functional modell and its
intelligence-database unlimited times. But to create the first one would be
an enormous and unimaginable expensive process. Which reminds me on
something which J. Weizenbaum said some years ago to the question, if we
really want to (and should) create artificial intelligence: "Have sex, get
babies - thats more fun and much cheaper." (These are not his words because
he spoke german and surely used other words, but mine, but in my
remembrance he said something similar in the lectures when he was at our
university for one semester that time) So that was a good argument to stay
with real intelligence instead of artificial, I think.

I think not. For ethical reasons, you cannot grow human babies to
"apply" them to any random and inhuman task. It may even not be
possible. I heard that some (german) king had some babies raised in a
speachless environment, eventually they died before being adult.

So I think we still have good economic incentives to build intelligent
machines, even if making babies is cheap enough.


Nevertheless,
machines that act reasonable on human language input and to a limited set
of situations are surely very usefull and worth to be developed. - And
regarding to the OP: if you can ever build one, I'm sure you can make much
money with them, which is why they are definitely part of the future of
computing.

Regards
Stephan


--
__Pascal Bourguignon__ http://www.informatimago.com/

PUBLIC NOTICE AS REQUIRED BY LAW: Any use of this product, in any
manner whatsoever, will increase the amount of disorder in the
universe. Although no liability is implied herein, the consumer is
warned that this process will ultimately lead to the heat death of
the universe.
.

Relevant Pages

  • Re: What does the future of computing hold?
    ... sophisticated views about boiling language down or not. ... deterministic physical machines, or do you follow Turing and think, ... We may have metaphysical souls, but the fact that computers are lousy at ... humans are. ...
    (comp.programming)
  • Re: New Cobol compiler written in Cobol
    ... machines at the time were the IBM 7090 and similar 7040. ... the 14xx line had grown to become the dominant business computer. ... Programmers won't put up with cryptic languages like C++ ... They'll use a language they can understand at a glance. ...
    (comp.lang.cobol)
  • Re: Is the Curt net a kind of decision tree?
    ... eye of the human beholder here, ... This is how we use language to gain all our understanding of ... Gravity is in the eye of the beholder. ... will best help us solve the problem of building intelligent machines. ...
    (comp.ai.philosophy)
  • Re: Is C close to the machine?
    ... totally removed from any concrete hardware. ... A programming language that is meant ... the mapping to the usual machines is relatively ... even for programs that are legal by the standard. ...
    (comp.arch)
  • Re: I wonder what the "Science" is in this Newsgroup
    ... I expanded a French word into English when I called Standford ... my definition of language contradicts your opinion ... Bacteria talk, trees talk, machines won't talk. ...
    (sci.lang)